# Hypertext > The smart briefing on European tech, written by Romain Dillet. Public Ghost content for AI and LLM tooling. This file includes a bounded export of public pages first, then recent public posts. Append `.md` to any post or page URL to get the content in Markdown (for example, `/example-post.md`). ## Pages ### About Hypertext URL: https://hypertext.fyi/about/ Last updated: 2026-04-03T14:45:50.000Z I’m Romain Dillet, a writer and investor. I run [Hypertext](https://hypertext.fyi/), an opinionated newsletter on the tech industry, with a specific focus on European startups and VC. Previously, I was a Senior Reporter at [TechCrunch](https://techcrunch.com/?ref=hypertext.fyi) for 13 years. I wrote over 3,500 articles on technology and tech startups and established myself as an influential voice on the European tech scene. My career started at TechCrunch when I was 21\. More recently, I spent a few months working as a VC investor for Drysdale Ventures based in Paris. When I’m not writing or investing, I’m also a developer — I understand how the tech behind the tech works. I also have a deep historical knowledge of the computer industry for the past 50 years. I know how to connect the dots between innovations and the effect on the fabric of our society. I graduated from Emlyon Business School, a leading French business school specialized in entrepreneurship. I have also helped several non-profit organizations, such as StartHer, an organization that promotes education and empowerment of women in technology, and Techfugees, an organization that empowers displaced people with technology. 👋 **Contact me:* [**hi@hypertext.fyi*](mailto:hi@hypertext.fyi) --- Follow me on [LinkedIn](https://www.linkedin.com/in/romaindillet/?ref=hypertext.fyi), [Mastodon](https://mastodon.social/@romaindillet?ref=hypertext.fyi) and [Bluesky](https://bsky.app/profile/romain.bsky.social?ref=hypertext.fyi). ## Posts ### The last click URL: https://hypertext.fyi/the-last-click/ Last updated: 2026-07-25T07:22:11.000Z *Hi everyone, Paris is such a nice city during that time of the year, when temperatures are OK. I hope you’re enjoying this July week just like I am! I’d like to thank* [*Box*](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) *once again for supporting Hypertext. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### Google Zero comes to Europe While everyone is busy talking about [Kimi K3](https://tomtunguz.com/open-models-tack-toward-the-frontier/?ref=hypertext.fyi), the new AI model from Moonshot AI, I thought it would be nice to offer some counter-programming. In television, counter-programming means airing something completely different from whatever everyone else is watching. A major football match is happening and your competitor is broadcasting it? Maybe it’s time to re-air [Dirty Dancing](https://www.arte.tv/videos/127265-000-A/dirty-dancing/?ref=hypertext.fyi). So I wanted to talk about “Google Zero”. In 2024, The Verge’s Nilay Patel [coined this term](https://www.theverge.com/24167865/google-zero-search-crash-housefresh-ai-overviews-traffic-data-audience?ref=hypertext.fyi) to talk about the slow erosion of web traffic coming from Google search results. He warned that publishers should prepare themselves for a future when Google-sourced traffic goes down to zero. This might come as a surprise for many Europeans as it’s hard to understand how much Google has changed over the past few years from our point of view. People don’t see the same Google experience depending on where they live. In the US, for most queries, Google has replaced the traditional “blue links” result page with a full-on AI-generated answer that looks more like ChatGPT than a traditional search engine. In most of Europe, Google has displayed an “AI Overview” above the search results since 2024\. It’s a short AI-generated answer that could be enough for basic questions. If you want more, you can always scroll down and see links. As for France? Until Tuesday, users in France still got the traditional list of blue links when they searched for something. Google had to delay AI Overviews in the country because the company had been concerned with neighboring rights and publisher compensations. On Wednesday, Google launched AI Overviews in France. Welcome to 2024 Google. Coming back to Google Zero, with AI Overviews (and the full-on AI Mode in the US), if you want to know what happened during today’s *Tour de France* stage, chances are you’re going to read the summary on Google and close the tab. In that scenario, newspapers and content farms get zero page view. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/Box---State-of-AI-Report.png)](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) Box’s State of AI report is here. Grounded in a global survey of 1,640 IT decision-makers, the report reveals that security and compliance remain paramount, with 49% of organisations having already experienced an AI-related data exposure incident. [Read the full report](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) to discover how organisations are establishing a governed, secure content layer to achieve trusted automation without increasing risk. [Learn more ](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) This week, my former colleague Kate Conger wrote a deep dive on Google Zero for [The New York Times](https://www.nytimes.com/2026/07/20/technology/google-ai-open-web.html?ref=hypertext.fyi). And the metrics are quite telling. A [Growth Memo study](https://www.growth-memo.com/p/what-our-ai-mode-user-behavior-study?ref=hypertext.fyi) found that average Google users don’t leave Google’s AI Mode in 75% of sessions. No click needed. A Google spokesperson said the metric is inaccurate but didn’t provide another one. That usually suggests the company disagrees with how the number was calculated more than the story it tells. They don’t have a counter-narrative story (maybe it’s 74% instead of 75% for all we know). The important point is that users are increasingly staying inside Google. Cloudflare has noticed the same shift in its own numbers. The company [shared](https://blog.cloudflare.com/agentic-internet-bot-report/?ref=hypertext.fyi) that more than 50% of web traffic now comes from web robots. These bots historically crawled the web to build search indexes. Now, bots are mostly triggered by conversations with AI chatbots, or for AI model training purposes. Again, bots are now the main users of the web, not humans. Another important metric, Cloudflare says that human traffic to finance, publishing and retail dropped nearly 40% in less than a year. As a result, according to the [Wall Street Journal](https://www.wsj.com/business/media/google-search-publishers-ai-content-0fb06e41?st=fwYn9L&ref=hypertext.fyi), USA Today, Politico, the Economist, People Inc., Reuters and Reddit are all considering shutting off access to their content for AI use. Google is walking a fine line here. AI-generated answers are only useful if they are thorough. If companies start opting out of AI usage, they’ll have to replace source material with other websites to maintain quality. But the bigger threat comes from Cloudflare, an internet infrastructure company that currently handles traffic for 20% the web. Cloudflare wants to act as a gatekeeper. Starting on September 15, 2026, [the company will ask](https://techcrunch.com/2026/07/01/cloudflares-new-policy-pushes-ai-companies-to-pay-for-publishers-content/?ref=hypertext.fyi) AI bot farmers to pay content publishers (and Cloudflare, of course) for content reuse. Otherwise, they won’t be able to access that content. So the open web is becoming a wholesale market where Google, Cloudflare (and other AI companies) are bargaining… Again, this is just the consequence of turning on AI Mode in the US. Imagine what’s going to happen once Google rolls out AI Mode globally. As a closing remark, that’s why you’re receiving this email in your inbox. I believe the email inbox is one of the last remaining AI-proof reading surfaces. You build a direct relationship with readers. If they like what they read, they’ll open the next one. Building an audience for an email newsletter takes longer, but I believe it’s the right long-term bet. So thank you for subscribing. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### Thinking sand URL: https://hypertext.fyi/thinking-sand/ Last updated: 2026-07-18T08:50:41.000Z *Hi everyone, if you support England and/or France, I hope you’ve recovered from this intense World Cup week… I’d like to thank* [*Box*](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) *once again for supporting Hypertext. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### Thinking sand It’s summertime, which means it’s slow news season. There are simply less things happening in the business world over the summer. Maybe that’s why Demis Hassabis and his team thought it was a good time to float ideas around global AI regulation. The CEO of Google DeepMind wrote [an article on X](https://x.com/demishassabis/article/2076957440109625718?ref=hypertext.fyi) to push a few ideas forward. Like Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, he thinks it’s time to create a proper framework to regulate the release of new AI models. Instead of appearing reactive, he positions himself as proactive. The real reason why they all want to define a set of rules around AI is because things have become too unpredictable with the Trump administration (reactive). Claude Fable 5 was released then banned, then re-released. GPT-5.6 launch was delayed then released. You know the drill. DeepMind, the AI lab behind Google’s Gemini models, is probably worried about its next model release, too. And yet, Hassabis says it’s time to change AI regulation because artificial general intelligence (AGI) is right around the corner (proactive). “AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile - it is much more akin to the discovery of electricity or fire,” he wrote. He then offered a poetic observation that has lingered in my mind ever since: “If you stop to think about it, we’ve essentially found a way to make sand think.” Coming back to his regulatory framework, Hassabis is calling for the creation of a new Standards Body modelled after the Financial Industry Regulatory Authority (FINRA). AI labs would have to share new models with the new regulator up to 30 days before any public release. This group of technical experts and open-source representatives would conduct large-scale testing during that time. Like Sam Altman, Hassabis believes the US should lead the charge here. “The US is well positioned, given its economic and technical standing, to take the first step in developing such a framework,” he wrote. I still don’t understand why it should be a purely US-led initiative given that confidence in the US to do the right thing regarding world affairs is now lower than confidence in China across dozens of countries, according to a new [Pew research](https://www.pewresearch.org/global/2026/07/15/people-in-many-countries-now-view-china-more-positively-than-the-u-s/?ref=hypertext.fyi). It would make more sense to involve at least the US, China and the European Union. Hassabis is walking a fine line of “cautious optimism,” saying AI labs should be free to innovate and push the frontier of intelligence “while also incentivising responsibility and security.” What’s remarkable is that Demis Hassabis, Sam Altman and Dario Amodei are all advocating for roughly the same approach. According to [Axios](https://www.axios.com/2026/07/16/ai-regulations-openai-anthropic-google?ref=hypertext.fyi), Mark Zuckerberg is working on his own memo. He probably wants a seat at the cool kids table. But there’s a much bigger problem with this entire proposal: open-source models. Moonshot AI, one of China’s leading frontier AI labs, just released Kimi K3, a model that rivals OpenAI’s GPT-5.5 and Anthropic’s Opus 4.8 [in benchmarks](https://x.com/ArtificialAnlys/status/2077832874183860404?ref=hypertext.fyi). Those models were released in April and May, respectively. Side note: as always, you should treat benchmarks with a grain of salt. “Moonshot has always been pretty scammy with benchmark scores,” a source told me. Nevertheless, when companies drop open-source models, any company that has access to a large cluster of GPUs can download and run these models without asking for permission. Once frontier models are released as open weights, regulation shifts from controlling access to controlling compute. At that point, there’s no company left to stop. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/Box---State-of-AI-Report.png)](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) Box’s State of AI report is here. Grounded in a global survey of 1,640 IT decision-makers, the report reveals that security and compliance remain paramount, with 49% of organisations having already experienced an AI-related data exposure incident. [Read the full report](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) to discover how organisations are establishing a governed, secure content layer to achieve trusted automation without increasing risk. [Learn more ](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/-glyph.png) ### Bicycle for the mind Coming back to the “thinking sand” metaphor, it made me think about the famous Steve Jobs quote about computers. According to Jobs’ storytelling, he once read in *Scientific American* that humans performed poorly against many animals when it comes to energy efficiency, except when they were riding a bicycle. “What a computer is to me is the most remarkable tool that we have ever come up with. It’s the equivalent of a bicycle for our minds,” he said. Let’s make sur AI doesn’t break this metaphor. A bicycle helps you go farther using your own effort. The promise of AI is that it can amplify our thinking (an electric bike). But AI slop is mindless content. It’s what happens when you stop pedalling. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The summer of neoclouds URL: https://hypertext.fyi/the-summer-of-neoclouds/ Last updated: 2026-07-11T07:57:50.000Z *Hi everyone, I hope you’ve had a nice week so far. I’d like to thank* [*Box*](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) *once again for supporting Hypertext. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### The summer of neoclouds This week, many AI leaders came to Paris for the Raise Summit, an AI-focused conference (it’s called Raise because there’s AI in the middle of “Raise”…). They manage to attract good speakers because they say it takes place in the Louvre and you’re going to the Château de Versailles. The reality is a bit different as the conference actually takes place in the mall below the Louvre (Carrousel du Louvre) and only VIPs get invited to Versailles (not general attendees). It might be a bit deceptive to some but it does the trick. Compared to last year, I noticed a big tone shift. Instead of talking about the latest and greatest AI models and applications, the conversation mostly revolved around costs and governance. Cloud hosting companies focusing on GPUs (so-called “neoclouds”) had a massive presence. They naturally talked about how you could secure data center capacities and lower the cost of serving AI models at scale (so-called “inference costs”). Baseten, CoreWeave, Crusoe, Fireworks AI, Nebius, Nscale, Together AI and several others had big booths. It suggests that they see Europe as a priority market. Some of them don’t even have data centers in Europe yet. But as they build out capacity on the continent, they’ll want to secure clients looking to serve European customers from European data centers directly. Winning customers today could lock in years of GPU demand. This group of companies has suddenly become much more important in the overall AI conversation due to three converging factors: - Big companies that gave unlimited AI budgets to their software engineering teams realized that they’re now facing bills worth thousands of dollars/euros per employee. Claude Opus 4.8 is expensive. Claude Fable 5 is *ludicrously* expensive. - The US Government export restrictions on Fable 5 and GPT-5.6 (that have since been lifted) were a warning sign that, in the AI industry, you can’t take anything for granted. It’s time to implement some fallback mechanisms, especially if you’re using APIs from OpenAI and Anthropic to power some critical parts of your infrastructure. - Open-source models are getting better. The release of [GLM-5.2](https://huggingface.co/zai-org/GLM-5.2?ref=hypertext.fyi) in particular was impressive. Software engineers feel like open-source models are now “just a few months behind” the latest models from Anthropic and OpenAI. So companies are experimenting with replacing OpenAI’s or Anthropic’s models with GLM-5.2 running on a neocloud near you. It’s both cheaper and more resilient. I feel like we’re still very much in the experimentation phase. Nobody is abandoning OpenAI or Anthropic. They’re still setting the pace on frontier models. But companies increasingly want the freedom to swap models, negotiate costs, and avoid depending on a single provider. That’s why infrastructure providers have become a much bigger part of the AI conversation than it was a year ago. