The Sora app died quietly on a Tuesday. No dramatic farewell post, no "we're evolving" corporate rebrand. Just a shutdown notice and a math problem that should unsettle every AI investor who hasn't been paying attention: OpenAI spent an estimated $15 million per day running Sora. It earned $2.1 million in total lifetime revenue. Not per day. Total. Ever.

That's the story of one of AI's most hyped consumer launches — and its swift, expensive collapse.

OpenAI announced the closure of Sora on March 24, 2026. Users have until April 26 to download their content before the app goes dark. The API follows on September 24. Sora had launched to enormous fanfare in late 2024, promising a new era of AI-generated video. Within months, it became clear that the era had a serious math problem.

Every 10-second clip Sora generated cost OpenAI roughly $130 in compute. Multiply that across millions of curious users experimenting with the technology, and the losses compound fast. Downloads peaked in November 2025, then collapsed — down 66% by the time the shutdown was announced. Even a planned partnership with Disney couldn't survive the economics. The deal would have brought $1 billion in investment from Disney and licensed over 200 of its characters for use inside Sora. No money ever changed hands. The deal simply died alongside the product.

The strategic pivot is robotics. OpenAI says Sora's research team will continue working on "world simulation" — the same physics modeling that made video generation work will now be used to train robots to understand and navigate physical environments. From a pure technology standpoint, that's a sensible move. World models are valuable for robotics. The problem is what this reframing admits: consumer AI video, at current compute costs, is not a business.

Here's the uncomfortable truth no one wants to say out loud: OpenAI didn't kill Sora because the technology failed. It killed Sora because the technology worked exactly as designed, cost exactly what cutting-edge video inference costs, and there is no consumer price point people will actually pay that covers those expenses. $130 per 10 seconds of video is not a rounding error. It's the product.

My Opinion

I'll be direct: the Sora collapse is the most honest data point we've had in months about where consumer AI is actually headed. OpenAI built something genuinely impressive — coherent, cinematic video from text prompts — and then discovered that impressive doesn't mean profitable at scale. The market wasn't broken. The economics were.

What bugs me is that this pattern keeps repeating. Companies build extraordinary AI capabilities, burn through billions in compute and investor capital, get viral adoption from people who love free or cheap access to novelty, and then discover that novelty wears off in about 12 weeks. Sora was the hottest app in November 2025. By March 2026, it was financially unjustifiable. That 66% download drop isn't a fluke — it's a signal about how durable AI consumer products actually are when the magic fades and the bill comes due.

The robotics pivot is where I think OpenAI is genuinely right. Physical AI — robots that can reason about and navigate the real world — has an obvious value proposition that doesn't depend on consumer novelty cycles. Industrial, medical, and logistics applications will pay real money for real capability. But the lesson from Sora isn't just "pivot to robotics." It's that the AI industry needs to stop pretending that any sufficiently impressive demo can be turned into a consumer subscription business. The next company that builds something people need enough to pay what it actually costs — not something they like enough to try once for free — will be the one that matters.

Author: Yahor Kamarou (Mark) / www.humai.blog / 29 Mar 2026