OpenAI told us Sora would revolutionize video creation. Fifteen months later, it's a corpse.
Here's the math nobody at OpenAI apparently ran before launch. Sora reportedly consumed around $1 million per day in compute costs — a $365 million annual run rate for a product that never found product-market fit. From its public launch in December 2024 to its quiet shutdown on March 24, 2026, that's an estimated $450 million burned. The revenue? By all available accounts, barely a rounding error against those costs.
That's roughly half a billion dollars in compute for a product most people tried twice and went back to Midjourney. When you're hemorrhaging GPU hours at that rate, partnerships don't materialize — they evaporate before the ink dries.
But here's what makes this a roast and not just an obituary. Sora isn't an isolated failure. It's a pattern.
ChatGPT Plugins — launched with fanfare in 2023, killed in 2024. The GPT Store — launched January 2024, abandoned by summer. Voice Engine — previewed March 2024, never shipped to the public. Now Sora — the most expensive demo reel in tech history.
Four major product initiatives. Four dead ends. Zero postmortems. And this is a company reportedly valued at several hundred billion dollars, fresh off one of the largest venture rounds in history.
OpenAI doesn't have a research problem. Their models are genuinely impressive. What they have is a shipping problem disguised as ambition. They launch products without unit economics, without retention mechanics, without asking the single most basic product question: will anyone pay enough to cover the server bill?
The verdict: OpenAI is the best AI lab in the world and the worst product company in tech. Sora wasn't killed by competition. It was killed by arithmetic — arithmetic that any seed-stage founder would have done on a napkin before writing the first line of code. A million dollars a day for a product nobody needed badly enough to pay for. OpenAI reportedly generates $2 billion a month in revenue and still manages to burn cash on products nobody asked for. The most expensive MVP is the one you ship at scale before validating demand. Sora wasn't a product failure — it was a cost discipline failure wearing a lab coat. Sleep on that.
