OpenAI Kills Sora: The Math Was Never There

Six months. That is how long it took for OpenAI to build, hype, scale, and ultimately abandon Sora — its TikTok-style AI video generation app. On March 24, 2026, the company announced it was shutting down both the app and its API, dissolving a landmark deal with Disney in the process, and signaling a strategic retreat from consumer-facing creative tools that, frankly, anyone paying attention should have seen coming.

I am not here to eulogize Sora. I am here to explain why it was always going to end this way.

The Rise: Impressive Numbers, Hollow Foundation

When Sora launched as a standalone app in late September 2025, it was, by conventional metrics, a success. It hit number one on Apple's App Store almost immediately, crossed one million downloads within five days, and peaked at approximately 3.3 million downloads across iOS and Google Play in November 2025. OpenAI shutters short-form video app Sora as company reels in costs

But download counts are not a business. The app generated roughly $2.1 million in total in-app purchase revenue over its entire lifetime — a figure so modest it barely registers as a rounding error for a company burning through billions annually and seeking to justify its $730 billion valuation as it prepares for an IPO. ¹ Meanwhile, ChatGPT — the company's flagship product — serves 900 million weekly active users, making the contrast in scale almost comically stark. ²

By February 2026, monthly downloads had collapsed to 1.1 million. The trajectory was unmistakable. The app had no retention engine, no monetization moat, and no compelling reason for users to return day after day. What it did have — in abundance — was problems.

The Chaos: Deepfakes, Copyright, and Creative Liability

Let us be precise about what Sora actually was in practice. Its flagship feature, originally called "cameos," allowed users to scan their faces and insert themselves — or anyone else — into AI-generated videos. The predictable happened immediately.

Deepfakes of Sam Altman, Martin Luther King Jr., and Robin Williams proliferated on the platform. According to TechCrunch, both King's and Williams' daughters went on Instagram to publicly ask users to stop making videos of their deceased fathers. ² Copyright violations were rampant: Mario, Naruto, and Pikachu appeared in AI-generated videos before any licensing agreements were in place. The legal entity Cameo took OpenAI to court over the "cameos" feature name; according to TechCrunch, Cameo prevailed, forcing a rename to "characters." ² (It is worth noting that the underlying facts of that litigation — whether it concluded with a formal judgment or a settlement — have not been fully disclosed publicly. What is documented is the outcome: the feature was renamed.)

In December 2025, Disney did something remarkable: rather than sue, it announced a planned $1 billion equity stake and a three-year licensing agreement covering more than 200 characters from Disney, Marvel, Pixar, and Star Wars — a deal that would have integrated Sora-generated content directly into Disney+. OpenAI Will Shut Down Sora Video App; Disney Drops Plans for $1 Billion Investment

It looked, briefly, like a lifeline. It turned out to be irrelevant.

The Economics: Compute Is the Whole Story

Here is the part that matters. Here is the part that was always going to matter.

Video generation is among the most computationally expensive tasks in the AI stack. Every second of generated video requires orders of magnitude more processing power than a text response or even an image. And OpenAI, for all its capital commitments, does not have enough compute to do everything it wants to do.

Bill Peebles, head of Sora, said the quiet part loud in October 2025: "We have been quite amazed by how much our power users want to use Sora, and the economics are currently completely unsustainable." OpenAI's Sora decision is all about AI compute

Consider what that means in practice — and let me make the arithmetic explicit, because the numbers are instructive even as rough estimates. Sora generated $2.1 million in lifetime in-app purchase revenue. OpenAI is losing billions annually. Video generation is, by any technical measure, among the most compute-intensive tasks in the AI stack — a single minute of high-fidelity generated video requires vastly more GPU time than a text exchange or even an image render. Now consider the usage side: at its peak, Sora had 3.3 million monthly downloads. If even 10% of those users generated just five minutes of video per month, that is 1.65 million minutes of video generation per month at peak. At even a conservative estimated compute cost of $0.10 per generated minute — a figure that industry practitioners suggest is, if anything, generous — that is $165,000 in monthly compute costs from a fraction of the user base, against a lifetime revenue figure of $2.1 million. The math does not work. It was never going to work. This is a back-of-the-envelope illustration, not a precise audit, but the directional conclusion is not in doubt: the revenue was not a rounding error relative to costs. It was not even in the same conversation.