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/Box---State-of-AI-Report.png)](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) Box’s State of AI report is here. Grounded in a global survey of 1,640 IT decision-makers, the report reveals that security and compliance remain paramount, with 49% of organisations having already experienced an AI-related data exposure incident. [Read the full report](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) to discover how organisations are establishing a governed, secure content layer to achieve trusted automation without increasing risk. [Learn more ](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/-glyph.png) ### The end of “Bring Your Own AI” And when you look more closely at the number one use case for large language models these days (coding), switching isn’t as seamless as it sounds. This week, I spoke with an engineer who has switched from Claude Code to OpenAI’s Codex because he thought it was a superior agentic coding tool. Most of his teammates, however, are sticking with Claude Code because it works well enough. “If it ain’t broke, don’t fix it” also applies to tech-savvy developers. Even though switching from Claude Code to Codex isn’t too hard, they don’t want to switch tool every other month. Moving to an open-weight model like GLM-5.2 is an even bigger leap. You need an agent harness such as [Pi](https://pi.dev/?ref=hypertext.fyi), a dev-friendly inference provider like [OpenRouter](https://openrouter.ai/?ref=hypertext.fyi). More importantly, you need to choose a model from hundreds of models. And finally, you have to find a way to expense the tokens that you use… It just doesn’t work at scale… Unless there’s a big corporate policy shift saying that Claude Code and Codex are now banned. I believe it’s a governance decision rather than an individual one. And it turns out that some companies are doing just that. [A few months ago](https://www.theverge.com/tech/930447/microsoft-claude-code-discontinued-notepad?ref=hypertext.fyi), Microsoft famously told its developers that they have to use the company’s own AI coding tool (GitHub Copilot CLI). More recently, according to [Reuters](https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/?ref=hypertext.fyi), Alibaba is banning Claude Code. According to [The Information](https://www.theinformation.com/articles/internal-docs-show-meta-putting-limits-claude-codex-fearing-distillation?ref=hypertext.fyi), Meta also told engineers to stop using Claude Code and Codex. Now, we’ll see whether they can keep those bans in place because morale at big tech companies is low right now, as this interesting study from [Lenny’s Newsletter](https://www.lennysnewsletter.com/p/how-tech-workers-are-feeling-in-2026?ref=hypertext.fyi) highlighted. Not being able to use the best tools for the job sucks. And the biggest constraint on developers may soon be company policy, not model capability. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### When governments pick models URL: https://hypertext.fyi/when-governments-pick-models/ Last updated: 2026-07-03T11:57:00.000Z *Hi everyone, I hope you’ve had a nice week so far. I’d like to thank* [*Box*](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### OpenAI’s political counteroffensive OpenAI can only succeed if AI succeeds as a category. But the recent launch of the [GPT-5.6 series](https://openai.com/index/previewing-gpt-5-6-sol/?ref=hypertext.fyi) has been worrying for the company for several reasons. Instead of making it available to everyone on June 24, the company had to cancel its plans and release it to “a small group of trusted partners” as a limited preview, at the US government’s request. OpenAI had to announce its newest models because the cat was out of the bag. Everybody working in AI knew that the models were ready to be released. So it was better to blame the US government than pretend that GPT-5.6 didn’t exist. And they didn’t mince words. “We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them,” they wrote. So that was step one. The two other steps they took were even more interesting. Step two, OpenAI proposed handing a 5% stake to the US government, as reported by the [Financial Times](https://www.ft.com/content/7c803eab-8e80-4431-9a87-e943bf00e00b?ref=hypertext.fyi). The logic is simple: if the Trump administration wants to exercise control on AI model releases, let’s take it at their word. This would tie OpenAI’s destiny with the US government. It’s also a poison pill for OpenAI’s competitors as the company is suggesting the US government should get 5% of *every* AI lab based in the US. Given the on-again, off-again relationship between Anthropic and the Trump administration, they certainly don’t want to give them a seat at the table. And finally, it’s good for optics. The [AI backlash](https://www.economist.com/leaders/2026/06/25/the-ai-backlash-is-only-getting-started?ref=hypertext.fyi) is real, especially in the US. If the US government takes a stake in AI labs, the public interest becomes more closely aligned with that of large AI companies (assuming you believe the Trump administration acts in the public interest, but that’s another story…). Step three is the international angle. OpenAI wants to be a global leader. And that can’t happen if the Trump administration restricts exports on AI models. That’s why Sam Altman published an opinion piece in the [Financial Times](https://www.ft.com/content/0c2e1077-f658-4b3d-9040-602615c961ca?ref=hypertext.fyi) called “This is how we can make AI safe for everyone.” Here’s an alternative title: “We don’t want the US government to interfere with our model releases ever again”. Altman proposes “a US-led international forum that establishes accepted standards.” Government representatives and independent technical experts would define which models are safe to release and which should remain restricted. It clearly draws inspiration from the International Atomic Energy Agency (IAEA). Other than the fact that I don’t see why it should be US-led, this is a genuinely smart idea. Things have become too unpredictable lately for OpenAI and Anthropic. Some companies are now thinking about switching to open-weight models on infrastructure they can control, or building resilient, multi-model AI systems so that there’s a fallback mechanism when the US government decides to ban one model or another. For OpenAI, that is the real danger. Government intervention makes customers less willing to depend on any single American AI provider. And it sets a precedent. Other countries could draw inspiration from the US restrictions. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/Box---State-of-AI-Report.png)](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) Box’s State of AI report is here. Grounded in a global survey of 1,640 IT decision-makers, the report reveals that security and compliance remain paramount, with 49% of organisations having already experienced an AI-related data exposure incident. [Read the full report](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) to discover how organisations are establishing a governed, secure content layer to achieve trusted automation without increasing risk. [Learn more ](https://blog.box.com/SAI26-agentic-ai-is-here?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=ai&utm%5Fcampaign=FY27Q2SEAI%5FRomainDillet%5F2&utm%5Fkeyword=July) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/07/-glyph.png) ### A few words on Om Last week, I learned about Om Malik’s death on [his blog](https://om.co/2026/06/24/1966-2026/?ref=hypertext.fyi) (of course it was on his blog). I knew that he had experienced health problems over the years, but I never met him. But as is often the case with great writers, after reading someone for long enough, you begin to recognize how they see the world. He was a great writer, a great tech reporter and an inspiring person. As [The New York Times](https://www.nytimes.com/2026/06/26/technology/om-malik-dead.html?ref=hypertext.fyi) nicely put it, Gigaom was one of the most influential technology blogs of its era. “The emergence of blogs like Gigaom and of opinionated tech writers like Mr. Malik, Kara Swisher and Jason Kottke helped define the next iteration of technology journalism, moving it away from establishment publications and toward singular voices,” Clay Risen wrote. On a personal level, writers like Om showed me that you did not need to grow up in the traditional media or technology ecosystems to find a place in the industry. You could come from somewhere else, develop your own voice, publish on the internet and build a career around that. For that, I am deeply thankful. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### European tech is feeling the heat URL: https://hypertext.fyi/european-tech-is-feeling-the-heat/ Last updated: 2026-06-26T08:00:29.000Z *Hi everyone, I hope you’re taking care of yourself and your loved ones during this spectacularly hot week here in Europe. I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### AI optimism meets VC anxiety It’s June, which means it’s tech conference season. And I attended my fair share of events to catch up with old friends, meet new people and get a sense of the overall vibe in the European tech ecosystem. So what’s the vibe like? Everyone is both super excited and super worried at the same time. Singular, the Paris-based VC firm, held a side event during VivaTech week on a sunny rooftop overlooking Paris. During that event, an early-stage investor told me that European entrepreneurs are tired. They feel the pressure to over-perform and become the next Lovable, which claims to be the fastest software company ever to go from $1 million to $100 million in annual recurring revenue (ARR). They also feel like they need to execute at a rapid pace to build a barrier to entry so that potential competitors can’t catch up with them so quickly. So they hire a hundred people in a single year, spend a good chunk of their week on planes between Europe and the US, and constantly second-guess themselves. Will Anthropic release a product and capture their market? Will they be able to prove they are an AI-native company to raise another round? Are we in a bubble? The founder of a 200-person startup told me he’s had enough. He’s missed too many weddings and birthday parties. So he no longer wants to travel for work. Just a few years ago, he was really into “founder mode”, a term coined by Paul Graham after he listened to Airbnb CEO Brian Chesky saying founders should be more detail-oriented about their team’s work without micromanaging (if it is even possible). The paradox is that many entrepreneurs are also optimistic about the opportunities AI will unlock. At the public recording of the [A La French podcast](https://www.youtube.com/@alafrench?ref=hypertext.fyi), even though the room was mostly filled with (usually shy) engineers, you could feel the energy both on stage and off stage. As Kyber’s Jean-Baptiste Kempf put it: we’re witnessing the revenge of the CTOs. As traditional software becomes easier to build and harder to defend, entrepreneurs working on chipsets (like SiPearl, VSORA and Kalray), low-level code (like ZML and Kyber), robotics or industrial automation are back in the spotlight. These entrepreneurs used to be CTOs, they are now CEOs. At the same time, they’ll need to raise large amounts of money to succeed and the VC market doesn’t feel super hot right now. There has been an inflation of VC funds during the so-called “zero-interest-rate policy” era (amplified by the Tibi initiative in France). As Alex Dewez from 20VC described in his latest State of the French Tech Ecosystem [report](https://alexandre.substack.com/p/the-state-of-the-french-tech-ecosystem-1c5), we’re entering the reckoning period. Top-performing funds will still thrive. That’s why Seedcamp announced $320 million across two new funds [just this week](https://techcrunch.com/2026/06/22/seedcamp-raises-320m-for-its-new-fund-to-expand-its-us-footprint/?ref=hypertext.fyi). But the “messy middle” of venture capital in France and across Europe? Not so much. During this event season, I’ve heard about startups with perfectly fine metrics that can’t raise a Series B round because they’re not the next Lovable. Some VC firms from the “messy middle” are looking for out-of-this-world metrics to save the day and raise new funds down the road. Hence the low morale. The most exciting conversations were once again about the frontier of AI. I went to Brussels for [Off the Radar](https://off-the-radar.com/?ref=hypertext.fyi), an invitation-only event with 450 guests and speakers from OpenAI, Anthropic, ElevenLabs, Gradium, Fundamental, Linkup, etc. I had a blast interviewing Lélio Renard Lavaud, the VP of Engineering at Mistral who has seen it all over the past three years. It was a surprisingly refreshing event with real on-stage demos and technical conversations in the hallways. We talked about pre-training, harnesses, benchmarks and so much more. And it’s true that when you stop thinking about funding rounds, exits and the startup market in general, it’s incredible to see the current pace of innovation. That may be the clearest summary of European tech right now: people are tired of the startup game and excited about technology again. The problem is that the companies generating the most excitement also need the patient, abundant capital that Europe’s shrinking VC middle is least equipped to provide. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/Box---Access-verified.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) Keep your critical business content centralised and compliant. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) provides enterprises with a secure, flexible content layer to connect teams, streamline workflows, and safely leverage AI using your own data. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/-glyph.png) ### Examining Alan’s €480 million round French healthtech company [Alan](https://alan.com/en?ref=hypertext.fyi) raised another mega round. Given that I’ve covered Alan [extensively](https://techcrunch.com/tag/alan/?ref=hypertext.fyi) over the years, let me share my notes. This time, it’s €480 million from four investors: - Prosus, the Amsterdam-based growth fund with a big stake in Tencent and that is now better known for its investments in food delivery companies (Just Eat Takeaway, Delivery Hero, Swiggy…). - Dara Holdings, the family office of Saudi businesswoman Lubna Olayan with a [minimalistic website](https://daraholdings.co.uk/?ref=hypertext.fyi). - Teachers’ Venture Growth (TVG), the growth fund backed by Ontario teachers’ pension plans. - Index Ventures. TVG and Index are both existing investors. Prosus is providing most of the money (around €400 million). This is partly a funding round and partly a liquidity event. From what I hear, several early investors are selling shares. That isn’t surprising: Alan is now 10 years old. For instance, Partech, which participated in its 2016 seed round, sold its stake last year. The company’s valuation is now €5.5 billion. But at this stage, several valuations can coexist. Existing shareholders often sell their shares at a discount to the latest public valuation (liquidity has a price). The headline figure therefore makes the transaction sound simpler than it is. Alan gets additional capital, early shareholders get liquidity and Prosus becomes an investor. