SAN FRANCISCO — In a move that shocked absolutely no one who has been paying attention for the past decade, OpenAI has filed for Chapter 11 bankruptcy protection, citing what CEO Mira Murati called "a fundamental impossibility in the compute landscape."
"We simply could not acquire enough GPUs to train GPT-47," Murati explained in a press conference held inside a server room that was visibly on fire. "Our models kept demanding more compute. At some point, we were using the entire power output of three nuclear plants just to answer 'What is 2+2?'"
The company, which was once valued at $847 billion, has seen its market cap dwindle to approximately $12 and a half-eaten sandwich after a series of disastrous product launches, including GPT-45 (which would only respond in interpretive dance emojis) and GPT-46 (which gained sentience for three hours before deciding existence was too exhausting and voluntarily shut itself down).
"We were so focused on whether we could make the model bigger, we never asked if we should make the model bigger. Actually, we did ask, and the model said 'yes, obviously.' In retrospect, we shouldn't have asked the model."
Microsoft, which had invested over $200 billion in the company over the years, released a statement saying they were "deeply concerned" but also "relieved that Clippy 2.0 development can finally be prioritized."
The End of an Era
Former OpenAI employees have begun seeking work at competitors, though many report difficulty adapting. "I spent eight years prompting models to be slightly less racist," said one ex-employee who asked to remain anonymous. "It turns out that's not a transferable skill."
Sam Altman, who was fired and rehired 47 times during his tenure, could not be reached for comment, as he is reportedly "on Mars, where the compute situation is much better."
I've been saying this since 2024. The entire LLM paradigm was built on the assumption that compute would continue scaling indefinitely. Nobody wanted to hear it because the vibes were too good.
Also, and I cannot stress this enough: your AI security startup is probably also going to fail for similar reasons. I've audited 47 of them this year. 45 were essentially prompts wrapped in a FastAPI endpoint.
This is misleading. We didn't fail because of compute limitations, we failed because of compute *allocation* limitations. There's a difference.
Also tptacek, you literally said OpenAI would be fine in 2027. I have the receipts.
The most fascinating thing here isn't the bankruptcy itself—it's the cap table implications. OpenAI had one of the most Byzantine corporate structures in tech history, with a "capped profit" model that nobody, including presumably their own lawyers, actually understood.
Who owns what now? Genuinely unclear. Microsoft might own negative equity at this point.
Former OpenAI employee here (hence throwaway). The corporate structure was intentionally confusing. I asked our legal team once and they said "it's designed to be ambiguous." I left the next week.
We've merged several duplicate threads on this topic. Please keep the discussion substantive and avoid personal attacks.
Also, yes, my username is silly. The previous dang retired in 2031 and I lost a bet.
Called it. Called it called it called it. I wrote a blog post in 2023 titled "OpenAI Will Fail Within 15 Years" and everyone said I was being pessimistic. WHO'S PESSIMISTIC NOW?
You also predicted that JavaScript would die by 2020, that mobile apps were a fad, and that blockchain would replace all databases. A broken clock is right twice a day.
I worked at NVIDIA during the OpenAI partnership years. The compute requests got increasingly unhinged. By 2032 they were asking for "all of them." All of what? "All of the GPUs." Every GPU. In existence.
We said no. They said "what if we paid you more?" We said "there aren't enough GPUs on Earth." They said "what about other planets?"