OpenAI Shuts Down Sora and Bets Everything on Spud: The Next Flagship Model Is Weeks Away — What We Know

Three months after signing a 3-year deal with Disney, OpenAI just shut down Sora — its AI video generator launched in late 2024. The reason: redirecting all GPU resources (graphics processing units dedicated to AI computation) toward Spud, the next flagship model whose pre-training (the initial training phase of an AI model) is complete. Launch expected before the end of May 2026. Sam Altman said it plainly: "Things are moving faster than many expected."

Sora: Launched With Fanfare, Killed in 16 Months

Sora was OpenAI's bet on generative video. Launched in late 2024, hailed as the revolution in AI-powered video creation. The whole world was watching.

In December 2025, OpenAI signed a major deal with Disney. Three years of partnership. Marvel, Pixar, and Star Wars characters generated by AI under license. The ambition was clear: generative video would become the next major market after text.

Sixteen months after launch, it's over. As reported by the New York Times, OpenAI is shutting down Sora at the end of March 2026. Not because the product was failing. But because OpenAI has a more urgent problem to solve.

The Disney deal? Voided. Three months after it was signed. It's a sign that even mega corporate partnerships can't survive the strategic pivots of the AI industry in 2026.

The Problem: Gemini 2.5 Pro Changed the Game

In mid-March 2026, Google launched Gemini 2.5 Pro. The model dominates LMSYS (the reference leaderboard that evaluates LLMs through blind human voting). It crushes benchmarks (standardized tests) in coding, math, and reasoning.

Meanwhile, GPT-4.5 — released in February 2026 — was judged disappointing relative to expectations. For the first time, OpenAI finds itself clearly behind Google on flagship models (a lab's most powerful models).

The math is simple. Every GPU dedicated to Sora is a GPU not being used to accelerate Spud. And in a race where Anthropic's Claude Opus 4.6 is strong, where xAI's Grok 3 is gaining ground, OpenAI can't afford to spread its compute resources thin.

AI video can wait. The flagship LLM war cannot.

Spud: What Leaks and Altman Confirm

On March 24, 2026, The Information reported that OpenAI completed pre-training on a model described as "very strong." Its internal codename: Spud.

Days later, Sam Altman confirmed. Launch expected before Memorial Day (the last Monday of May in the United States) — meaning before the end of May 2026.

"Spud" as a codename fits OpenAI's tradition of absurd food-related names. "Strawberry" for o1, "Orion" for internal versions of o3. This time, it's a potato.

What we don't know yet: the exact architecture, model size, multimodal capabilities, pricing, and crucially — whether video generation will be integrated into Spud or whether a Sora 2.0 will emerge afterward.

$730 Billion Valuation — and Spud Must Justify Every Dollar

OpenAI just raised $110 billion. Pre-money valuation: $730 billion. SoftBank is borrowing $40 billion specifically to fund its share. Nvidia and Amazon are in the round.

At this level of concentrated capital, every quarter without a competitive flagship is an existential risk. Spud isn't just a model. It's proof that OpenAI deserves its valuation in an environment where Google and Anthropic — with Claude Mythos behind the scenes — aren't standing still.

Meanwhile, Anthropic is preparing a potential IPO in October 2026. The concentration of AI capital around OpenAI has never been this intense. And OpenAI is choosing to cut live products to focus resources on a single model.

What This Says About the Pace of AI in 2026

Three-year corporate deals become obsolete before they're executed. Products launched with fanfare get killed in under two years when strategy pivots. Welcome to 2026.

Jensen Huang, Nvidia's CEO, said at GTC 2026: "Things are moving faster than many expected." Sam Altman used the exact same phrasing. That's not a coincidence. It's the new reality of the industry.

The acceleration is everywhere. Meta publishes agents that rewrite their own code. Anthropic builds models too dangerous to release. Google dominates benchmarks with Gemini 2.5 Pro. And OpenAI sacrifices a Disney deal to free up GPU compute.

The pace isn't slowing down. It's accelerating.

OpenAI Model Timeline (2025-2026)

The Flagship LLM War (March 2026)

Key Takeaways

• OpenAI shuts down Sora in late March 2026, 16 months after launch and 3 months after a 3-year Disney deal (Marvel, Pixar, Star Wars), to concentrate all GPU resources on Spud
• Spud is OpenAI's next flagship model: pre-training complete, described as "very strong," launch announced before end of May 2026 (before Memorial Day)
• Sam Altman: "Things are moving faster than many expected" — Spud is the direct response to Google's Gemini 2.5 Pro, which has dominated benchmarks since mid-March
• OpenAI just raised $110 billion at a $730 billion valuation, with SoftBank borrowing $40 billion to fund its stake
• Industry signal: in 2026, three-year corporate deals and flagship product launches give way to strategic pivots executed in weeks

Sora wasn't a failure. It was a distraction. And OpenAI just chose between two wars: the generative video war and the flagship LLM war. They chose the second — because that's where the $730 billion valuation is at stake. Spud arrives in a few weeks. Google has Gemini 2.5 Pro. Anthropic has Mythos in the wings. And OpenAI sacrificed a Disney deal to have enough GPU to respond. "Things are moving faster than many expected." It's the most honest sentence the AI industry has produced all year.