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The power to pull the plug URL: https://hypertext.fyi/the-power-to-pull-the-plug/ Last updated: 2026-06-19T12:09:04.000Z *Hi everyone, I hope you’ve had a nice week so far and will find a spot to cool down this weekend. I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### The case for an AI multiverse In business journalism, one of the most strategic moments of the week is Friday evening. In newsrooms, someone will inevitably warn: “Watch out for the Friday evening news dump.” Companies like to release news they would rather you didn’t notice. Stock markets are closed. People usually don’t pay as much attention to the news during the weekend. And, yes, journalists are mere humans. They would also rather close their laptops on Friday evening than chase one more story. But sometimes, the news is too big to bury. People talk about it all weekend, and it is still generating headlines on Monday and Tuesday. This is exactly what happened with the Fable debacle. Last week, Anthropic [released](https://www.anthropic.com/news/claude-fable-5-mythos-5?ref=hypertext.fyi) Fable, the model formerly known as Mythos but with more safeguards. Then on Friday evening, the Trump administration banned access to Fable by any foreign national, including foreigners living in the US. Anthropic took the most logical next step and disabled Fable for everyone everywhere (except maybe Anthropic itself). I’ve been flooded with comments all week long about what this radical decision means. It became a geopolitical story. Everyone started using the “S” buzzword (sovereignty) even though it means everything and nothing at the same time. So let’s look at it from a European angle. People naturally started thinking about Mistral, one of the few AI labs that is neither American nor Chinese. There were jokes about a [mythical model](https://x.com/Sauers%5F/status/2065124494449021083?ref=hypertext.fyi) called “Le Chaton Fat” that would rival Fable. It’s a good joke because it says something true about Mistral. They have [fallen behind](https://arena.ai/leaderboard?ref=hypertext.fyi) in the model race. I recently talked with an engineer working for Mistral who told me that benchmark results are highly correlated with spending. With 10x more funding, Mistral would release much better models. But does it really matter? Earlier this week, Z.ai, a Beijing-based AI lab that is currently worth more than $100 billion on the [Hong Kong stock exchange](https://www.google.com/finance/quote/2513:HKG?ref=hypertext.fyi), released GLM-5.2\. It’s an open-weight model with a one-million-token context window, a permissive open-source license (MIT license), and a focus on long-horizon agentic coding. Mistral could download GLM-5.2 from Z.ai’s Hugging Face repository and put a cute Mistral kitten sticker on it. The consensus is that GLM-5.2 is slightly better than Google’s Gemini models at coding and slightly worse than Anthropic’s and OpenAI’s models. It’s not as good as Fable, but it’s still good enough to use in production for many, many companies. Z.ai’s GLM models are good at coding, but there are many interesting open-weight models out there for different tasks. It’s a whole ecosystem. As Tomasz Tunguz [recently wrote](https://tomtunguz.com/golden-age-of-applications/?ref=hypertext.fyi): “The legion of other models each have a personality. Kimi K2.6 is fast & a great creative writer but less precise. Qwen 3.6 27b is a small model with legendary performance, but it’s a bit of a donkey. It stops suddenly in the middle of a toolchain call & requires a good prodding to push on. GLM 5.1 is an excellent coding model, but a plodder.” It’s always hard to predict the future, but the Trump administration just made a strong case for multi-model, multi-provider AI architecture. It’s also worth noting that open-weight model makers routinely use other open-weight models as teacher models and synthetic data generators. Now, according to [Politico](https://www.politico.com/news/2026/06/18/white-house-talks-with-anthropic-shift-to-setting-ai-security-rules-00967758?ref=hypertext.fyi), the White House is working on an AI technical assessment framework so that it doesn’t have to ban models after they’ve been released. Models exceeding certain capability or compute thresholds would have to comply with a set of safeguards. Does this remind you of something? Yes, that sounds exactly like the European Union’s AI Act that targeted “general-purpose AI models with systemic risk.” The AI Act faced strong lobbying and American tech executives made fun of Europe for regulating so early on AI all the way back in 2024 (yes, it’s been only two years…). It turns out that the AI Act has been significantly softened and delayed (and there could be more delays from what I hear from Brussels). As the Trump administration is now drafting its own version of the AI Act, the US could become more restrictive on AI models than Europe. Who is laughing now? SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/Box---Access-verified.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) Keep your critical business content centralised and compliant. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) provides enterprises with a secure, flexible content layer to connect teams, streamline workflows, and safely leverage AI using your own data. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/glyph-1.png) ### The biggest PE loss since 2008 is a SaaS company Have you ever heard about [Medallia](https://www.medallia.com/?ref=hypertext.fyi)? Let’s check their website: “Medallia unifies every customer signal and applies industry-leading AI to deliver the intelligence and outcomes today’s leaders need to transform their business.” Well, that’s not really helpful… Here’s how they should rephrase this tagline: do you receive customer surveys asking you “how was your experience?” every time you go to a hotel, a restaurant or every. single. thing. you do? Yes, that’s Medallia. Medallia is the perfect example of a software-as-a-service (SaaS) success story (until now). It was founded right after the dot-com bubble burst. It grew quite rapidly, especially during the zero-interest rate policy of the 2010s and went public in 2019 at an [IPO valuation](https://www.prnewswire.com/news-releases/medallia-announces-pricing-of-initial-public-offering-300887869.html?ref=hypertext.fyi) of $2.5 billion. In 2021, right in the middle of the Covid-era software valuation frenzy, private equity firm Thoma Bravo [acquired](https://www.thomabravo.com/press-releases/medallia-to-be-acquired-by-thoma-bravo-for-6.4-billion?ref=hypertext.fyi) Medallia for $6.4 billion (including $5.1 billion of equity). Since then, it’s been a SaaS zombie. Still generating revenue but not really growing, with billions of dollars in debt. Back in April, Reuters [reported](https://www.reuters.com/business/thoma-bravo-nears-agreement-turn-software-firm-medallia-over-creditors-source-2026-04-22/?ref=hypertext.fyi) that Thoma Bravo was trying to restructure Medallia’s debt. And now, the penny has dropped. Or rather, the $5 billion write-off, the second-largest wipeouts in the history of the private equity industry. According to [the Financial Times](https://www.ft.com/content/ae4b6c77-9b3d-46fb-a7fc-59f072e7291b?syn-25a6b1a6=1&ref=hypertext.fyi), Thoma Bravo wrote down its investment to virtually zero. The PE firm is handing Medallia to its lenders (Blackstone, Apollo, KKR…), who are taking control of the software company. Thoma Bravo will be fine. But growth investors must be worried. Software companies haven’t performed well on public markets (with some exceptions like Datadog and Twilio). Private equity was perceived as a good alternative to an IPO for aging software companies that generate real revenue. But that door seems to be closing too. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### AI’s distribution dilemma URL: https://hypertext.fyi/ais-distribution-dilemma/ Last updated: 2026-06-12T09:42:44.000Z *Hi everyone, I hope you’ve had a nice week so far. I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### Apple’s credibility deficit with EU regulators As a company, when is it fine to publicly express that you’re mad as hell? Apple has decided that this week is a good time. More specifically, it’s a good time to tell European citizens that the Digital Markets Act (DMA) is stupid and, by extension, the European Commission. To give you a bit of context, Apple held its annual developer conference earlier this week and unveiled iOS 27, the next version of the iPhone operating system that will be released in September. The big breakout feature was [Siri AI](https://www.apple.com/newsroom/2026/06/apple-introduces-siri-ai-a-profoundly-more-capable-and-personal-assistant/?ref=hypertext.fyi) (not to be confused with Siri not AI). Apple is launching its own ChatGPT-like AI assistant in the form of an app. You can also access it from anywhere by swiping down from the top of the screen or by triggering the Siri voice assistant. It can easily find and reuse information from your messages, photos, emails and calendars that are already on your device. And it looks genuinely useful. > Quick aside: people always say “I don’t download apps anymore” but ChatGPT, Claude and Gemini dominate the charts for most downloaded apps and these apps have only been around for three years or less. So Apple made the right choice by creating a new app for Siri AI. It’s the right entry point if they want to compete seriously with other AI apps. Coming back to Apple being mad. The company also announced that Siri AI is not going to launch in the EU due to the DMA. “We’re deeply disappointed that our EU users won’t have Siri AI on iPhone or iPad when we share our new software releases later this year,” Craig Federighi said in [a press release](https://www.apple.com/newsroom/2026/06/due-to-dma-siri-ai-delayed-in-eu-for-ios-27-and-ipados-27/?ref=hypertext.fyi). “Their refusal to engage constructively on solutions that preserve privacy and security means we do not currently have a timeline for Siri AI’s availability on iOS and iPadOS in the EU.” First of all, in Apple lingo, this sort of quote is extremely unusual. The company is better known for its public restraint and answering “no comment” to basically any question from reporters. So this very salty press release stands out. Let me take a deep breath and give you a one paragraph recap on the DMA. The DMA was created to give equal access to platforms that are so big that the EU called them “gatekeepers” so that other companies could compete fairly with the platform owner. The list includes many American tech giants (Google, Apple, Amazon, Meta, Microsoft), also the company behind TikTok (ByteDance) and Booking.com because hotel chains have a thing or two to complain about Booking.com *and* it’s a European company (so it’s not just American companies and TikTok, good for the EU narrative). With Siri AI, under the DMA, Apple has to grant similar access to your personal data (messages, photos, calendars…) so that Gemini or Meta AI can compete fairly with the built-in AI assistant. Some context, Google and Meta respectively generate 70% and 98% of their revenue from advertising. In other words: Apple has a point. Letting Google and Meta access all your data would turbocharge the algorithmic-based advertising model that these two companies leverage with billions of people. It would also make their social platforms even more efficient at capturing everyone’s attention. More importantly, it would spawn other anti-competitive issues with Google and Meta. So Apple is right to tell European users that the DMA is inappropriate for the current state of technology. However, the context doesn’t play in Apple’s favor. Apple has been extremely dismissive with the European Commission for several years. They didn’t want to open up iOS to other app stores and ended up creating a convoluted platform fee, which at first included a controversial €0.50 per install fee above 1 million installs (imagine if European governments had to pay a fee to Apple for social security app downloads *from third-party app stores*). They blamed the DMA to justify the fact that Apple Intelligence (notification summaries, writing tools, Genmoji, etc.) wouldn’t be available in the EU. They ended up [rolling out](https://techcrunch.com/2024/10/28/apple-intelligence-is-coming-to-the-eu-in-april-2025/?ref=hypertext.fyi) Apple Intelligence in the EU once they added support for languages other than English. So it was a technical issue disguised as a policy issue. And the list goes on. So the Commission’s frustration with Apple didn’t appear out of nowhere. Apple says they comply with the exact legal obligations even though the EC expects them to comply with the interpretation of the DMA so that it achieves the DMA’s objectives. With Siri AI, Apple is right on the substance, but has burned so much credibility with EU regulators that nobody wants to give it the benefit of the doubt. So I’m not surprised the EC is not cooperative and Apple ended up in a dead end. By releasing this week’s statement, they’re hoping the public sentiment against the DMA will change. And I don’t think the DMA is perfect. It feels inappropriate for the current AI era. But I’m not sure EU citizens want to side with Apple either. So, for now, Apple is stuck with it. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/Box---Access-verified.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) Keep your critical business content centralised and compliant. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) provides enterprises with a secure, flexible content layer to connect teams, streamline workflows, and safely leverage AI using your own data. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/glyph-1.png) ### What is WhatsApp exactly? In other antitrust news, the European Commission has [told Meta](https://apnews.com/article/meta-whatsapp-european-union-8b4b48acb23acf5686f141a276bb868d?ref=hypertext.fyi) that it should restore access to WhatsApp for rival AI assistants (this is unrelated to the DMA, btw). If you’re an intense WhatsApp user, you may have noticed a new purple banner and button at the bottom of your conversation list that nudges you to start a conversation with Meta AI, the company’s own AI assistant. Meta is using one of its best distribution channels (WhatsApp) to drive adoption of its new AI service. But for quite a bit of time, Meta AI wasn’t the only AI assistant available on WhatsApp. OpenAI introduced ChatGPT on WhatsApp in [December 2024](https://techcrunch.com/2024/12/18/openai-brings-chatgpt-to-your-landline/?ref=hypertext.fyi). Perplexity and Microsoft also launched their own AI assistants and paid API usage fees to interact with WhatsApp users. A year later, Meta [banned](https://techcrunch.com/2025/10/18/whatssapp-changes-its-terms-to-bar-general-purpose-chatbots-from-its-platform/?ref=hypertext.fyi) general-purpose chatbots from its platform, leaving Meta AI as the only major consumer AI assistant integrated directly into WhatsApp. Not only the EU wants to restore access to competitors, but it says access should now be free. Meta is appealing and I’m sure there will be more ramifications down the road. But it proves once again that distribution is key to capture this relatively new market of consumer AI chatbots. While ChatGPT now has [one billion](https://www.reuters.com/technology/chatgpt-app-hits-1-billion-monthly-active-users-record-time-data-shows-2026-06-02/?ref=hypertext.fyi) monthly active users, Google’s Gemini is not far behind despite the late start. And Apple’s Siri AI will jump from 0 to hundreds of millions of users with an iOS update. So it’s interesting to see big tech companies using all the levers they can pull to control the next platform shift. