The Clip That Cost $18

OpenAI spent more money per day running Sora than most startups raise in a seed round. When the shutdown announcement dropped on March 24, the surprise wasn't that it was dying — the surprise was that anyone expected it to survive.

#The arithmetic that killed a product

A single 25-second Sora 2 video cost OpenAI roughly 18 in raw compute. Users paid 4 to 8. That gap produced a burn rate north of 120 million per month, or roughly 15 million per day at peak usage. Against that, the platform generated a grand total of 2.1 million in lifetime revenue. Not per month. Total.

Downloads hit 3.3 million across iOS and Android in November 2025, the first month after public launch. By February 2026 they'd dropped 66%, to about 1.13 million. Active users cratered from a million to under half a million. OpenAI was subsidizing a shrinking audience at catastrophic scale — a flywheel spinning in the wrong direction.

#The Disney detail

Disney had pledged $1 billion and character licensing in December 2025. No contract was signed. No money moved. When the shutdown went public, Disney reportedly found out less than an hour before everyone else did. A billion-dollar partner doesn't get blindsided by a press release unless the billion was always aspirational.

#Why the IPO pulled the trigger

OpenAI's public offering is expected late 2026 or early 2027, targeting somewhere between 830 billion and 1 trillion. A product hemorrhaging $15 million daily with negligible revenue doesn't survive due diligence. This wasn't a technology failure — the model was technically impressive. It was a balance sheet problem. The kind institutional investors highlight in red ink before asking "what else are you subsidizing?"

The consumer app dies April 26. The API stays live until September 24, giving developers a six-month migration window that, judging by the Reddit threads, most are treating as a three-week sprint.

#What a dime buys you elsewhere

The alternatives are dramatically cheaper to operate, which makes the whole fiasco look even more avoidable.

But here's the thing nobody putting together comparison tables wants to acknowledge: the real cost of any of these tools is 3–5x the per-clip price. Roughly 80% of your credits go to failed generations, re-rolls, and experiments that never produce publishable output. That 0.70 Kling clip is really 2–3 once you account for the four duds you generated first. The table above is a menu, not a bill.

#Where the users scattered

Kling saw global weekly active users jump 4% to 2.6 million the week after the announcement. Not a stampede, but a clear signal. The community consensus across migration threads breaks down roughly like this:

Runway Gen-4.5 inherited the creators who prioritized cinematic feel — it's the closest visual match to what made early Sora demos go viral. Kling 3.0 picked up the practical users, largely because it generates clips up to two minutes long, nearly five times what Sora ever managed. Seedance 2.0 from ByteDance has the most raw momentum among new entrants, partly because CapCut integration makes it immediately useful rather than a standalone toy. And Luma Dream Machine carved out a niche in physical simulation — lighting behavior, material interaction, atmospheric effects — that none of the others prioritize.

Nobody is calling any of these a drop-in replacement. Each has its own quirks, pricing structure, and quality envelope. Creators who built workflows around Sora's specific aesthetic are adapting their expectations, not porting their prompts.

#Every tool on this list is subsidized

That's the uncomfortable part of the post-Sora landscape. Kling's $0.07-per-second pricing isn't profitable — it's Kuaishou buying market share. Runway's pricing model assumes continued venture capital. Google can absorb Veo's compute costs because they round down to zero against cloud revenue.

The difference between Sora and its competitors isn't that one was subsidized and the rest aren't. It's that Sora was subsidized at a scale that made the subsidy impossible to hide. 15 million a day shows up on a balance sheet. 500K a day doesn't — at least not until the next funding round.

At some point, every one of these platforms will need to either raise prices or find architectural efficiency gains — distillation, quantization, better attention mechanisms — that make current pricing sustainable on actual revenue. The ones built on architectures that can optimize will survive. The ones that scaled a demo without ever making inference cheap will follow the same trajectory, just on a longer fuse.

Two and a half weeks until April 26. Download your Sora videos if you care about them. Scope your API migration if you've been procrastinating. And the next time you're evaluating which platform to bet a production workflow on, maybe spend less time on quality benchmarks and more time on the company's burn rate. The prettiest clip in the world doesn't help when the service evaporates.