When OpenAI launched Sora to the public in late 2025, it felt like the future had arrived. Type a prompt, get a cinematic video. Disney signed a $1 billion partnership to let users create videos with Marvel and Star Wars characters. The hype machine was running full throttle.

Six months later, Sora is dead. And its demise tells us more about the real economics of AI than any earnings call ever could.

The Numbers That Killed It

The math was brutal. According to the Wall Street Journal, Sora burned through roughly $1 million per day in compute. Some estimates from Cybernews put the figure at $15 million daily when factoring full inference costs.

Against that burn rate? Total lifetime in-app revenue of $1.4 million, per Sensor Tower data shared with the BBC. Sora’s entire six-month haul roughly equaled one day of operating costs.

Meanwhile, ChatGPT pulled in $1.9 billion over the same period. It doesn’t take an MBA to see which product deserved the GPU cycles.

User numbers told the same story. Downloads cratered from 3.3 million in November 2025 to 1.1 million by February 2026. Active users collapsed from roughly one million to under 500,000. The novelty wore off. The server bills didn’t.

Why Video AI Is So Absurdly Expensive

Text generation is cheap by comparison. Every Sora request triggered complex neural network processes across multiple stages — understanding the prompt, generating individual frames, ensuring temporal consistency, rendering at high resolution. Each video consumed GPU cycles that could have powered thousands of ChatGPT conversations.

This isn’t unique to OpenAI. Every company in AI video faces the same fundamental challenge: generating even a few seconds of coherent video requires orders of magnitude more compute than generating pages of text. The technology works — it just doesn’t work economically.

Forrester analyst Thomas Husson called it plainly: Sora was “a resource black hole” with “limited monetisation.”

Disney Got 60 Minutes of Warning

Perhaps the most jaw-dropping detail: Disney found out Sora was being killed less than one hour before the public announcement.

A billion-dollar partner. Less than sixty minutes of warning. According to the Wall Street Journal, Disney’s tech team learned about the “strategy pivot” on Monday night with zero opportunity for negotiation or transition planning.

Disney’s response was pure corporate fury wrapped in diplomatic language — they “respect OpenAI’s decision” and “will continue to engage with AI platforms.” Translation: we’ll never trust you again.

This incident will send ripples through every boardroom considering AI partnerships. If OpenAI can nuke a billion-dollar deal with less than an hour’s notice, every executive has to recalculate the risk of building on top of AI platforms.

Anthropic Was Eating Their Lunch

Here’s the competitive context that made Sora’s death inevitable. While OpenAI burned compute on video generation, Anthropic’s Claude Code was quietly winning the war that actually matters: enterprise revenue.

Software developers and enterprise clients are the most valuable customer segment in AI. They pay real money, they stick around, and they build workflows that create lock-in. Claude Code was capturing this market at an alarming rate.

OpenAI’s own Codex has since caught up — surpassing $1 billion in annualized revenue by January 2026. But every month spent bleeding compute into Sora was a month where Anthropic gained ground in the race that actually pays the bills.

The lesson is clear: in the current AI landscape, specialized tools for professionals crush consumer entertainment plays.

The IPO Pivot

Sora’s death isn’t about one failed product — it’s about OpenAI growing up. CEO Sam Altman historically ran the company like Y Combinator, placing broad bets across video, browsers, hardware, robots, and coding agents. That scatter-shot approach worked for a research startup. It doesn’t work when you’re preparing for an IPO.

CFO Sarah Friar told CNBC the company needs to be “ready to be a public company.” Translation: cut the money pits and tell a coherent growth story to Wall Street.

The new strategy centers on a “super app” combining ChatGPT, Codex, and Atlas (OpenAI’s browser) into a unified consumer interface. The Sora research team is being redirected toward “world simulation research to advance robotics” — a domain with a much clearer path to enterprise value than consumer video.

What This Means for AI

Sora’s failure doesn’t mean AI video generation is dead. It means it’s not ready for primetime as a standalone business. The technology needs several more years of hardware improvements and efficiency gains before it can operate at consumer scale without hemorrhaging money.

For the broader AI industry, Sora highlights the tension every company is navigating: the gap between what’s technically impressive and what’s commercially viable. Generating photorealistic video from text is genuinely remarkable. It’s also genuinely unprofitable at current compute costs.

The Sora shutdown also raises uncomfortable questions about trust. If the hottest AI company on the planet can kill a billion-dollar deal with less than an hour’s notice, every company building on AI platforms needs to build in more redundancy, negotiate better terms, and never assume any AI product is permanent.

The Bottom Line

Sora was a beautiful demo that became a terrible business. It cost too much to run, attracted too few paying users, and consumed resources OpenAI desperately needed elsewhere. Sam Altman made the ruthless but rational call to kill it.

The real story isn’t about video generation — it’s about the brutal economics of AI in 2026. Compute is finite. Revenue matters. And even the most jaw-dropping technology has to answer the question every business does: who’s paying for this?

For Sora, nobody was. And that’s why it’s gone.