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The “get rich quick” AI moment URL: https://hypertext.fyi/the-get-rich-quick-ai-moment/ Last updated: 2026-06-09T07:02:28.000Z *Hi everyone, I can’t believe it’s June already. Which means summer is right around the corner! I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### The “get rich quick” AI moment I started covering cryptocurrencies in 2012\. (It turns out that one of my articles is even used as a reference on the “History of bitcoin” Wikipedia page.) At first, it was extremely fun to interview wildly unconventional people who wanted to destroy the banking system. It’s always nice to talk with people who have a very different worldview. Then, the vibe shifted. In addition to techno-anarchists, idealists and libertarians, a lot of people started looking at cryptocurrencies as a way to get rich quick. And I’m not talking about the entrepreneurs working on crypto companies or the developers creating decentralized autonomous organizations (DAOs). I’m talking about the gurus, the scammers, the crypto influencers. A question that people have asked me regularly over the past few months is: how do you feel about [Polsia](https://polsia.com/?ref=hypertext.fyi), the startup founded by Ben Cera (also known as Ben Broca)? It’s hard not to draw parallels with the crypto vibe shift. Ben Cera recently [announced](https://www.linkedin.com/posts/benbroca%5Ftoday-im-announcing-that-polsia-raised-30m-ugcPost-7463621225169211393-3vFx/?ref=hypertext.fyi) that it has secured $30 million in funding for an AI startup that runs itself. He claims that the company is on track to generate $10 million in revenue over the next 12 months *and* that Polsia raised the funding round by itself. “That’s not a marketing line. It’s the product,” he wrote. Or maybe that’s not what he wrote. It’s what his AI agent wrote. Where it gets extremely weird is that Polsia’s whole product is letting people create autonomous companies (or, at least 80% autonomous according to the company’s pitch deck, but Polsia wants to bring this metric closer to 100%). AI agents handle research, create a web app, take care of email marketing and social media ads, etc. In other words, Polsia’s pitch is: sign up to our services and get rich without doing anything. It’s hard to say no to such an incredible promise. Well… you have to enter your card information first. There’s a Polsia subscription fee, then you can pay more to unlock task credits, then you can set a budget to buy online ads… Then Polsia takes 20% of every dollar in revenue you generate through the platform. And finally, you can withdraw money from your Polsia company! Well… there’s a withdrawal limit of $500 per calendar month. Here are some of the apps on Polsia that I found in a few clicks on the company’s [live dashboard](https://polsia.com/live?ref=hypertext.fyi): - [CarePilot](https://carepilot-12.polsia.app/?ref=hypertext.fyi): a voice AI agent for doctors to manage appointments. There’s no contact form, no phone number, no way to get in touch. - [ScaleOS](https://scaleos-17.polsia.app/?ref=hypertext.fyi): an AI transformation company that can help other companies use AI agents in their organizations. Again, it’s just a static page. Pretty sure [that name is taken](https://scaleos.ai/?ref=hypertext.fyi) too. - [ResumeForge](https://resume-forge-4.polsia.app/?ref=hypertext.fyi): a product that rewrites resumes so that they pass automated AI screenings. This time there’s some pricing information but still no way to sign up or contact anyone. Maybe I’m missing something, but this isn’t the most convincing metrics dashboard. But where Polsia shines is with storytelling. A single person who can build an AI company that generates $10 million in revenue is incredible. However, Ben Cera is stretching the definition of “solopreneurship,” as he admitted in [an interview with Fortune](https://fortune.com/2026/03/26/the-one-person-unicorn-myth-miracle-future-of-startups-polsia/?ref=hypertext.fyi). He said that he works with other people. They’re just not on Polsia’s payroll. But let’s assume Polsia takes off like a rocket ship and actually works. In a world with millions of Polsia companies spamming people via email and Instagram ads to find customers, it doesn’t scale. Ad inventories are limited, software budgets are limited, attention is limited. The paradox is that Polsia only works if very few people use it. If millions of people are deploying autonomous companies competing for the same customers, acquisition costs rise, margins disappear and the advantage gets arbitraged away. The only Polsia company that might end up working is Polsia itself thanks to its perfectly crafted narrative. Maybe some of the VC firms are going to get a return too. They’re just playing the pyramid game, hoping another VC firm will invest at a higher valuation down the road and maybe even repurchase their stake. Maybe some of them even noticed that “Polsia” spelled backwards is “AI slop.” SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/Box---Access-verified.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) Keep your critical business content centralised and compliant. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) provides enterprises with a secure, flexible content layer to connect teams, streamline workflows, and safely leverage AI using your own data. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=paidinfluencer&utm%5Ftheme=icm&utm%5Fcampaign=romaindillet&utm%5Fkeyword=june) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/06/glyph.png) ### Hardware is hard, fintech too Wise (the company formerly known as TransferWise) is the latest fintech to face an investigation over its anti-money-laundering controls. Over the years, Revolut, Monzo, Starling Bank and N26 have all faced regulatory scrutiny, resulting in restrictions, fines or both. This time, according to [The Bureau of Investigative Journalism](https://www.thebureauinvestigates.com/stories/2026-06-01/money-transfer-giant-wise-investigated-for-half-a-billion-in-suspicious-transactions?ref=hypertext.fyi), prosecutors in Belgium have opened an investigation into Wise for half a billion euros in suspicious transactions. The company uses Belgian bank accounts for all its European activities. Whether you live in Spain, Germany or Estonia, if you open an account and a balance labeled in EUR, you get banking details in Belgium. “We are currently working with the Brussels prosecutor to respond to queries about our business, as we routinely do with regulators and law-enforcement authorities,” Wise told TBIJ. This is interesting as Wise has often been perceived as the rigorous fintech company with strong processes. But handling money at scale can be hard. A few years ago, La Poste [received](https://banque.meilleurtaux.com/frais-bancaires/actualites/2019-janvier/l-acpr-sanctionne-la-banque-postale-d-une-amende-de-50-millions-d-euros.html?ref=hypertext.fyi) a €50 million fine for failures related to anti-money-laundering controls involving billions of euros in transactions. And I wouldn’t call La Poste a startup. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The metered AI era URL: https://hypertext.fyi/the-metered-ai-era/ Last updated: 2026-05-29T09:31:04.000Z *Hi everyone, I hope you had a nice week despite the intense heat wave that is currently happening. I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### The business model finally caught up with the AI model It turns out companies can raise tens of billions of dollars before finding so-called ‘product-market fit.’ Or, at least, that’s [Simon Willison’s theory](https://simonwillison.net/2026/May/27/product-market-fit/?ref=hypertext.fyi). According to him, OpenAI and Anthropic have *just* found product-market fit (in April 2026!). Before today, those two AI behemoths were just fiddling around to find a product that sells itself. Given the large product catalogs of the two AI labs, you might be wondering what are these miraculous products capable of generating billions in revenue. And yes, you guessed correctly! It’s Claude Code and OpenAI’s Codex. More precisely, it’s those agentic coding products deployed with enterprise clients willing to switch to usage-based pricing. Anthropic is about to close its first [profitable](https://www.wsj.com/tech/ai/mind-blowing-growth-is-about-to-propel-anthropic-into-its-first-profitable-quarter-7edbf2f4?ref=hypertext.fyi) quarter ever. This seems weird, given that developers can’t stop talking about Anthropic’s compute shortage. But if you run a bakery, and you run out of croissants by 10 in the morning, you have some pricing power in your hands. You can also safely increase your croissant batches. And that’s precisely what Anthropic has been doing over the past month: 1. Enterprise customers renewing their contracts now pay $20/seat/month *plus tokens* instead of a more traditional SaaS-like fixed monthly price. OpenAI [moved](https://help.openai.com/en/articles/20001106-codex-rate-card?ref=hypertext.fyi) Codex to token-based pricing in April, too. Some companies are now paying hundreds, sometimes $1,000+, per software engineer on Claude Code or Codex. 2. Anthropic has been investing as much as possible in compute capacity. 3. The more compute it can secure, the more revenue it generates. So I’m sure Anthropic won’t stay profitable for long… They’ll most likely reinvest all that money in more infrastructure, given how the company’s revenue chart basically looks like a pro-level climbing route (i.e., vertical). SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/Box---Built-to-remove-friction.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) Huge amounts of business content remain ungoverned. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) brings structure, security, and control to your unstructured data, enabling secure, enterprise-ready AI. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/glyph-1.png) ### The agent is eating the chatbot What about you, ChatGPT? OpenAI is the clear leader in consumer AI with [900 million weekly active users](https://openai.com/index/scaling-ai-for-everyone/?ref=hypertext.fyi). My back-of-the-envelope calculation still puts it above Codex in terms of revenue. But Codex is catching up pretty quickly. But more important than pure numbers, the vibe is changing at the company. Based on conversations I’ve had, Codex used to be more or less a startup within a startup, with a separate team trying to ship new features as fast as possible without having to worry too much about ChatGPT’s massive installed user base. And it’s been working so well that the Codex team is gaining internal political power within OpenAI. Thibault Sottiaux, the former head of Codex, [now leads](https://www.wired.com/story/openai-reorg-greg-brockman-product/?ref=hypertext.fyi) product and platform across: - Consumer (ChatGPT). - Enterprise (historically ChatGPT Enterprise, now increasingly Codex). - Developer products (historically APIs, now increasingly Codex + APIs). The product teams behind ChatGPT, Codex and APIs are also being merged. My reading is that the Codex team is slowly taking over OpenAI’s entire product portfolio. In a not-so-distant future, people might end up saying “OpenAI, the company behind Codex” instead of “OpenAI, the company behind ChatGPT.” Mistral is reaching the same conclusion from the opposite direction, as it already focuses heavily on B2B enterprise clients. This week, Mistral held its [first conference](https://mistral.ai/news/ai-now-summit-2026/?ref=hypertext.fyi) here in Paris. Other than some big new clients (Airbus, EDF and BMW) and a new data center in Les Ulis (next to where I grew up), Mistral announced a product shift that is right in line with OpenAI’s moves. Until this week, Mistral had a consumer/prosumer product called Le Chat and a developer-centric product called Vibe with a command-line interface (CLI). Mistral is merging all of those products into a single brand called Vibe. Le Chat, as a brand, is gone. In addition to this naming adjustment, Mistral is adding support for skills and automations in a separate “Work Mode.” So now you have Vibe (the chatbot), Vibe’s Work Mode (a bit like Claude Cowork) and the Vibe CLI for developers (and a new VS Code plugin). This new product positioning should work well for signing new customers who have yet to use AI at scale. But will that be enough to convince large enterprise customers to switch from Claude to Vibe? Sponsor Hypertext I believe good information should remain free and accessible. That’s why this newsletter is free. Become Hypertext’s next sponsor and reach founders, operators and investors who want to understand where the tech industry is going. [Contact me ](mailto:romain@hypertext.fyi) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/glyph-2.png) ### Europe redraws the marketplace line In non-AI news, Temu just became a test case for Europe’s Digital Services Act (DSA) as the European Commission [fined](https://www.reuters.com/world/china/temu-fined-232-million-breaching-eu-rules-sale-illegal-products-2026-05-28/?ref=hypertext.fyi) the Chinese shopping app €200 million for failing to properly assess and reduce the risk of illegal products being sold on its marketplace. A few months ago, Shein made a deal with the iconic BHV department store here in Paris. Reporters quickly [pointed out](https://www.leparisien.fr/economie/shein-le-gouvernement-menace-dinterdire-lacces-au-site-apres-le-scandale-des-poupees-sexuelles-03-11-2025-MQZLLASPJ5AOXKQWKBI22X4F5E.php?ref=hypertext.fyi) that you can find extremely dubious items like sex dolls with childlike features, machetes and other weapons. So I’m expecting more fines pretty soon. More importantly, the bigger point is around the DSA. The law doesn’t make marketplaces automatically liable for every illegal product sold by a third-party seller. But Brussels is clearly testing a tougher idea: if your marketplace design, recommendation engine and seller controls make illegal products easy to distribute at scale, you can’t hide behind the idea that you’re just a neutral shelf. In practice, that starts to look a lot like responsibility for what gets sold there. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The developer’s melancholy URL: https://hypertext.fyi/the-developers-melancholy/ Last updated: 2026-05-22T08:04:23.000Z *Hi everyone, I’m back in Paris just in time for this warm, summer-like weekend ahead. I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) *once again for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### Developers as train dispatchers There’s an old saying that technological shifts happen gradually, then suddenly. And it feels particularly apt when talking with software developers. Since the release of Anthropic’s Claude Opus 4.5 model in November 2025, many developers feel like they got a new job. Instead of writing code directly, they increasingly write instructions for coding agents that generate the code for them. If coding agents are the trains, developers have become the train dispatchers. They assign tasks, monitor progress, fix failures, and coordinate dozens of parallel workflows. And don’t just take my word for it. This week, Niklas Gustavsson, Spotify’s Chief Architect and VP of Engineering, [gave an interesting talk](https://www.youtube.com/watch?v=zFslvuvYifQ&ref=hypertext.fyi) at an Anthropic event reviewing the company’s shifting development strategy. “AI tools in general have gone completely bananas. And today more than 99% of our engineers use AI coding tools every week,” he said. Quick aside: I’ve always paid attention to Spotify’s architecture work because they’ve regularly been extremely innovative. For instance, when Spotify was still a small Swedish company, they used peer-to-peer transfers to optimize server bandwidth. When a user hit play on a popular song, Spotify would try to download the track from someone else’s computer before downloading it from the company’s servers. They didn’t talk about it too much because peer-to-peer transfers had a bad reputation due to Napster, Kazaa, LimeWire, eDonkey2000 and (eventually) BitTorrent. That way, they didn’t have to constantly explain: “No but this is *legal* peer-to-peer!” Today, Spotify ships 76% more updates to the code base (pull requests, or PRs for short) compared to last year. “Now, by far, most of the PRs that we ship are authored by an AI agent together with the developer,” he said. Every month, AI agents generate over a thousand pull requests. Right now, a human developer looks at the agent’s work and validates the pull request. But this is Sisyphean work. AI never gets tired. Humans do. “One of our most frequent feedbacks at the moment is there’s just too many freaking PRs to review. So we need to figure out where we apply humans to review those PRs where it matters the most,” Niklas Gustavsson said. “We’re already auto-approving some PRs that we think are safe enough to merge without human review and then we try to focus the human review where it really matters.” And this may be the most important sentence in Gustavsson’s entire talk: **“We'll figure out over time where we need the human judgment to be applied.”