Sora's last day: $1M/day burn, $2.1M lifetime revenue, no replacement

On April 26, 2026, OpenAI shut down Sora’s web app and mobile apps. The numbers OpenAI disclosed in the lead-up made the post-mortem painfully clear:

• ~$1 million per day in compute costs to run Sora at peak
• Peak ~1 million active users, falling to under 500,000 by the time the shutdown was announced (March 24)
• $2.1 million in total lifetime in-app revenue

Run that math: a single year of Sora compute, at $1M/day, is $365M/year. Lifetime revenue: $2.1M. The product was unprofitable by a factor of roughly 170×.

The reasons it died

Cost ratio. $1M/day in compute against $2.1M total lifetime revenue is the simplest possible explanation. No path to break-even existed at any plausible scaling.

User decline. Halving from 1M to 500K active users in roughly 12 months is the steepest decline of any flagship OpenAI product. Compare to ChatGPT, which has scaled from 100M to 400M+ over the same period.

Copyright litigation. Generative video that produced Disney characters, Pixar characters, and other copyrighted IP became an existential legal liability. The December 2025 Disney $1B partnership was supposed to neutralize this. Instead, Disney walked back the deal and Sora was sunsetted.

Strategic redirect to enterprise. OpenAI’s $122B funding round in March was framed around enterprise readiness ahead of an IPO. Consumer-facing video that lost money was not on the IPO narrative. Coding tools, agents, and Codex Enterprise — products with paying enterprise customers — are.

What “no official replacement” actually means

OpenAI has not announced a Sora 3 or a successor. The video category at OpenAI is, for the foreseeable future, dead.

That doesn’t mean OpenAI is leaving generative video forever. Two things to watch:

1. GPT-5.5’s native multimodality. The model already handles video understanding and the rumored GPT-5.6 is expected to have video generation as a model-native action class — not a separate product. If/when it arrives, it’ll be inside ChatGPT, not as “Sora 3.”

2. Image-to-video integration with ChatGPT Images 2.0. ChatGPT Images 2.0 shipped this week with “thinking”-enabled generation. The architecture for adding short-clip animation to existing images is in place. Months, not years, before that’s plausible.

For users right now: there is no successor. You move to a competing platform.

The actually-comprehensive replacement decision tree

Today, with all the news of the past week factored in:

Photorealistic short clips with synchronized audio: Veo 3.1 via Google AI Pro ($19.99/mo) — only mainstream tool with native audio.

Cinematic professional video, Sora-style aesthetic: Runway Pro ($28/mo) — Gen-4.5 produces the highest-quality cinematic output of anything currently available.

Long single-take clips (up to 2 minutes): Kling Pro ($26.99/mo) — nothing else comes close on duration.

Artistic short-form clips at lowest cost: Midjourney V1 Video — shipped this week, 5-second clips, ~25× cheaper than Sora was. If you already have Midjourney Standard, this is essentially included.

Free-tier experimentation: Veo 3.1 free tier (10 generations/month per Google account), Kling free tier (66 daily credits).

TikTok / Reels / Shorts editing on top of generated clips: CapCut free.

The full ranked breakdown is on the best AI video tools page.

The lesson for AI buyers

If you build a workflow on a single AI product, that workflow can disappear with two weeks’ notice. Sora’s user base — small as it was — was given 33 days from announcement (March 24) to shutdown (April 26). For anyone whose video pipeline depended on Sora, that’s not a lot of time to migrate.

Three rules I now apply when picking AI tools:

1. Is this a vendor’s main product line, or a side bet? Main product lines survive. Side bets get cut. Sora was a side bet. Runway is Runway’s main product line. Veo is one of Google’s main product lines. ChatGPT is OpenAI’s main product line.

2. Is the unit economics plausible? $1M/day compute against $2.1M lifetime revenue should have been visible from the outside if anyone had bothered to estimate it. Today, the same question applies to Sooth Labs, Project Prometheus, and a dozen other “no clear monetization” AI startups.

3. Is there an open-weight or self-hostable fallback? For categories where there is one (text generation: Llama, Mistral, DeepSeek; image generation: FLUX, Stable Diffusion), vendor-lock-in risk is bounded. For video, the open-weight option (Mochi, CogVideoX, Wan 2.1) is still meaningfully behind closed alternatives. That’s a real risk premium baked into Veo / Runway / Kling for any production workflow.

What I’m actually doing tomorrow

Backing up my own Sora exports today (server load is climbing, export endpoint here — do it now, not at 11pm tomorrow). Migrating my one ongoing Sora project to Veo 3.1 since it needs voiceover. Keeping Runway Pro for cinematic work. Trying Midjourney V1 Video this week since I already pay for Midjourney Standard.

Not subscribing to a new product to replace Sora. The lesson Sora taught is that I had too many AI subscriptions in the first place.

Related:

• Sora shutdown migration plan (April 23 article)
• Midjourney V1 Video launch
• Best AI video tools in 2026
• Full Sora review

Questions or corrections? Email Pick Right. Want the full list? See all news.