OpenAI's Sora Shuts Down April 26 After Burning $15M a Day Against $2.1M in Revenue

By Hector Herrera | April 15, 2026 | Business

OpenAI's Sora video generation app shuts down on April 26, with the API following on September 24. Less than two years after its high-profile launch, Sora is being discontinued — the product burned an estimated $15 million per day in compute costs against lifetime revenue of just $2.1 million. It is the most expensive failed product launch in AI history.

The shutdown is not just about one product. It signals a fundamental strategic reorientation at OpenAI: away from consumer multimedia and toward enterprise coding tools.

Background: Sora's Promise and Its Reality

OpenAI unveiled Sora in February 2024 to immediate industry shock. The model could generate cinematic, temporally consistent video from text prompts — a capability that had seemed years away. The demos were stunning. Coverage was wall-to-wall.

Sora launched publicly in December 2024. By early 2025, the gap between demo and product was clear. Generation quality was inconsistent, speeds were slow, and the use cases that justified premium pricing — professional video production, advertising, film — remained elusive for most users.

According to The Decoder, the WSJ reported Sora consumed an estimated $15M per day in compute costs at peak. Against that burn rate, lifetime revenue of $2.1M represents a staggering misalignment between cost and monetization.

The Numbers

• Daily compute cost (est.): $15 million
• Lifetime revenue: $2.1 million
• Active users at shutdown: Under 500,000 (down from 1 million at peak)
• App shutdown date: April 26, 2026
• API shutdown date: September 24, 2026

The two-stage shutdown — app first, API months later — gives enterprise developers and businesses that built on Sora's API a runway to migrate. September 24 is a hard deadline.

Why Video AI Is So Expensive

Video generation is fundamentally more compute-intensive than text or image generation. A single second of high-quality video requires generating hundreds of frames with temporal consistency — each frame spatially coherent and consistent with the frames before and after it.

At OpenAI's scale, serving Sora to hundreds of thousands of users required maintaining enormous GPU infrastructure running at high utilization. The economics only work if users pay enough per generation to offset that cost — and at the price points where users are willing to pay, the math does not work.

The core problem: Video generation costs are largely fixed (you need the compute regardless of demand), while Sora failed to build the user density to spread those costs. Under 500,000 active users is not enough to sustain a product at this cost structure.

What OpenAI Is Doing Instead

The retreat from Sora is not a retreat from product ambition. OpenAI is explicitly shifting resources toward:

• Enterprise coding tools — the successor to the Codex API and the foundation of GitHub Copilot's continued evolution
• Productivity tools — document analysis, knowledge management, and workflow automation aimed at enterprise buyers
• ChatGPT Pro — the subscription tier that has shown clearer monetization than consumer media tools

This pivot reflects where the money actually is. Enterprise software buyers have established procurement processes and can pay meaningful recurring fees. Consumer video generation requires either a massive user base at low price points or a small user base at very high price points — neither of which Sora achieved.

Implications for the AI Video Industry

Sora's shutdown will be read as a cautionary tale, but it should not be read as evidence that AI video is a dead end. Several dynamics are still in play:

• Compute costs are falling. The economics that killed Sora at 2024-2026 infrastructure prices will look different as GPU efficiency improves and competition drives cloud costs down
• Specialized competitors remain. Runway, Kling, and Pika are continuing to develop AI video tools with more targeted use cases and pricing models
• The enterprise use case is still being discovered. Advertising, training video, and entertainment post-production all have genuine demand for AI video — the path to monetization there is clearer than consumer

For businesses that built on Sora's API: The September 24 API deadline is firm. Start evaluating Runway's API, Kling's enterprise tier, and Google's Veo 2 now. Waiting until September creates unnecessary migration risk.

What to Watch

OpenAI's product announcements through summer 2026 will reveal how aggressively they're pursuing the enterprise coding and productivity market. Expect expanded Codex capabilities and potentially a dedicated enterprise coding product to fill the strategic space Sora was supposed to occupy. Also watch whether any of Sora's underlying research is open-sourced — OpenAI has occasionally released research from discontinued products, and the technical work behind Sora's video consistency is genuinely valuable.

Hector Herrera is the founder of Hex AI Systems and editor of NexChron.