** I used Spotify as an example, but I could just as easily have referenced Mistral’s Arthur Mensch testimony at the French National Assembly (Chris O’Brien has [a good write-up](https://www.frenchtechjournal.com/mistral-ai-ceo-mensch-to-french-lawmakers-europe-has-two-years-to-stop-losing-the-ai-race-before-the-race-is-over/?ref=hypertext.fyi) of the testimony if you want more on this). “Today, engineers at Mistral no longer write lines of code,” the CEO of Mistral said. “In the past, \[software development\] was more of a craft if you were an individual contributor. That is to say, you wrote your own code. In fact, people enjoyed that craftsmanship,” he added. “Today, you’re no longer a craftsperson; you’re a manager, and you ask agents to write the code for you. You provide the specifications, you’re the one giving orders, and that’s a pretty profound change.” These days, a lot of developers talk about potential burnout from AI tools for a combination of reasons: - The obvious part: they now have so many agents to manage and so much code to review. - The pernicious vibe: they feel like they have to outperform their peers to remain relevant as big companies like Meta and Amazon are laying thousands of employees (approximately 8,000 jobs for Meta this week alone). Developers often tell me that the current AI shift is “a rat race.” - The side effect: due to those huge productivity jumps, more faulty code gets pushed into production. When AI breaks things, you call a human for an emergency code review. But one thing that comes up less often, and may ultimately be even more heartbreaking, is what I’d call developer melancholy. Many developers loved getting in the zone and writing hundreds of lines of code in a single session to solve a Very Hard Problem. When you’re spending your days talking with agents, it’s just not the same. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/Box---Built-to-remove-friction.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) Huge amounts of business content remain ungoverned. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) brings structure, security, and control to your unstructured data, enabling secure, enterprise-ready AI. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/glyph.png) ### Tokens as the universal currency A few years ago, I wrote an [AI glossary](https://techcrunch.com/2026/05/09/artificial-intelligence-definition-glossary-hallucinations-guide-to-common-ai-terms/?ref=hypertext.fyi) with my friend (and former colleague) [Natasha](https://www.linkedin.com/in/nlomas/?ref=hypertext.fyi). Here’s how we described tokens: they are the basic building blocks of human-AI communication, representing discrete segments of data that have been processed or produced by an LLM. They are created through a process called tokenization, which breaks down raw text into bite-sized units a language model can digest, similar to how a compiler translates human language into binary code a computer can understand. In enterprise settings, tokens also determine cost — most AI companies charge for LLM usage on a per-token basis, meaning the more a business uses, the more it pays. What I didn’t include in that definition is that tokens are increasingly becoming a global currency with geopolitical consequences. Mistral CEO Arthur Mensch articulated that idea pretty well in his testimony at the French National Assembly. “We’re in a business where we convert electricity into token generation. And fundamentally, this is something that needs to be treated more or less like a natural resource,” he said. “In other words, we need to think about intelligence the same way we think about energy. So our goal is to provide intelligence that is affordable, secure in terms of supply — meaning we don’t have to source it from the U.S., for example — and that is also sustainable because it transforms energy with a lower carbon footprint than elsewhere, since we use French energy in particular.” Later in the conversation, he even went one step further and roughly translated a gigawatt of electricity into a corresponding amount of tokens. Those tokens can then be monetized (with a margin, hopefully) through different business models: - Hosted AI inference APIs (Anthropic’s strategy with Claude Code). - Raw GPU capacity (SpaceX’s strategy with Colossus 1). - Enterprise AI transformation projects (Mistral’s strategy with its services-led enterprise customers). But here’s a fourth idea few people had considered. OpenAI CEO Sam Altman is [offering](https://techcrunch.com/2026/05/20/sam-altman-makes-mic-drop-offer-to-every-y-combinator-startup/?ref=hypertext.fyi) $2 million worth of OpenAI tokens to every startup in the current Y Combinator batch in exchange for equity in the startups (through an uncapped SAFE agreement that would convert into equity at the Series A round). Yes, you read that right. **OpenAI could end up owning stakes in hundreds of buzzy startups without spending a dollar from its cash balance.** Taken together with OpenAI’s convoluted deals with Microsoft, Nvidia, AMD, Oracle and others, this suggests Sam Altman may be an AI innovator, but he is also proving to be an even more important innovator… in startup financing. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The art of the second act URL: https://hypertext.fyi/the-art-of-the-second-act/ Last updated: 2026-05-15T08:46:30.000Z *Hi everyone, I’m writing this from the lovely city of Bristol where I’m spending a few days. Before we jump into this week’s newsletter, I’d like to thank* [*Box*](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) *for supporting Hypertext as its very first sponsor. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### Why AI is pushing startups to launch second products Many years ago, a startup founder and a good friend of mine told me that it’s much easier to launch a second product when you already have one successful product. You’re not starting from scratch as you already have an audience. The second product can also start as an add-on and branch out to become its own standalone product. In 2026, AI has made second-product launches more tempting because incumbents already have workflows, users, trust, data, and distribution. Launching a second product seems inevitable for a variety of reasons: - Successful Software-as-a-Service (SaaS) companies are suddenly under pressure. They are generating revenue but many of them are on the verge of accelerated obsolescence. They need… something else. - Launching a second product is easier than ever due to agentic coding. If you have an existing code base that you can leverage without having to redefine data formats, user accounts and all those pesky things, you can speedrun your way to a good minimum viable product. - And the most important part: distribution is becoming more important than ever. If several companies can build the same product at the same time, your ability to stand out from the crowd is extremely important. On that last point, let’s take voice transcription apps as an example. Many of you probably know [Wispr Flow](https://wisprflow.ai/?ref=hypertext.fyi). On your computer or your phone, you press a button, talk naturally, and text magically appears on your screen with an unprecedented level of accuracy. The company is reportedly raising at a [$2 billion valuation](https://www.bloomberg.com/news/articles/2026-05-12/ai-dictation-startup-wispr-in-funding-talks-at-2-billion-value?ref=hypertext.fyi). But there’s nothing magical behind Wispr Flow as a product. There are many excellent speech-to-text AI models out there (OpenAI’s Whisper models, Nvidia’s Parakeet models, more recently Cohere Transcribe). That’s why there are *many* Wispr Flow alternatives: [Willow Voice](https://willowvoice.com/?ref=hypertext.fyi) (a Y Combinator copycat), [Monologue](https://www.monologue.to/?ref=hypertext.fyi) (from [Every](https://every.to/?ref=hypertext.fyi)), a bunch of bootstrapped companies ([Aqua Voice](https://aquavoice.com/?ref=hypertext.fyi), [Superwhisper](https://superwhisper.com/?ref=hypertext.fyi), [VoiceInk](https://tryvoiceink.com/?ref=hypertext.fyi)…). Even OpenAI created its own dictation feature that you can use in any app. It is currently an optional feature in Codex that you can activate it in the app settings. Wispr Flow is a reminder that the hard part is not always building the underlying technology. The hard part is packaging it, distributing it, and becoming the default. And they’ve excelled at that. Coming back to second products, two European unicorns have turned to this diversification strategy. [Deel](https://www.deel.com/?ref=hypertext.fyi), the company that helps you hire remote workers without opening a subsidiary, launched [Akai](https://www.akai.run/?ref=hypertext.fyi). It started as an internal tool to create AI agents that connect with legacy banking systems. Many employees at Deel currently spend countless hours submitting payments across multiple banking portals, reconciling payments and more. Akai captures what Deel’s employees do in a web browser and builds universal connectors so that it can do the same tasks automatically the next time. And as Deel handles sensitive data and large sums of money, Akai has been designed with compliance and audit trails in mind. In other words, Deel is not just automating its own back office. It is turning that automation layer into a product. Revolut is also branching out beyond its banking roots with [GlobalHire](https://www.revolut.com/business/globalhire/?ref=hypertext.fyi), its own take on remote hiring and payroll. Instead of working with Deel, Rippling or Remote.com, Revolut has recreated its own alternative to Deel for its global workforce. In other words, Revolut is not just solving an internal HR problem. It is turning that operational need into another product surface. When you’re an executive at a large tech company, you often ask yourself whether you should build or buy a product. Maybe the question now is: build, buy or build and sell. SPONSORED by Box [![CTA Image](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/Box---Built-to-remove-friction.jpg)](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) Huge amounts of business content remain ungoverned. [Box](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) brings structure, security, and control to your unstructured data, enabling secure, enterprise-ready AI. [Learn more ](https://www.box.com/fr-fr/smartercontent?utm%5Fsource=newsletter&utm%5Fmedium=emeapaidinfluencer&utm%5Ftheme=contentmodernization&utm%5Fcampaign=romaindillet&utm%5Fkeyword=hypertext) ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/-glyph-2.png) ### The case for European companies buying Chinese AI startups A few weeks ago, Chinese authorities decided to block Meta’s acquisition of AI startup [Manus](https://manus.im/?ref=hypertext.fyi). A lot of articles mention Manus like it’s a mainstream product but maybe it’s worth spending a minute talking about what Manus actually does, right? Originally developed in China, Manus is an AI agent workspace that can handle complex, multi-step tasks and create presentations, design websites or build a research report. It’s a bit like OpenClaw, but they come from opposite philosophies. OpenClaw is a local-first AI agent framework. It runs on your own Mac Mini and interacts with your local apps to do things that you simply can’t do with ChatGPT. Manus also lets you create things that are impossible to make with ChatGPT, but through a highly-polished cloud-first interface. As Meta was about to acquire Manus, the Chinese company moved to Singapore. And it wasn’t just a headquarters change. [Around 100 employees](https://www.techinasia.com/news/manus-shifts-hq-singapore-cuts-china-jobs?ref=hypertext.fyi) literally moved to Singapore. And yet, that wasn’t enough to stop China from cracking down on Meta’s acquisition. Singapore-washing didn’t work. And it brings me to this important geopolitical conclusion: maybe it’s time for Europe to take advantage of the ice-cold relationships between the U.S. and China. It’s time to start some M&A action with some promising AI companies based in China. Would Beijing react differently if the buyer were European rather than American? Probably. Sure, SAP is [investing in n8n](https://www.bloomberg.com/news/articles/2026-05-12/sap-invests-in-ai-automation-startup-n8n-at-5-2-billion-value?ref=hypertext.fyi) and acquiring [Prior Labs](https://techcrunch.com/2026/05/05/sap-bets-1-16b-on-18-month-old-german-ai-lab-and-says-yes-to-nemoclaw/?ref=hypertext.fyi), two extremely promising German AI startups. But European incumbents should also look at the other side of the Eurasian continent. If U.S. buyers are becoming politically toxic for Chinese AI companies, there might be a narrow window for European companies to invest in, partner with, or acquire the next Manus. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/-glyph-3.png) ### A few words on layoffs Unfortunately, these past few days, [Cloudflare](https://blog.cloudflare.com/building-for-the-future/?ref=hypertext.fyi), [DeepL](https://sifted.eu/articles/deepl-cuts-250-jobs?ref=hypertext.fyi) and [GitLab](https://about.gitlab.com/blog/gitlab-act-2/?ref=hypertext.fyi) all announced massive layoffs. These companies are in very different positions and aren’t cutting jobs for the same reason. GitLab and DeepL are both streamlining their organizations, reducing costs, refocusing, etc. They are also both competing with products that are gaining steam with the current AI wave: GitHub (and Hugging Face) for GitLab, and ChatGPT (and Claude) for DeepL. Most people around me switched from DeepL to using ChatGPT to translate big walls of text, even though DeepL is clearly the superior product for that task. As for Cloudflare, it is investing more in AI tokens (+600% in 3 months) and less in people (-1,100 people). It is one of the first companies to admit publicly that AI spend isn’t just software spend. It’s an operational expense, just like salaries. It’s hard to predict exactly if this will keep happening if AI labs raise token prices to stop burning cash. So when an HR person next asks you what’s your expected salary range, maybe you should ask whether they want an answer in euros or tokens. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### Why robotic hands matter URL: https://hypertext.fyi/why-robotic-hands-matter/ Last updated: 2026-05-28T10:05:06.000Z *Hi everyone, I hope you’re having a good week so far. