OpenAI Killed Sora -- What the Shutdown Tells Us About AI Video Economics

Six Months

When OpenAI first unveiled Sora in February 2024, the demo videos went viral instantly. Text-to-video that looked cinematic. The hype was enormous. But it took 19 months to ship the actual product, and just 6 months after that September 2025 launch, OpenAI pulled the plug.

On March 24, the company announced that the Sora iOS app, Sora.com website, and Sora 2 API would all be discontinued. The timeline details are still coming, but the direction is final.

The Backstory

Sora's trajectory tells the whole story. The February 2024 demo was arguably the most successful viral marketing moment in AI history. But by the time the product shipped in September 2025, competitors like Runway, Pika, and Kling had already staked out the market.

At launch, Sora became the top app in the Photo and Video category. It peaked in November with 3.33 million downloads across iOS and Android. But by February 2026, downloads had cratered to 1.13 million, a 66% decline in three months.

Period	Downloads (iOS+Android)	Status
September 2025 (launch)	Category #1	Explosive interest
November 2025 (peak)	3.33M	Maximum usage
February 2026	1.13M	66% decline
March 24, 2026	--	Shutdown announced

A 66% drop is not normal post-launch cooling. It means people tried Sora once or twice and did not come back.

Breaking It Down

The Unit Economics Problem

OpenAI officially cited "unsustainable economics of video generation at scale." AI video generation consumes orders of magnitude more GPU resources than text generation. A single video generation burns through compute equivalent to hundreds of text conversations.

With Sora bundled into the $20/month ChatGPT Plus subscription, the math was brutal. The more users actually used Sora, the more money OpenAI lost. A product that punishes its own success.

The IPO Calculation

The bigger play here is OpenAI's upcoming initial public offering. Companies preparing for IPO obsess over profitability metrics. Killing a GPU-hungry video service and reallocating that compute to higher-margin services like coding, reasoning, and text generation is financially rational.

This is not a technology failure. It is a business model failure. AI video generation works, but it does not make money.

The Model Survives

Sora is not completely dead. The Sora 2 model itself remains available within ChatGPT paid subscriptions. What is dying is Sora as a standalone product -- the dedicated app, website, and API. The technology lives on as a feature, not a product.

The Bigger Picture

Sora's shutdown raises uncomfortable questions for the entire AI video market. Runway raised $450M in its Series D. Pika raised $130M. Are they facing the same economics problem?

The core issue is this: AI video generation is technically impressive but has not found its killer use case. Short viral clips? Fun, but users do not come back. Professional video production? Quality is not there yet. Ad creative generation? Promising, but the unit economics do not scale.

If OpenAI could not solve this problem, smaller startups face an even steeper climb.

What This Means for You

If you were building on the Sora API, you have 30 days to find alternatives. Runway's Gen-4, Google's Veo 2, and open-source CogVideo are the realistic options.

More broadly, this is AI growing up. Not every AI capability needs to be a standalone product. Sometimes the right form factor is a feature inside a larger platform. Sora's technology will survive inside ChatGPT. It just could not justify existing on its own.

And yes, people will say "AI bubble." But Sora's shutdown does not mean AI is collapsing. It means the market is starting to filter what actually makes money from what just looks cool in demos.

References

• OpenAI's Sora was the creepiest app -- now shutting down (TechCrunch)
• OpenAI Discontinues Sora App (Bloomberg)
• OpenAI shutting down Sora (NBC News)
• OpenAI shutting down Sora app and API (VideoCardz)