Fordel StudiosOpenAI shuts down Sora after $15M per day burn rate
What Happened
OpenAI shut down Sora, its text-to-video model, on March 24, 2026 after the product reached $15M per day in inference costs. The closure terminated a $1 billion partnership with Disney. Compute resources have been redirected to an internal project codenamed 'Spud.'
Our Take
Called it. Anyone doing the math on Sora's inference costs knew this was coming — $15M/day is $5.5B/year to run a single product. That's not a pricing problem, that's a physics problem.
Video generation is just fundamentally more expensive than text. You're not predicting the next token — you're synthesizing thousands of frames with temporal coherence. The compute curve doesn't flatten the way LLM inference has.
The Disney deal unraveling is the real tell here. That wasn't a small content experiment — that was $1B anchored to the assumption Sora would get cheaper fast. It didn't. So now OpenAI's redirecting compute to 'Spud' (whatever that is) and hoping nobody notices the pivot.
Honestly? This is good news for builders who were skeptical of video AI as a feature. The companies that went all-in on Sora integrations just got burned. If you were waiting to see how the economics shook out — you were right to wait.
For us, this kills the "just add video generation" impulse for a while. The alternatives — RunwayML, Kling, Wan — are still standing, but they're all running the same math just with different VCs footing the bill.
What To Do
If you have any Sora API dependencies, migrate to Kling 1.6 or RunwayML Gen-4 this week — both have stable APIs and lower per-second costs (~$0.05-0.08/s vs Sora's rumored $0.40+/s).
Cited By
React