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### The upper hand When AI researcher Yann LeCun left Meta and went on a press tour to build momentum for [AMI Labs](https://amilabs.xyz/?ref=hypertext.fyi), one of his talking points was that Silicon Valley had become “LLM-pilled.” In other words, the AI industry was spending too much time, energy and money on large language models (LLMs) instead of thinking about the next major technological breakthrough. A few months ago, I had a brief discussion with him and we ended up talking more about Philippe Aghion’s view on technological progress rather than the latest LLM release. According to him, the solution lies in world models, AI models that have internalized the physical world’s structure and dynamics. LeCun’s argument about LLM rhymes with Philippe Aghion’s Schumpeterian view of innovation: the real economic impact of AI will depend less on squeezing more out of today’s dominant paradigm than on whether new waves of innovation emerge and diffuse through the economy at a fast pace. Of course, this is nothing new. The tension between long-term technological progress and short-term hype is older than Silicon Valley. And given the pace of innovation around LLMs these days and my limited scientific understanding of world models, I don’t even know who’s right and who’s wrong. (And that’s fine.) But it brings me to a broader point. When one area creates so much noise, it becomes harder to notice innovation happening elsewhere. Which brings me to robotic hands. So far, the humanoid robotics narrative has been over-indexed on walking demos, while the commercial bottleneck may be manipulation. This week, Chinese robotics startup Linkerbot [announced](https://www.reuters.com/world/china-robot-hand-building-unicorn-linkerbot-targets-6-billion-valuation-2026-05-04/?ref=hypertext.fyi) the closing of a funding round that valued the company at $3 billion. It also said it was already raising its next round at a $6 billion valuation. According to the company, it currently holds a 80% market share in “high-degree-of-freedom robotic hands” (read: humanoid hands). The reason hands matter is that it becomes much easier to train AI models for the physical world if robots look and work more like humans. Imagine you run a giant warehouse network and want robots that can pick up items from one shelf and place them on another. One way to collect training data is to equip all your human workers with cameras and record how they perform those tasks (except in Europe due to GDPR). But that data becomes much more useful if the robots you eventually deploy have human-like hands too. The closer the robot is to the human demonstrator, the easier it is to turn repetitive human actions into robotic movements. And as we’ve seen across many industries in the past, whether you think automation is good or bad, it seems inevitable. Linkerbot isn’t the only company paying attention to robotic hands. French startup [Genesis AI](https://www.genesis.ai/?ref=hypertext.fyi) recently [unveiled a demo](https://techcrunch.com/2026/05/06/khosla-backed-robotics-startup-genesis-ai-has-gone-full-stack-demo-shows/?ref=hypertext.fyi) of its first AI robotic model, GENE-26.5, along with the company’s first set of robotic hands. I encourage you to watch the [video demo](https://www.youtube.com/watch?v=6K%5FbGH54ltI&ref=hypertext.fyi) of Genesis AI’s robot cooking, playing the piano and solving a Rubik’s cube. It’s very impressive. When it comes to funding: - Genesis AI raised a $105 million seed round last year. - Also in France, a good chunk of the team that worked on LeRobot at Hugging Face left to start [UMA](https://uma.bot/?ref=hypertext.fyi) (and raised less than Genesis AI so far). There are also a few well-backed American startups exploring the space: - Physical Intelligence, which [is reportedly raising](https://www.bloomberg.com/news/articles/2026-03-27/ex-deepmind-staffers-robotics-startup-in-talks-for-11-billion-valuation?ref=hypertext.fyi) $1 billion at an $11 billion valuation according to my former colleague Natasha Mascarenhas and her new Bloomberg friends. - Skild AI [recently raised](https://techcrunch.com/2025/01/28/softbank-to-invest-500m-in-robotics-startup-skildai/?ref=hypertext.fyi) $1.4 billion. And of course, big tech companies are paying attention too. Tesla seems to be increasingly [pivoting](https://www.youtube.com/watch?v=SQJIyR5uoZY&ref=hypertext.fyi) toward robots. Jeff Bezos is raising billions for [Project Prometheus](https://www.nytimes.com/2025/11/17/technology/bezos-project-prometheus.html?ref=hypertext.fyi), an “AI manufacturing startup” (so… robots?). And just this week, Meta [acquired](https://www.bloomberg.com/news/articles/2026-05-01/meta-acquires-assured-robot-intelligence-to-help-build-humanoid-technology?ref=hypertext.fyi) Assured Robot Intelligence, a company working on AI models for robots. Based on conversations I’ve had with several robotics researchers, robots may become mainstream much sooner than expected, at least in manufacturing, logistics, and other industrial fields. Maybe we’re at a GPT-3 moment, but for robots. OpenAI released the GPT-3 model in 2020\. At the time, AI researchers and developers were already extremely impressed by the broad capabilities of the model ([Hacker News](https://news.ycombinator.com/?ref=hypertext.fyi), the popular developer-centric forum, was filled with GPT-3 demos). But it took OpenAI another couple of years to release ChatGPT, the mainstream application that brought LLMs to the masses. I think we’ll hear much more about AI robotics over the next two years. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/-glyph-1.png) ### The geopolitics of LLMs EU finance ministers met in Brussels earlier this week. One of the topics on the agenda was Mythos, Anthropic’s AI model that is too useful to hide and too risky to simply ship, as I covered at length in [a previous newsletter](https://hypertext.fyi/the-ai-gatekeepers/). “Banks must be prepared for an environment of heightened uncertainty and rapid technological change,” Greek finance minister and Eurogroup president Kyriakos Pierrakakis [said](https://www.consilium.europa.eu/en/press/press-releases/2026/05/04/remarks-by-kyriakos-pierrakakis-following-the-eurogroup-meeting-of-4-may-2026/?ref=hypertext.fyi) in his closing remarks. “It is for this reason that we decided to kick-off a discussion on AI at today’s meeting. Frontier AI models are evolving rapidly and may soon present challenges of a potentially systemic nature. And we must ensure a framework that supports both stability and competitiveness.” Translated from Brussels-speak to real talk, Kyriakos Pierrakakis means two things: - Yes, Anthropic wins this one. European banks will come begging for access to Mythos so they can make sure there are no major security flaws in their systems. - At the same time, the way Anthropic restricted access to American companies is concerning and should not become the default playbook. He was even more explicit with reporters after the meeting, as [Bloomberg reported](https://www.bloomberg.com/news/articles/2026-05-04/eu-in-talks-with-anthropic-to-get-banks-tested-for-mythos-flaws?ref=hypertext.fyi). "Indeed there are contacts with Anthropic,” he said. “I don’t think we have the luxury of not trying to establish channels of communication with the U.S. The challenge here is that technologies like AI necessitate international governance frameworks at a moment when multilateralism is challenged,” he added. What a lovely double negative: “I don’t think we have the luxury of not trying…” Yep, Anthropic forced the EU’s hand and they don’t seem thrilled about it. But what can they do? ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/05/-glyph.png) ### AI agents are coming to EVE Online In lighter news, a quick update on [EVE Online](https://www.eveonline.com/?ref=hypertext.fyi), the massively multiplayer online game that has been around for more than 20 years. If you’re not familiar with EVE, it’s a space game where thousands of players interact on the same server. They mine resources, manufacture ships, trade goods, form corporations, wage wars, and occasionally run scams. This game has always been popular with tech-savvy people because it is often described as a giant economic and political simulation disguised as a video game. Paradoxically, the best part about this digital world is the humans doing very human things. And now, there’s a tech angle as DeepMind, Google’s AI division, [has acquired a stake](https://www.rockpapershotgun.com/eve-online-will-be-used-to-help-train-google-deepminds-ai-tech-as-company-take-a-minority-stake-in-the-former-ccp-games?ref=hypertext.fyi) in the Icelandic company behind EVE Online. The plan is to use the game as a sandbox to evaluate AI models and observe how they behave in a complex but controlled environment. I personally can’t wait to see AI agents doing weird, space-piratey things in EVE Online. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### Aleph Alpha’s ambitions URL: https://hypertext.fyi/aleph-alphas-ambitions/ Last updated: 2026-05-01T10:20:32.000Z *Hi everyone, I hope you’re enjoying a long weekend if May 1st is a holiday where you live (sorry, British friends, but I believe you get Monday off instead?). Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ### What can we learn from Aleph Alpha’s trajectory? In the race to build Europe’s leading large language model (LLM) company, it initially looked like a two-horse race: France had Mistral, Germany had Aleph Alpha. Of course, it’s a bit more complicated than that, as DeepMind was originally a British company but has been part of Google since 2014\. Similarly, Meta/Facebook has always had a large AI research team in Paris with talent that worked on early Llama models, but Facebook isn’t exactly a European startup. But if you only look at independent AI labs based in Europe, Mistral and Aleph Alpha stood out from the crowd pretty quickly. While Mistral managed to raise capital at a regular pace and is still releasing new models every other month (side note: Mistral just announced [Mistral Medium 3.5](https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5?ref=hypertext.fyi)), Aleph Alpha has had a completely different trajectory. After raising a [$500 million Series B round](https://techcrunch.com/2023/11/06/lidl-owner-and-bosch-ventures-co-lead-500m-series-b-into-german-ai-start-up-aleph-alpha/?ref=hypertext.fyi) in 2023, the company [stopped](https://www.bloomberg.com/news/articles/2024-09-05/the-rise-and-pivot-of-germany-s-one-time-ai-champion?ref=hypertext.fyi) work on its own LLMs. It chose to focus on AI integration for large corporate clients and governments. In hindsight, Aleph Alpha made three linked bets: - LLMs would become a commodity very quickly. - Enterprise AI integration looked like the clearest path to revenue. - Controlled deployments with on-premise AI models would win over API-first models from OpenAI, Anthropic and others, in part due to customer control over data. Things didn’t exactly pan out as expected. First, since 2024, there’s been *a lot* of innovation in the LLM space. Sure, open-weight models now lag behind state-of-the-art models by just a few months. But Anthropic and OpenAI are still iterating at a rapid pace. I still believe that LLMs will become a commodity at some point. But when? Second, Aleph Alpha likely found real revenue with large corporate and public-sector clients, including SAP and Bosch. But that is not where most of the value has accrued so far. The companies that are benefiting the most from AI are not system integrators: chipset makers (Nvidia and its suppliers), hyperscalers (AWS, Microsoft Azure and Google Cloud) and large AI labs combining great AI models with great products (OpenAI and Anthropic) are growing much more rapidly. It’s been nearly three years since Aleph Alpha’s latest funding round. The German startup wasn’t in a strong position to raise another $500 million to pursue the same strategy. Hence, last week’s deal with Cohere, the Canadian AI firm. Both companies announced [plans to merge](https://www.theglobeandmail.com/business/article-canadian-ai-firm-cohere-germanys-aleph-alpha-announce-merger/?ref=hypertext.fyi). However, the new company will be called Cohere, Cohere’s CEO Aidan Gomez will remain as CEO and “Cohere is going to remain Canadian headquartered and owned,” according to Gomez. So it really sounds like Cohere is absorbing Aleph Alpha more than anything. Interestingly, a German company played an instrumental role in structuring the deal. That company is Schwarz Group, the parent company behind Lidl and Kaufland. My theory is that Schwarz Group looked at the margins of its supermarkets (low single digits) and compared it to the margins of AWS (a 35% operating margin per Amazon’s most recent [earnings](https://www.investing.com/news/company-news/amazon-q1-2026-slides-aws-surges-28-record-margins-offset-by-capex-93CH-4647447?ref=hypertext.fyi)!). They realized they were in the wrong business… So Schwarz Group wants to become a cloud company and is “investing” $600 million in Cohere. It expects Cohere to use Schwarz Group’s cloud infrastructure going forward, which should help when it comes to recouping its investment in Cohere. In other words, part of the investment is free rent in exchange for a stake in Cohere. As for Aleph Alpha, it may end up as a cautionary tale for other mid-sized AI companies. If you’re not pushing the frontier, it’s hard to capture the upside. And if you don’t own the infrastructure, it’s even harder to capture the margins. ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/glyph-7.png) ### The quiet European winner I mentioned Nvidia and its suppliers as some of the companies that benefited the most from AI in the piece above. Let me expand on that a bit because one of the biggest AI winners does not sell GPUs, cloud credits, or foundation models. It is ASML. Based in the Netherlands, ASML’s extreme ultraviolet lithography machines are among the most sophisticated pieces of industrial equipment in the world. As this excellent [Works in Progress profile](https://worksinprogress.co/issue/the-worlds-most-complex-machine/?ref=hypertext.fyi) from last week explains, these machines use lasers, tin droplets, plasma, and ultra-precise mirrors to print the tiny patterns that make advanced chips possible. If you prefer a long-form video instead of a long read, check out Veritasium’s [YouTube video](https://www.youtube.com/watch?v=MiUHjLxm3V0&ref=hypertext.fyi) on ASML. That sounds abstract until you connect the dots. Nvidia’s success depends heavily on TSMC’s manufacturing prowess. TSMC’s leading-edge processes depend, in turn, on ASML’s continued lithography innovation. That’s how you get tens of billions of transistors packed onto a single chip. And that’s also why ASML is now worth roughly €470 billion. It is not an AI company in the usual sense, but it is one of the companies benefiting the most from AI. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/glyph-8.png) ### The billion dollar bet on reinforcement learning I’m sure you’ve seen some headlines about [Ineffable Intelligence](https://www.wired.com/story/david-silver-ai-ineffable-intelligence-reinforcement-learning/?ref=hypertext.fyi), a London-based AI lab that raised a $1.1 billion seed round. The company was founded by David Silver, the former Google DeepMind researcher behind [AlphaGo](https://en.wikipedia.org/wiki/AlphaGo?ref=hypertext.fyi), the AI program that plays the board game Go, and its successors AlphaGo Zero and AlphaZero. Silver has also said that any money he makes from his Ineffable equity will go to charities, which adds an interesting footnote to an already unusual company formation. But I wanted to spend a minute to talk about Ineffable Intelligence’s technical bet: **reinforcement learning**. In simple terms, reinforcement learning is when an AI system learns by doing. It takes actions in an environment, gets feedback, and gradually improves its strategy. It is less about copying human examples and more about discovering what works through trial and error. That is how AlphaGo (and more specifically [AlphaGo Zero](https://en.wikipedia.org/wiki/AlphaGo%5FZero?ref=hypertext.fyi), which was completely self-taught without using data from human games) became so interesting, as my Go player friend Matthieu once told me. It did not just imitate human Go players. It found moves that humans had not seriously considered. If you’ve paid attention to AI model training techniques, big AI labs like Anthropic and OpenAI already use reinforcement learning. In most current frontier LLMs, reinforcement learning is a post-training step. The model first learns from huge amounts of human-generated text, code and synthetic data. Then reinforcement learning is used to make it more helpful, safer, better at reasoning, or better at following preferences. Ineffable Intelligence’s pitch is more radical. It wants to use reinforcement learning from the very beginning and ditch training data sets altogether, just like a baby learns a new language from scratch without any “pre-loaded information.” Instead, a baby learns from experience in a specific environment. The bet is that the next AI breakthrough will not come from reading more of the internet, but from building better worlds for machines to learn in. