OpenAI Killed Sora in Six Months. What That Actually Tells You About the AI Race.
A $1 billion Disney deal. A product that burned $15 million a day. And a company that pulled the plug without warning.
Six months ago, OpenAI launched Sora to the fanfare that makes you feel you’re witnessing history. Imagine a TikTok-like feed where all the content is produced by artificial intelligence. You could drop yourself into cinematic scenes, make your friends dance the polka in exotic locations, or pour a bucket of paint over your own head and watch the physics render perfectly across your hair. I tried a few of those videos back in October. I’ll admit: I was genuinely impressed. It felt like a glimpse of something that shouldn’t exist yet.
Then last week, OpenAI announced it was shutting Sora down entirely. The app goes dark on April 26, 2026. The API follows on September 24. That’s it. Six months. Done.
The Cost of Running a Magic Trick Nobody Needed
The number that makes this story genuinely surreal is buried in reporting from Forbes: Sora was burning approximately $15 million every single day in inference costs. Not in total development. Not in marketing. Servers needed to be kept active so users could produce videos. Every 10-second clip reportedly costs OpenAI roughly €1.30 to produce. And it was free for everyone to use.
What was the total revenue the app generated across its entire six-month life?
According to app analytics data relayed by TechCrunch, $2.1 million. The app was spending what amounts to $450 million a month and making $2.1 million back over half a year. This wasn’t a struggling product. This was a financial black hole with a beautiful interface.
In November 2025, downloads briefly hit a peak of 3.3 million before plummeting to 1.1 million by February. That trajectory — explosive curiosity followed by near-instant abandonment — tells you everything you need to know about the gap between “impressive demo” and “product people actually keep using.”
Disney Found Out With Less Than an Hour’s Notice
The part of this story that still stuns me is what happened to Disney.
Back in December 2025, OpenAI and Walt Disney Company signed a three-year licensing agreement considered genuinely groundbreaking. It allowed Sora users to generate videos featuring characters from Disney, Marvel, Pixar, and Star Wars, covering more than 200 licensed characters. Disney, famously protective of its intellectual property, had effectively handed OpenAI a primary key to its entire creative universe. In return, Disney committed to a $1 billion stake in OpenAI, structured entirely in stock warrants rather than cash.
It was the deal you spent months negotiating, years imagining.
According to reporting from Bloomberg, Disney found out Sora was being shut down less than an hour before the public announcement. A Disney spokesperson confirmed the deal was dead with a statement so composed it bordered on passive-aggressive: “We respect OpenAI’s decision to exit the video generation business.”
That sentence carries more weight the more you read it.
Where the Real Strategy Lies
OpenAI’s official statement on the shutdown focused not on the losses but on what comes next. The Sora research team, they said, will now be redirected toward world simulation research to advance robotics.
That phrase deserves more attention than it’s getting.
World simulation, or world modeling, is distinct from video generation. A world model doesn’t just produce something that looks realistic. It understands physics. It knows that if you place a cup on the edge of a table and nudge it slightly, the cup will fall. It can predict what happens when objects interact, when forces are applied, and when materials deform.
It’s this fundamental ability that enables robots to be practical in actual situations, not just visually appealing in controlled presentations.
When OpenAI first published Sora in 2024, its own researchers specifically noted that it appeared to develop an internal model of physical reality as a byproduct of learning to generate coherent video. They described it as a model that “truly understands the physics of the world.” That wasn’t marketing language. That was a signal about where the real value sat — not in the social feed, but in the spatial reasoning engine underneath it.
The move to pivot Sora’s team toward world simulation could represent a much smarter long-term bet than the app itself ever was. Whether that points toward robotics hardware, physical AI agents, or something none of us has quite imagined yet remains genuinely open. But it’s a better use of the technology than generating dance videos.
The Competition That Forced OpenAI’s Hand
There is another reason Sora had to go, and it’s one the official statements won’t say plainly.
Sora was outdone by Chinese AI companies in terms of video quality, and they accomplished this more affordably. Google’s Veo 3 had already established a higher quality ceiling for generated video. According to TechCrunch’s investigation, while OpenAI’s Sora team was working on the social app, Anthropic was quietly winning over the software engineers and enterprises that actually drive revenue. Claude Code, in particular, was eating into OpenAI’s technical audience at a speed that demanded an internal reallocation of compute resources.
The math became brutal: continue spending $450 million a month on a video product that can’t catch its competitors, or free up those GPU resources and point them at the market that’s actually converting.
They chose the latter.
ChatGPT vs. Claude: The Numbers People Keep Getting Wrong
This is the eventual outcome of the Sora story, and it’s crucial to be exact with the data because the prevailing accounts are inaccurate in both directions.
ChatGPT is enormous in a way that is genuinely hard to conceptualize. OpenAI reported 900 million weekly active users as of late February 2026, with approximately 5.35 billion monthly site visits. Along with YouTube and Google, ChatGPT is one of the top five most visited websites worldwide. The brand recognition is so embedded in public consciousness that “ChatGPT” has become a generic verb for AI use, the way “Google” became a verb for search two decades ago. That kind of cultural penetration doesn’t evaporate because a competitor has better benchmark scores.
Claude, by comparison, has roughly 18.9 million monthly active users. OpenAI has approximately 60 times more users. That gap is real, and it matters.
But here’s where the story gets genuinely interesting. Claude’s daily downloads briefly overtook ChatGPT’s in the US App Store in early March 2026. Not in total users. Not in monthly visits. In daily downloads, the metric that measures where new intent is going. Anthropic reported that Claude’s paid subscriptions more than doubled this year. Enterprise penetration tells an even sharper story: eight of the Fortune 10 are now Claude customers, and over 500 companies spend more than $1 million annually on Claude. Anthropic hit a $14 billion annualized revenue run rate in February 2026.
The dynamic playing out is not “Claude is winning.” The dynamic is:
ChatGPT owns the mass market, and Claude is becoming the tool that the people who influence everyone else actually trust. These are developers, senior engineers, technical writers, researchers, and founders. When those populations shift, the broader market follows slowly but reliably.
At the end of this contest, the most favorable result is for AI tool users. OpenAI is forced to improve continuously. Anthropic is forced to scale beyond its technical niche. Google is forced to justify its infrastructure investments with actual product quality. None of them can afford to coast. The only people who genuinely win in a market this competitive are the ones using the products.
What killing Sora actually means
Closing a product after six months while absorbing $450 million in losses and terminating a $1 billion Disney deal is not conventional behavior. By most measures, it looks like failure — and it will be reported as failure.
I’d argue it’s actually closer to rational triage under extreme pressure.
The Silicon Valley move-fast-and-kill-it culture gets caricatured endlessly, but the logic underneath it is sound: test fast, measure honestly, stop spending on things that don’t convert. OpenAI launched Sora at scale, watched the usage curves, watched the competitor quality curves, and made a clean decision. That takes a kind of institutional courage, especially when a $1 billion Disney deal is sitting in the balance, and your team has been working on the product for over a year.
The real question is what they do with the compute they’ve freed up, and whether the world simulation bet pays off. If those researchers build something that gives robots functional physical reasoning, the Sora detour will look like a very cheap experiment in retrospect.
If they don’t, it’ll just be another line item in what is shaping up to be the most expensive product pivot in technology history.
What’s your read on this? Is killing Sora a smart reallocation or a sign that OpenAI can’t focus long enough to win anything? Curious where you land on this in the comments.