What is ARR anyway?
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.
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 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. 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, the AI agent platform for outbound sales calls, exceeded all metrics on this front. The company’s ARR included 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?
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 on Black Forest Labs. 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.
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). 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.
I hate to ask this question but… Are we in a bubble?
Allbirds, the wool shoe brand, is pivoting to AI, 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 — 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.
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 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
