The Real Numbers Behind OpenAI Killing Sora — $15M/Day Cost, $2.1M Revenue
OpenAI's decision to shut down Sora makes a lot more sense when you look at the numbers. According to analysis by Appfigures and independent researchers, the AI video generation platform was costing an estimated $15 million per day in compute — while generating roughly $2.1 million in total lifetime consumer revenue. The math was never going to work. Sora accumulated only around 450,000 downloads in six months, a rounding error compared to ChatGPT's tens of millions, and the product was losing ground to faster-moving competitors like Runway and Kling despite the enormous ongoing infrastructure cost.
The collateral damage from the shutdown is significant. A Disney partnership worth up to $1 billion was cancelled before it could be finalized, a deal that had been presented as a flagship example of responsible enterprise AI deployment. OpenAI CEO of AGI deployment Fidji Simo captured the company's new posture bluntly: "We cannot miss this moment because we are distracted by side quests." The compute previously allocated to Sora is now being redirected toward agentic AI workloads and ChatGPT enterprise products — areas where OpenAI is generating meaningful revenue and defending territory against Google and Anthropic.
The Sora shutdown is a useful data point in the broader debate about which AI verticals can actually sustain themselves economically. Video generation turned out to be an enormously expensive capability to run at scale, with a consumer demand curve that didn't come close to justifying the infrastructure. As OpenAI accelerates toward a 2026 IPO, the pressure to demonstrate margin discipline — not just technical ambition — is clearly shaping product decisions in ways that would have seemed uncharacteristic just eighteen months ago.