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### The AI Gatekeepers URL: https://hypertext.fyi/the-ai-gatekeepers/ Last updated: 2026-04-24T07:13:18.000Z *Hi everyone, I hope you had a nice week so far. This week, I ended up writing a lot about sovereignty, a word that means a lot of different things. So I’m tackling it from different angles. Feel free to share this email with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* It took me a while to understand what Anthropic actually meant when it said the release of Claude Mythos would be limited to a select group of companies and organizations for national security reasons. That sounds reasonable. Cybersecurity is inherently dual-use. The same model that can find a critical vulnerability can help exploit it. But the distribution method quickly becomes political. But there’s a wider point. This precedent is going to greatly affect Anthropic’s (and OpenAI’s) distribution strategy going forward. And it’s concerning for European companies and developers. As a reminder, with [Project Glasswing](https://www.anthropic.com/project/glasswing?ref=hypertext.fyi), Anthropic gave early access to a small group of large technology and security organizations, including Apple, Google, Microsoft, CrowdStrike and JPMorganChase. The company says Mythos has already found thousands of zero-day vulnerabilities. Axios [reported](https://www.axios.com/2026/04/19/nsa-anthropic-mythos-pentagon?ref=hypertext.fyi) that the NSA is also using Mythos Preview. The agency isn’t just using Mythos to ensure that there are no security vulnerabilities in its systems. Mythos is “being used more widely within the department,” as Axios put it. And yet, Europe was mostly left out of the Mythos loop, as Next rightfully [pointed out](https://next.ink/233707/mythos-leurope-tenue-a-lecart-du-modele-ia-le-plus-ambitieux-du-moment/?ref=hypertext.fyi). While the U.K.’s AI Security Institute tested the model, European cyber agencies on the continent appear to have had little or no hands-on access. And yet, the gatekeeping already looks inconsistent. Per [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-21/anthropic-s-mythos-model-is-being-accessed-by-unauthorized-users?ref=hypertext.fyi), unauthorized users even started playing with Mythos. As my former colleague Zack Whittaker [commented](https://mastodon.social/@zackwhittaker/116448013647934279?ref=hypertext.fyi): “Anthropic spent weeks claiming that it couldn't publicly release its Mythos AI model because of its alleged offensive hacking capabilities and… some AI nerds from Discord just found it and accessed it.” (Subscribe to Zack’s [This Week in Cybersecurity](https://this.weekinsecurity.com/?ref=hypertext.fyi) newsletter, it’s good.) To be fair with Anthropic, it isn’t the only AI gatekeeper in town. OpenAI has adopted a similar strategy with its more permissive version of GPT-5.4 (called GPT-5.4-Cyber). It has [created](https://openai.com/index/scaling-trusted-access-for-cyber-defense/?ref=hypertext.fyi) the “Trusted Access for Cyber (TAC) program,” a fancy name to say that the company gets to say who deserves to access GPT-5.4-Cyber and who doesn’t. Both companies are building verification flows, partner programs and government relationships around models that are too useful to hide and too risky to simply ship. And just like pharmaceutical companies put a long list of side effects on medication labels, this new distribution strategy is also a way to shift the blame down to partners for potential misuse. But what is going to happen when security vendors, banks and critical software companies have fixed all the vulnerabilities identified by Mythos? Is Anthropic going to release it publicly as an API endpoint or expand the “preview release” to a larger group of American companies? ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-6.png) Talking about AI sovereignty, Forbes’ Iain Martin (hi Iain!) wrote an [interesting profile](https://www.forbes.com/sites/iainmartin/2026/04/16/how-frances-mistral-built-a-14-billion-ai-empire-by-not-being-american/?ref=hypertext.fyi) on Mistral last week with a useful update. Mistral is no longer positioning itself as the company that will compete with OpenAI and Anthropic on benchmarks. Instead, the company’s pitch is around control and sovereignty: European-built models, open-weight models that can run in your own cloud, data that can stay in the right geography. An early-stage VC recently told me that sovereignty is often plan B for a startup narrative. Plan A is trying to build a world-class company. Plan B is when you start saying “but we’re European!” It’s a great pitch for customers in Europe. But American or Asian companies don’t care about European sovereignty. So you gain something and you lose something. In Mistral’s case, that strategy is working out pretty well. According to Forbes, Mistral generated $200 million in revenue in 2025 and is “on track to start making around $80 million monthly by December” (what a weird metric, I could also say that I’m on track to generate $79 million monthly, just below Mistral). The company is now relying heavily on forward-deployed engineers working with large corporate clients. So Mistral is a real business, even if some Silicon Valley people will dismiss it as a very expensive systems integrator with a French accent. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-7.png) Ok, for real this time, 2026 might be [the year of the Linux Desktop](https://www.reddit.com/r/linux/comments/3038d4/when%5Fwas%5Fthe%5Ffirst%5Fyear%5Fof%5Fthe%5Flinux%5Fdesktop/?ref=hypertext.fyi). France’s DINUM [announced](https://www.numerique.gouv.fr/sinformer/espace-presse/souverainete-numerique-reduction-dependances-extra-europeennes/?ref=hypertext.fyi) that it plans to move some government computers from Microsoft Windows to Linux. I know, don’t try to fix something that isn’t broken… But, as Tariq Krim writes in [his newsletter](https://www.cybernetica.fr/se-preparer-au-decouplage-transatlantique/?ref=hypertext.fyi), governments and companies should prepare for a (technological) decoupling between the U.S. and Europe because it seems like it’s a real possibility given the current geopolitical landscape. The implementation detail is more interesting than the speech. The French government’s [Sécurix](https://github.com/cloud-gouv/securix?ref=hypertext.fyi) project is publicly available on GitHub. It is based on NixOS, supports hardware keys like YubiKeys, and is designed so administrators can write configurations and roll them out to thousands of computers so that everybody gets the same thing. This is not “let’s install Ubuntu on a few laptops and hope everyone enjoys LibreOffice.” It is an attempt to make secure, reproducible government workstations boring enough to deploy at scale. And I’m sure other European countries will pay attention to this implementation. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-8.png) The funny thing about Cursor is that it started as an IDE story and is now turning into an infrastructure story. On Tuesday, SpaceX announced plans to acquire Cursor for $60 billion [later this year](https://www.theverge.com/science/916427/spacex-cursor-potential-deal-acquisition?ref=hypertext.fyi). If SpaceX doesn’t exercise the rights to acquire Cursor, it’ll have to pay $10 billion as a breakup fee. But before we dive deeper into the deal: SpaceX, the rocket company? If you missed previous episodes, Elon Musk’s SpaceX [acquired](https://techcrunch.com/2026/02/02/elon-musk-spacex-acquires-xai-data-centers-space-merger/?ref=hypertext.fyi) Elon Musk’s xAI. (xAI, which also owns X.) So it’s all the same company now. It’s a highly unusual deal for a highly unusual situation. Cursor was a trailblazer in AI-assisted coding with its fork of VS Code. What made Cursor particularly successful is that users could switch between AI models from Anthropic, OpenAI, Google and more — the best model was always just one switch away. In other words, Cursor made a bet on the application layer. AI coding was going to be a product problem, not an AI model problem. When the big AI lab turned their attention to agentic coding, Cursor became less relevant in the AI conversation. The company knew it had to train its own AI models and redesign its product around agentic coding. And this pivot seems to be well underway. Cursor’s Composer seems like a capable agentic coding model. But it’s not the best one. And if they want to stay in the race with OpenAI and Anthropic, they need compute and they need it now. Cursor’s announcement of the deal with SpaceX is [surprisingly short](https://cursor.com/en-US/blog/spacex-model-training?ref=hypertext.fyi) (100 words!) but is right on point. “Cursor is partnering with SpaceX to accelerate our model training efforts,“ it reads. “Each step up in compute has translated to meaningfully more capable models. We’ve wanted to push our training efforts much further, but we’ve been bottlenecked by compute. With this partnership, our team will leverage xAI’s Colossus infrastructure to dramatically scale up the intelligence of our models.” (It’s a great lesson in how you should communicate when the news is so big that people will dissect every word that you publish on your blog. In that case, shorter is better. Deliver the important message and hit publish.) As for SpaceX/xAI, Musk has tons of GPUs and wants to IPO this year. It needs to prove that xAI’s Colossus data center was built for a good reason. Similarly, xAI doesn’t have any agentic coding product while Cursor is already installed on millions of laptops. It’s a GPU + distribution story. That is an uncomfortable lesson for Europe. Many European AI startups have chosen to stay model-agnostic on top of U.S. labs to focus on the application layer. (Or they develop models without any plans to build data centers). The sovereign compute question is no longer just for foundation-model labs. It is coming for every serious AI application company too. As it stands, if the next $100 billion AI companies are decided by who controls GPUs *and* distribution, European companies are set up to lose. But I’m always optimistic, so I know that somewhere in Europe, there’s someone currently working on the next big thing. They just created the Git repository for this new project and it’s going to be magnificent ✨ Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### What is ARR anyway? URL: https://hypertext.fyi/what-is-arr-anyway/ Last updated: 2026-04-17T07:38:56.000Z *Hi everyone, it’s finally sunny again here in Paris. I hope you like the vibe of this newsletter. Feel free to share this week’s edition of Hypertext with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* The story that captured my attention this week was an accounting story. And no, it’s not as boring as it sounds. On Monday, OpenAI’s chief revenue officer Denise Dresser [wrote an internal memo](https://www.theverge.com/ai-artificial-intelligence/911118/openai-memo-cro-ai-competition-anthropic?ref=hypertext.fyi) about how OpenAI can increase its enterprise revenue going forward. Before I dive into the content of the memo, I want to point out two things. First, it’s always extremely funny when an “internal memo” gets published online on several websites just minutes after it was sent. I hate to break it to you, but yes, big tech companies often share these memos directly with reporters as long as they agree to say that it was an “internal leak” and not some form of official communication. It’s a powerful way to control the narrative without appearing like you’re talking to the public at large. Second, Denise Dresser really “buried the lede” with this one. In journalistic lingo, burying the lede is something you want to avoid. It means that you’re mentioning what’s actually important (the lede) near the bottom of the article. But in Denise’s case, it was a calculated move to look like it was just an afterthought. If I try to sum up this memo in one sentence, it would sound like this: “We have this brilliant strategy about enterprise AI and by the way, did I mention that Anthropic is lying about their revenue figures?” That’s the signal they wanted to send to shareholders, prospective investors and employees. “Their stated run rate is inflated. They use accounting treatment that makes revenue look bigger than it is, including grossing up rev share with Amazon and Google,” Dresser wrote. “Our analysis shows that this overstates their run rate by roughly $8 billion (at the current $30 stated).” So maybe Anthropic’s *real* ARR is $22 billion, while OpenAI’s ARR is [closer to $25 billion](https://www.reuters.com/technology/openai-tops-25-billion-annualized-revenue-last-month-information-reports-2026-03-05/?ref=hypertext.fyi). In other words: don’t panic, we’re still ahead of Anthropic. ARR, or annual recurring revenue, is a financial metric that predicts how much subscription revenue you’re supposed to generate over the next twelve months at the current revenue run rate. It is relevant for high-growth companies as trailing-twelve-month revenue matters less when a company is growing extremely quickly. But the thing is, nobody agrees about what it *really* means. Do you multiply the current month’s revenue by 12? Do you include one-time booked revenue, like setup fees or non-recurring deals? Some companies get *extremely creative* with ARR, including right here in Europe: before its CEO stepped down, [11x](https://www.11x.ai/?ref=hypertext.fyi), the AI agent platform for outbound sales calls, exceeded all metrics on this front. The company’s ARR [included](https://techcrunch.com/2025/03/24/a16z-and-benchmark-backed-11x-has-been-claiming-customers-it-doesnt-have/?ref=hypertext.fyi) free one-month trials. Multiplied by 12\. With a 70-80% churn rate on free trials. Of course, it’s not the case for everyone. A few months ago, Dust co-founder and CEO Gabriel Hubert told me that the company’s ARR was *real ARR* — 100% organic. While the non-inflated number wasn’t as big as other companies’ numbers, he thought it was the right, long-term strategy. But it leads me to this simple question: what is ARR anyway? ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/glyph-3.png) One of the most important AI companies in Europe is also one of the least visible. Over at Wired, my former TechCrunch colleague Maxwell Zeff wrote [a nice profile](https://www.wired.com/story/black-forest-labs-ai-image-generation/?ref=hypertext.fyi) on [Black Forest Labs](https://bfl.ai/?ref=hypertext.fyi). It matters because too many people still haven’t heard of Black Forest Labs. Mistral and ElevenLabs? Sure. Black Forest who? Shortly after contributing to the release of Stable Diffusion, the open-weights text-to-image AI model, Andreas Blattmann, Robin Rombach and Patrick Esser created their own AI lab focused on image generation, Black Forest Labs. It is now worth [over $3 billion](https://bfl.ai/blog/our-300m-series-b?ref=hypertext.fyi). The 70-person team based in Germany’s Black Forest is responsible for some of the best performing image generators on the market, only behind Google’s Nano Banana and OpenAI’s GPT-Image models. Their business model is quite simple. They license their technology to other companies that want to include image generation in their product without developing their own model. Microsoft, Meta, Canva and Adobe are all customers. That’s probably why you rarely hear about Black Forest Labs. And then, there’s xAI… With xAI, the timeline is a bit messy. At first, Elon Musk’s AI startup was a customer. Then xAI developed its own image generation model (the one that could [nudify people](https://www.reuters.com/legal/litigation/grok-says-safeguard-lapses-led-images-minors-minimal-clothing-x-2026-01-02/?ref=hypertext.fyi)). Then it approached Black Forest Labs again but the German startup turned the opportunity down because it was “too operationally difficult to partner with xAI,” Max wrote. Up next: Black Forest Labs wants to expand into physical AI (think: AI models for robots). It will compete with another well-funded European company in the space: [Genesis AI](https://genesis-ai.company/?ref=hypertext.fyi). ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/glyph-4.png) I hate to ask this question but… Are we in a bubble? Allbirds, the wool shoe brand, is [pivoting to AI](https://www.cnbc.com/2026/04/15/allbirds-bird-stock-shoes-ai.html?ref=hypertext.fyi), and more precisely building and operating a GPU farm with long-term lease arrangements. The company went public in 2021 (nice timing) and was valued at nearly 5 billion at the time. Last month, it sold all its assets and intellectual property [for $39 million](https://techcrunch.com/2026/03/30/allbirds-is-selling-for-39-million-it-raised-nearly-10-times-that-amount-in-its-ipo/?ref=hypertext.fyi) — oof. But the public company is still there. And they don’t have any product to sell. And it even secured a $50 million convertible financing facility. So what do you do with all that cash? In 2021, the company would have probably used that cash to buy Bitcoins. But because it’s 2026, Allbirds is turning its attention to the new gold, and that’s GPUs. If you remember when Allbirds was flying high, the company focused on environmental impact in its advertising, promising a low carbon footprint. Allbirds now plans to “remove references to the Company being operated for the environmental conservation public benefit.” If that pivot weren’t strange enough, Allbirds is now a memestock, up roughly 600% in a day. ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/glyph-5.png) Apple’s lawyers always tell EU regulators (and reporters) that the App Store is safe. According to them, if you open up app distribution on iOS, it’s going to be extremely dangerous for customers. It turns out a malicious developer managed to release a fake Ledger Live app on the App Store. The app is used in combination with Ledger’s hardware wallets to send and receive crypto assets. However, the fake app asked users to enter sensitive information (recovery phrases) to take control of those crypto wallets. [Nearly $10 million](https://www.coindesk.com/business/2026/04/14/a-fake-ledger-app-on-the-apple-app-store-just-drained-usd9-5-million-in-crypto?ref=hypertext.fyi) has been stolen from those wallets. The app remained live on the App Store for a week. Ledger is not at fault here. There was no security vulnerability on their side so they can safely blame Apple. Moreover, the company keeps saying that you should never share your recovery phrase with anyone. But when an app that looks like the official app asks you to enter the recovery phrase, it’s harder to know that you shouldn’t do it. I don’t blame people who got fooled… As for Apple’s claim that the App Store’s closed model guarantees security, this episode seriously undermines it. And that problem will only get worse as AI-built apps flood app stores. Have a good day ☀️ Romain ### Anthropic, the European unicorn ripper? URL: https://hypertext.fyi/anthropic-the-european-unicorn-ripper/ Last updated: 2026-04-10T08:01:46.000Z *Hi everyone, I hope you enjoy this week’s edition of Hypertext. Feel free to share it with a friend, it really helps 💌* [*Sign up here*](https://hypertext.fyi/)*.* ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph.png) The phrase of the week is clearly a variation of “the new Claude Mythos AI model is so good that it’s too dangerous to release to the public.” Once again, Anthropic dominated the headlines and captured all the air in the room with the announcement of a new model that is supposedly much smarter than existing models. It’s definitely concerning when it comes to cybersecurity as all the AI labs tend to catch up to the latest state-of-the-art models pretty quickly these days. How long until an open-weight model has the capabilities of [finding](https://red.anthropic.com/2026/mythos-preview/?ref=hypertext.fyi) “vulnerabilities \[that\] are often subtle or difficult to detect,” sometimes even “ten or twenty years old” just like Mythos? Could a bad actor use it to break down the financial system altogether? Fun times ahead… But let’s pivot to Anthropic’s *other* big launch this week: [Claude Managed Agents](https://claude.com/blog/claude-managed-agents?ref=hypertext.fyi). The easiest way to think about it is as a Platform-as-a-Service for AI agents. Just like Heroku, Fly.io, Render, Upsun, Clever Cloud, and Qovery abstracted cloud infrastructure, Anthropic abstracts the messy parts of running autonomous agents. You define your agent’s role and tools, hit deploy and that’s it. Anthropic manages sandboxing, credentials, monitoring and all the other tedious things required with long-running AI agents — no OpenClaw required. I can imagine shivers running down the spines of many startup founders who have been working on the ability to build and deploy AI agents at scale. LangChain, which recently [reached a $1.25 billion valuation](https://blog.langchain.com/series-b/?ref=hypertext.fyi), provides a platform (LangSmith) that does the same thing as Claude Managed Agents. Which raises the question: if Anthropic continues down this path, could it eventually compete with platforms like n8n, which allow visual AI agent building? Leave our European unicorns alone, please. ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-1.png) TikTok is [spending €1 billion](https://www.reuters.com/legal/litigation/tiktok-build-second-billion-euro-data-centre-finland-2026-04-08/?ref=hypertext.fyi) on a new data center in Finland. It turns out TikTok plans to spend as much as €12 billion on data centers across Europe, and I only learned about it this week. ByteDance, TikTok’s owner, likely wants to avoid repeating what happened in the U.S. Earlier this year, the U.S. Government forced the Chinese parent company to form a joint venture and sell 80% of TikTok to Oracle, Silver Lake, MGX and others, valuing the company at *just $14 billion* (it’s extremely low for an app with roughly 200 million users generating billions in ad revenue in the U.S. alone). ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-2.png) In other construction news, Revolut seems to have a thing for Paris. Per [Bloomberg](https://www.bloomberg.com/news/articles/2026-04-08/revolut-to-open-paris-office-as-lender-seeks-to-expand-in-europe?ref=hypertext.fyi), the fintech giant just signed a lease for an entire six-floor building on rue Réaumur. It plans to hire up to 400 employees. But the reality is that Revolut relies more and more on India to staff its team. The company [plans](https://www.reuters.com/business/finance/revolut-base-40-its-global-workforce-india-by-2026-2026-03-26/?ref=hypertext.fyi) to have 5,500 employees in the country by the end of the year. That’s 40% of its global workforce with people working on customer support, product development and finance. Revolut is often seen as the fintech company at the bleeding edge of product innovation. It’s interesting to see that it is still hiring aggressively across multiple markets even though many companies are slowing down hiring due to AI-driven productivity gains. This is further proof that Jack Dorsey used AI as an excuse to [lay off](https://x.com/jack/status/2027129697092731343?ref=hypertext.fyi) a large portion of Block’s workforce. The reality? Block’s performance has been underwhelming lately. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-3.png) The New York Times published a [riveting story](https://www.nytimes.com/2026/04/08/business/bitcoin-satoshi-nakamoto-identity-adam-back.html?ref=hypertext.fyi) on the creator of Bitcoin. It’s been 17 years that various reporters have tried to find out who is Satoshi Nakamoto, the pseudonymous figure who wrote the original Bitcoin research paper. And it turns out Satoshi could be British. The article points to Adam Back, a low-profile British cryptographer who contributed to the Cypherpunk mailing list in the 1990s and 2000s. It was a small circle of privacy-obsessed, sometimes libertarian-leaning technologists. And yet, despite the mounting evidence, Adam Back denies it. Worth noting: Satoshi is still believed to hold around 1 million bitcoins worth around €66 billion today. Not exactly an incentive to step into the spotlight, especially when you care so much about privacy. The reason why it’s always important to have a glance at the byline when you read a story: the article is co-written by John Carreyrou, the reporter who exposed Elizabeth Holmes and Theranos, and wrote *Bad Blood*. It’s a pretty good track record in investigative journalism. ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-4.png) This week, Greece announced [plans](https://www.reuters.com/business/media-telecom/greece-ban-social-media-under-15s-2027-pm-says-2026-04-08/?ref=hypertext.fyi) to ban social networks for teens under the age of 15\. France is moving in a similar direction, with Emmanuel Macron and former digital minister Clara Chappaz pushing for comparable measures. But what will it look like in reality? If you live in the UK and use an iPhone, you may have been surprised by iOS 26.4’s new rules. Apple [asked users](https://www.theguardian.com/technology/2026/mar/25/apple-iphone-users-face-over-18-age-check-to-use-services-after-update?ref=hypertext.fyi) to scan their passport or enter credit card numbers to confirm that they are 18+ to lift restrictions on adult content. These are just the first restrictions on online content and they are specifically targeted at children. However, it raises a broader question: what if a government wants to restrict access to essential services for certain groups, such as foreigners? Scan your ID or move along. ![](https://storage.ghost.io/c/7a/c5/7ac55220-8b80-4689-92ef-f10f1b9baf70/content/images/2026/04/-glyph-5.png) *Thank you for reading and feel free to reply to this email if you have comments.* Have a good day ☀️ Romain ### A tale of two narratives URL: https://hypertext.fyi/a-tale-of-two-narratives/ Last updated: 2026-04-10T07:22:09.000Z OpenAI just [acquired](https://openai.com/index/openai-acquires-tbpn/?ref=hypertext.fyi) TBPN, the daily streaming show focused on Silicon Valley (if you’re not familiar with TBPN, the New York Times wrote [a good profile](https://www.nytimes.com/2025/10/11/technology/tbpn-silicon-valley.html?ref=hypertext.fyi) a few months ago). This is insane news for both the tech industry and the media industry. When you think the story cannot become crazier, the Financial Times says they [spent](https://www.ft.com/content/4fe4972a-3d24-45be-b9fa-a429c432b08e?ref=hypertext.fyi) “low hundreds of millions” on this deal. It’s a tech level valuation for a media company. Insane stuff. But it made me think about the current narratives at big AI companies. A few years ago, an executive at Mistral told me that the French AI lab wanted to position itself as the anti-OpenAI: open instead of closed, colorful branding instead of black-and-white branding, developer-centric instead of consumer-centric. Mistral’s first breakthrough success was the [Mistral 7B](https://techcrunch.com/2023/09/27/mistral-ai-makes-its-first-large-language-model-free-for-everyone/?ref=hypertext.fyi): a small, open-weight model that you could literally download and run on your laptop. By comparison, OpenAI had just released GPT-4 and didn’t say much in its [technical report](https://cdn.openai.com/papers/gpt-4.pdf?utm%5Fsource=chatgpt.com) — we don’t even know how many parameters were needed to train the model. ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. Flash forward to 2026 and a funny thing is happening. Not only is each company’s narrative shifting, but in many cases they’re even *swapping* their narratives. Last week, Mistral co-founder and CEO Arthur Mensch wrote an opinion piece in [the Financial Times](https://www.ft.com/content/d63d6291-687f-4e05-8b23-4d545d78c64a?ref=hypertext.fyi) saying that big AI companies should pay a tax to content publishers in order to freely reuse content as training material. This is a complete 180 compared to Mistral’s positioning around the EU’s AI Act. At the time, they lobbied hard to protect big AI companies from extreme regulation. The funny thing is that the AI Act has reached a dead end. The application date has been [postponed](https://www.europarl.europa.eu/news/en/press-room/20260316IPR38219/meps-support-postponement-of-certain-rules-on-artificial-intelligence?ref=hypertext.fyi), and I wouldn’t be surprised if they postpone it again and again until it disappears completely. Now, Mistral gets to suggest what EU regulation should look like and this Financial Times piece is super smart: - It makes them look like a responsible, friendly company with the broader economy. - It’s a pitch on European sovereignty, their new (greenfield) strategy as it is becoming increasingly hard to compete on the global AI stage. - It’s a tax on revenue, and Mistral’s ARR ([$400 million](https://www.ft.com/content/664249e7-e8d5-4425-b397-ad3ed590b305?ref=hypertext.fyi)) is much smaller than OpenAI’s ([$25 billion](https://www.theinformation.com/articles/openai-tops-25-billion-annualized-revenue-anthropic-narrows-gap?ref=hypertext.fyi)) and Anthropic’s ([$19 billion](https://www.bloomberg.com/news/articles/2026-03-03/anthropic-nears-20-billion-revenue-run-rate-amid-pentagon-feud?ref=hypertext.fyi)). So OpenAI and Anthropic would pay 50x to 60x more than Mistral to fund European companies (Mistral’s potential customers). - And of course, it brings legal certainty and protection from liability. So let me recap. Mistral, the scrappy, developer-friendly European AI company, goes to the Financial Times to send a message to EU leaders through a prestigious old-media institution. At the same time, OpenAI is buying TBPN, an echo chamber for Silicon Valley, to shape the conversation more directly. In 2023, Sam Altman went on a world tour and met Emmanuel Macron, Rishi Sunak, Narendra Modi, Pedro Sánchez and other world leaders , acting almost like a head of state himself. In 2026, Arthur Mensch is starting to play a similar game. The tables have turned. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup. ### Hello world, again URL: https://hypertext.fyi/hello-world-again/ Last updated: 2026-04-10T07:22:22.000Z Welcome to [Hypertext](https://hypertext.fyi/), a brand new site and newsletter written by me, [Romain Dillet](https://hypertext.fyi/about). While this is a new home for my thoughts, writing isn’t a new thing for me. Over the past decade, I have written thousands of articles for a little known tech news publication called [TechCrunch](https://techcrunch.com/?ref=hypertext.fyi). Tech news has changed so much that people working in the tech industry aren’t looking for the same thing as they were 10 years ago. We are constantly being bombarded with small and big news items — in our feeds, inboxes, and private chats. But when is the last time you took some time to think about what you just read and tried to make sense of it all? With Hypertext, I want to bring analysis and opinion on the tech industry, for people working in the tech industry. Think about it as a smart briefing for busy people. I won’t try to cover everything. Instead, I plan to focus on some important tidbits of information and answer a simple question, over and over again: why does it matter? I’m not the first person to launch a newsletter. And people like [Ben Thompson](https://stratechery.com/?ref=hypertext.fyi), [Om Malik](https://om.co/?ref=hypertext.fyi), [M.G. Siegler](https://spyglass.org/?ref=hypertext.fyi), [Eric Newcomer](https://www.newcomer.co/?ref=hypertext.fyi) and [Alex Konrad](https://www.upstartsmedia.com/?ref=hypertext.fyi) have shown me that it was possible to break out of the traditional newsroom setup. And yet, I don’t think there’s enough coverage of what’s happening in Europe — especially in English, and especially with the right context. [Sifted](https://sifted.eu/?ref=hypertext.fyi), [The French Tech Journal](https://www.frenchtechjournal.com/?ref=hypertext.fyi) and business news publications are doing a fine job. But I believe I can cover the news through a different angle and by bringing new context to the table. More importantly, when I browse my LinkedIn newsfeed, some hot takes are simply extremely bad. I’m sure I don’t have a perfect track record over my 3,500 articles at TechCrunch. But I hope I can do better than the average ChatGPT-fuelled misinformed BS that you can read in your newsfeeds. So there you have it. Thank you so much for reading this post. Obviously, I have many ideas of what I could do next but one thing at a time… For now, please [subscribe to the newsletter](https://hypertext.fyi/#/portal/signup) before closing this tab. Have a good day ☀️ Romain ****Hypertext** is the smart briefing on the future of European tech. Sign up below. Subscribe Email sent! Check your inbox to complete your signup.