Meanwhile, demand for AI compute across the industry has surged dramatically. Data from OpenRouter's model platform — which tracks inference request volume across major AI systems and serves as a useful, if partial, proxy for industry-wide demand trends — shows that AI usage more than tripled in just 2.5 months, a sign of explosive and accelerating demand. ³ Supply has not kept pace. Building new data centers has become harder due to local opposition, energy constraints, and shortages of key components such as memory chips. OpenAI has committed hundreds of billions of dollars to data center and chip deals and still lacks sufficient capacity. ³

Every compute dollar spent rendering someone's deepfake of Sam Altman walking through a slaughterhouse is a compute dollar not spent on ChatGPT — which serves 900 million weekly active users and is the company's clearest path to revenue. ²

The math was never there. It was never going to be there.

The Pivot: Focus as a Business Strategy

Earlier this month, Fidji Simo — OpenAI's CEO of applications — held an all-hands meeting with staff. The message was unambiguous. "We cannot miss this moment because we are distracted by side quests," she said. The "everything era," she declared, is over. ³

Today's announcement is the operational execution of that strategy. Alongside Sora's shutdown, OpenAI announced it is also discontinuing its Instant Checkout shopping feature and consolidating its web browser, ChatGPT app, and Codex coding app into a unified desktop "super app." The direction is clear: enterprise, productivity, and AI agents — not consumer social platforms built around AI-generated video content. OpenAI shutters short-form video app Sora as company reels in costs

The competitive pressure makes this pivot more urgent, not less. Anthropic's Claude has captured significant market share among developers and enterprise clients. Google's Veo video generation tools have continued to advance, leaving Google as essentially the only player in AI video generation with true scale. ByteDance's SeeDance 2.0 has drawn significant attention with Hollywood-quality generated scenes. OpenAI is also preparing to release a new model internally referred to as "Spud." OpenAI to shut down Sora app months after launch as focus shifts to agents

OpenAI cannot fight every battle simultaneously. Sora was a battle it chose to stop fighting.

The Fallout: Disney, Developers, and What Comes Next

The Disney deal is dead. To be precise about the deal's status: the $1 billion investment was announced in December 2025, but according to CNBC, the transaction never closed — no money had changed hands before the collapse. ¹ What dissolved was not a signed and fully executed partnership but an announced deal that never reached financial close. Whether a formal term sheet or conditional agreement existed has not been publicly disclosed. Disney issued a characteristically diplomatic statement: "We respect OpenAI's decision to exit the video generation business and to shift its priorities elsewhere." OpenAI is shutting down its Sora video app just months after launch

The Sora API is also being discontinued, cutting off developer access entirely. OpenAI has committed to sharing timelines and providing guidance on preserving user-created content — small consolation for developers who built pipelines around a model that is now being retired.

The underlying Sora 2 technology will persist, redirected toward world simulation research in support of robotics applications. OpenAI confirmed this directly in its shutdown statement: "As we focus and compute demand grows, the Sora research team continues to focus on world simulation research to advance robotics that will help people solve real-world, physical tasks." ⁴ This is, in its own way, a more coherent application of the technology. Generating photorealistic video of physical environments to train robotic systems is a legitimate and potentially lucrative use case. Generating videos of Pikachu doing ASMR is not.

Conclusion

There is a version of this story that frames Sora's shutdown as a failure. I do not find that framing particularly useful or accurate. Sora was an experiment in whether AI video generation could anchor a consumer social platform. The answer was no — not primarily because the technology was insufficient, but because the economics were structurally broken, the moderation challenges were severe, and the opportunity cost of the compute was too high. One could argue that with better moderation infrastructure, a different monetization model, or a more permissive regulatory environment, the outcome might have differed. Perhaps. But those are not the conditions that existed, and OpenAI is not a company that can afford to wait for better conditions.

What is worth noting is the speed and abruptness of the decision. On Monday, March 23, 2026 — one day before the shutdown announcement — OpenAI published a blog post titled "Creating with Sora safely," outlining new safety improvements for the app. The Guardian reported that OpenAI gave no indication it was winding Sora down as recently as that Monday post. OpenAI shutters AI video generator Sora in abrupt announcement That is not a graceful exit. That is a company moving fast and accepting the reputational friction that comes with it.

OpenAI is preparing for an IPO. It is losing billions annually. It has a consumer AI product in ChatGPT that represents a clear and scalable revenue path. In that context, shutting down a deepfake-adjacent social video app that made $2.1 million in its entire lifetime is not a difficult decision.

It is, in fact, the only decision that makes sense. ¹