The power to pull the plug
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 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.
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 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 called “Le Chaton Fat” that would rival Fable.
It’s a good joke because it says something true about Mistral. They have fallen behind 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, 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: “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, 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?
Keep your critical business content centralised and compliant. Box provides enterprises with a secure, flexible content layer to connect teams, streamline workflows, and safely leverage AI using your own data.
The biggest PE loss since 2008 is a SaaS company
Have you ever heard about Medallia? 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 of $2.5 billion.
In 2021, right in the middle of the Covid-era software valuation frenzy, private equity firm Thoma Bravo acquired 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 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, 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