OpenAI Killed Sora and Bet Everything on a Model Called 'Spud'

Thirty minutes. That's how much warning Disney got before OpenAI killed the product they'd agreed to invest a billion dollars in. On the evening of March 23, Disney and OpenAI teams were actively working together on a Sora integration. Half an hour later, Disney was told Sora was dead. A source close to Disney called it "a big rug-pull."

The Timeline

The speed of this unraveling is remarkable even by Silicon Valley standards. Sora — OpenAI's photorealistic video generation tool — launched publicly in September 2025 after eighteen months of demos that stunned the industry. By October, Sora 2 added voice synthesis, lip-syncing, and the ability to insert real faces into generated videos. That last feature triggered immediate backlash when users created deepfakes of MLK Jr., Mister Rogers, and Tupac.

In December, Disney announced a $1 billion investment tied to bringing 200+ Disney characters to Sora. On Monday, March 23, OpenAI published a blog post updating Sora's content policy. The very next morning, they announced Sora was dead — the app, the API, the ChatGPT integration, all of it. The Disney deal, which reportedly never moved beyond a "verbal understanding," collapsed with no money having changed hands.

OpenAI's farewell was brief: "To everyone who created with Sora, shared it, and built community around it: thank you. What you made with Sora mattered, and we know this news is disappointing."

Why Kill Your Most Famous Product?

The answer is GPUs. Every chip running Sora was a chip not running something else. Video generation is extraordinarily compute-intensive, and Sora reportedly generated only around $15 million in revenue — a rounding error for a company hiring 8,000 employees.

At an all-hands meeting days before the shutdown, executives reportedly used the phrase "distracted by side quests" to describe projects like Sora. The company is pivoting hard toward enterprise and coding tools — the space where Anthropic's Claude Code has been eating OpenAI's lunch. The Sora team will transition to "world simulation" research aimed at robotics applications.

Enter Spud

On the same Tuesday that Sora died, Sam Altman sent an internal memo announcing that pretraining had been completed on a new model codenamed "Spud." He described it as a "very strong model" that can "really accelerate the economy," adding: "Things are moving faster than many of us expected."

It's unclear whether Spud becomes GPT-6 or GPT-5.5. No technical details — parameter count, architecture, benchmarks — have been disclosed. Altman said the model would be ready "in a few weeks," meaning post-training (RLHF, safety evaluation) is still underway. Spud may serve as the foundation for OpenAI's planned desktop "superapp" combining ChatGPT, the Codex coding agent, and the Atlas browser.

The Bigger Reorganization

The Spud announcement came wrapped in a broader restructuring. Altman is stepping back from day-to-day product oversight to focus on fundraising and data center construction. Safety now reports to research chief Mark Chen. Security moves under Greg Brockman's scaling organization. And in perhaps the most telling rename, Fidji Simo's product division changed from "Applications" to "AGI Deployment."

OpenAI has reportedly been in an internal "Code Red" state since December 2025, after Anthropic and Google gained significant ground. The company that once set the pace is now playing catch-up in agent-based AI — the category that's actually generating enterprise revenue.

What This Tells Us

Killing Sora isn't just about GPU allocation. It's a confession that OpenAI spread itself too thin. Creative tools, social media features, content partnerships — none of it generated revenue at scale. Meanwhile, Anthropic built Claude Code into a tool that developers actually pay for, and Google kept shipping Gemini updates at a pace that keeps Jensen Huang talking about AGI rather than OpenAI.

The bet on Spud is that a sufficiently powerful foundation model can win back the narrative. Whether that bet pays off depends on what "in a few weeks" actually means — and whether the enterprise customers OpenAI lost to Claude Code are willing to come back.