OpenAI Killed Sora: $15M a Day Will Do That

Fifteen million dollars a day. That's what it cost OpenAI to keep Sora running. Total lifetime revenue from the product? $2.1 million. Not per day. Not per month. Total.

On March 24, OpenAI announced it's pulling the plug. The Sora app shuts down April 26. The API follows in September. Disney's billion dollar partnership built around the tool? Being wound down entirely.

This is the most expensive failed product in AI history. And it tells you everything about where the industry actually stands.

The math was never going to work

Video generation is a GPU furnace. Every second of output requires orders of magnitude more compute than a text response. OpenAI's own models show that a single Sora request consumed resources equivalent to hundreds of ChatGPT queries.

Fidji Simo, OpenAI's applications CEO, told employees the company couldn't afford "side quests" anymore. Every GPU cycle spent rendering video was a cycle not spent on ChatGPT, Codex, or enterprise API calls. Products that actually make money.

The unit economics paint a brutal picture. At $15 million per day in inference costs, OpenAI would need roughly $5.5 billion in annual Sora revenue just to break even on compute. They generated $2.1 million across the product's entire lifespan.

Cheap for users, ruinous for providers

Here's the paradox. AI video generation is transformative for buyers. It costs 90 to 99 percent less than traditional video production. A thirty second commercial that runs $50,000 with a production crew costs under $100 with AI tools. 72 percent of marketers have already adopted the technology.

But someone has to pay for the GPUs. Google charges roughly $30 per minute for Veo 3. Budget tools run $0.50 to $2 per minute. Those prices don't reflect the true cost of inference. They reflect what the market will tolerate while companies subsidize growth with venture capital.

The gap between what users pay and what providers spend is the central tension of the AI video market. Sora proved that even the best funded company in the space couldn't bridge it.

What the Disney deal collapse tells you

In December 2025, Disney announced a three year, billion dollar investment centered on Sora. It was the kind of partnership that validated the entire category. Hollywood's biggest studio betting real money on AI video generation.

Four months later, that deal is unwinding. Not because the technology failed creatively. Because the economics failed structurally. Disney will find other tools. But the signal matters. When the flagship partnership of a flagship product dissolves this fast, it exposes how fragile the business model was from day one.

Is this a crack in the AI bubble?

VideoCardz asked the question directly. The answer is more nuanced than a yes or no.

Text generation has proven its economics. ChatGPT, Claude, and Gemini all demonstrate that language models can run profitably at scale. The compute per query is manageable. The willingness to pay is established.

Video is a different beast. The compute requirements are exponentially higher. The willingness to pay hasn't caught up. And unlike text, where inference costs keep dropping, video generation hasn't found its efficiency breakthrough yet.

Sora's shutdown doesn't mean AI video is dead. It means AI video at current cost structures is unsustainable without massive subsidies. That's a solvable problem. But solving it will take years, not months.

What this means for your roadmap

If you're building products or workflows around AI video, three things to consider.

Treat every AI video provider as a temporary dependency. Sora's two stage shutdown gives developers six months to migrate. That's generous by startup standards. The next tool that disappears might give you thirty days. Build abstraction layers. Don't couple your pipeline to a single provider.

Watch the compute economics, not the demos. Every AI lab can produce stunning video demos. The question is whether they can do it at a price that sustains a business. Before you commit to a platform, look at their pricing trajectory. If costs are going up or staying flat, the service is being subsidized. Subsidies end.

Focus on use cases where the ROI is already clear. Short form marketing content, product demos, social media clips. These are the segments where AI video pricing works today. Feature length content and real time generation are still waiting for their cost curve moment.

The AI video market will survive Sora's death. But it will look different on the other side. Leaner providers with better unit economics will replace the brute force approach. The companies that build for that reality now won't have to scramble when the next shutdown notice arrives.