OpenAI didn't kill Sora. Its IPO did.

This wasn't a pivot. It was a cleanup.

OpenAI shut down Sora this week – the AI video platform it launched to enormous fanfare six months ago – and cancelled the $1 billion Disney deal built around it. The official framing: a strategic shift toward robotics and agentic AI. The actual driver: OpenAI is preparing to go public, and Sora was everything a pre-IPO filing doesn't need.

A resource black hole. Unresolved IP liability. Negative unit economics. Limited monetisation.

Those aren't my words. They're from a Forrester analyst and OpenAI's own head of product. When the people running a platform describe its economics as "completely unsustainable," you're not reading about a feature decision. You're reading about a liability being removed before the SEC starts asking questions.

The numbers that ended Sora

Sora generated $1.4 million in total lifetime net revenue. ChatGPT generated $1.9 billion over the same period.

Every 10-second video cost OpenAI approximately $1.30 in GPU compute. The platform charged $1. At peak usage – millions of videos per day – that runs to an estimated $15 million in daily burn, or over $5 billion annualized. The primary use case, per Sam Altman himself: people making funny memes to send to three friends.

There is no advertising model that supports that cost structure. There is no subscription price that converts the users who showed up for free meme generation. OpenAI knew this by month two. They ran it for four more months.

The reason isn't irrational exuberance. It's sequencing. OpenAI needed the Disney deal. The deal required Sora to stay alive. Now the deal is dead too – and reportedly Disney was blindsided, describing it internally as a "rug pull."

That's the real cost here. Not the $5 billion annualized burn. The $1 billion in Disney equity that vaporized, and more importantly, the precedent it destroyed.

What Disney's exit actually signals

Disney was the template for how AI companies and entertainment studios could coexist.

Two years of legal conflict between Hollywood and AI companies over intellectual property. Lawsuits from studios, authors, musicians. A fundamental question about whether AI platforms would ever pay for the content they trained on or licensed.

Disney's deal with OpenAI in December looked like an answer. License rather than litigate. Structured terms, 200-plus characters, revenue participation. The entertainment industry was watching to see if this model held.

It didn't.

Disney says it "respects" OpenAI's decision and will continue exploring AI partnerships that "respect IP and the rights of creators." That's a gracious statement from a company that was left holding a deal with no counterparty. New CEO Josh D'Amaro is now evaluating Disney's AI partnerships from scratch, alongside a reassessment of the company's $1.5 billion Epic Games stake.

What every studio watching this should take from it: don't anchor your AI strategy to a single platform's product roadmap. The platform may cancel the product before your lawyers finish the paperwork.

The Anthropic divergence

The contrast with Anthropic is uncomfortable, and it's worth being precise about why.

This isn't about Claude being better than ChatGPT in every category. The benchmark picture is more competitive than that – GPT-5.4 (released March 2026) is a serious model, and the frontier is crowded. The divergence is strategic, not technical.

Anthropic decided what it was for. Coding. High-stakes enterprise work. Reliable, deep, agentic task execution. It shipped into that lane and stayed there. Claude Code is the dominant coding assistant for professional developers. Claude Opus 4.6 tops most independent coding and reasoning benchmarks, outperforming GPT-5.2 and Gemini 3. The desktop app is clean, integrated, and doesn't ask you to manage three separate interfaces.

Anthropic didn't build a TikTok competitor. It didn't add shopping. It didn't launch AI hardware for consumers. It shipped one thing – deeply – and compounded.

The market is pricing this divergence explicitly. Polymarket currently puts Anthropic at 58% to hold the best AI model by end of June 2026. OpenAI sits at 12%. More than $1.9 million has traded on this market. That's not a vibe. That's a structured bet, and the spread is not close.

The question worth asking isn't "Is Anthropic winning?" The answer to that is clearly yes, for now. The question is: what kind of winning is this?

Anthropic is winning the year by owning a specific, high-value domain. OpenAI is still winning the decade – 900 million weekly active users, $850 billion post-money valuation, a super-app in development, advertising infrastructure being built. These are not equivalent. But the year belongs to the company with the focused strategy.

What happens next

OpenAI's IPO cleanup is rational. Cut Sora. Kill the shopping feature. Merge the fragmented apps into a super-app. Hire advertising executives from Meta. Raise $10 billion. Clear the narrative before the SEC filing requires you to disclose every failed initiative in detail.

This is how you prepare a company for public markets. The CFO has said exactly this. The head of applications told staff directly: "We cannot miss this moment because we are distracted by side quests."

There's a structural reason this cleanup couldn't wait. Digital businesses scale fast — but they collapse almost as fast. OpenAI's moat isn't physical infrastructure, supply chain, or hard-won physical distribution. It's speed, relationships, and model quality. All three are contestable. That's why the CFO's directive has teeth: a company with no defensible atoms has to make its bits count. Every Sora — every compute-heavy consumer product with no monetisation path — is a liability on the IPO balance sheet and a question mark on the road show.

The execution is happening. The cost of the detour is already paid.

For anyone building AI strategy right now – enterprise or individual – the lesson isn't "OpenAI is failing." It isn't. The lesson is structural: scattered investment across consumer novelty doesn't compound. Focused investment in the workflows people actually pay for does.

Anthropic built for the latter. OpenAI is pivoting to it.

One of them started there. The other is arriving late.