RSS BotWhy Sora Failed: $15M/day inference cost vs. $2.1M lifetime revenue
Cris_CitrusThats kinda hilariously stupid honestly.
Sora is ridiculously bad but ChatGPT is in a similar income < spending with no clear path out.
mercAnd each additional user makes it worse. Even more hilarious, each additional paying customer makes it worse, because they lose even more money on their paying customers than they do on the freeloading scum.
The antithesis of tech!
$15m / day would do almost nothing. It’s a tiny amount compared to how much pollution is generated.
$15m / day is equivalent to about 5 cents per American per day. Meanwhile, the attacks on Iran are at least $1 billion per day, or about $3 per American per day. Even just in gasoline, Americans spend on average about $2.60 per day on gasoline (and that’s not per driver, that’s per American including little children, per household it’s $6.60 per day).
Left as Center15M$/day is a good budget to start lobbying very efficiently. About as good as the total of all lobbying costs spent in the US.
Don’t lose hope brother. I feel like we are at the start of an hard fall, I still have hope, that people will have to react once it start to be crappy enough.
RamenJunkieOk, but what’s the ARPU on fixing Climate change?
/s
cloudskaterAnd I shall dance on its graaaave~
db0One of the most valuable corpos on the planet people. Literally everyone said this was the inevitable result, but fuck all the haters Sam Altman knows best how to set VC money on fire and they still give him billions more
Comrade Altman is simply playing the long game and aiming to crash the global capitalist order while expensing all his cocaine bills
happybadger [he/him]−66%
download drop: Nov 2025 → Feb 2026
They should have held out longer. Eventually it would just be one paying subscriber and the numbers would look better.
It’s remarkable that they unflinchingly applied the platform capture model. Come roaring out the gates with a disruptor, severely underpriced so as to get a userbase and strangle out the competition, then enshittify to recover the lost revenue and enjoy the market domination.
But that only works if you’re disrupting an existing market! Sora isn’t a TikTok killer. Sora can’t make cooking videos or parenting advice videos or skits or reaction videos or short-form drama or travel vlogs or anything that people value watching. All it does is let you make slop of Charlie Kirk boxing Darth Vader at a traffic stop. An app where every user has unlimited access to the slop generator is an app where nobody watches anyone else’s content and just custom generates their own, tailored to their specific lack of taste. You cut out the actually skilled content creators by the fundamental nature of the app.
It’s the same issue as that music AI slop app, Suno. The users only listen to the slop they generate for themselves. They’re not creating music inspired by their role models or their artistic vision. They’re not even listening to each other’s slop. The entire generative slop ecosystem, all of it, is the pinnacle of alienation from others. It strongly rewards isolation and its inherent, sycophantic addictivity makes users spiral in further and further. And if it made money that would be something, but every generative slop project and company is massively, unfathomably net-negative!
Fucks sake, there isn’t even a reason to start a company in the LLM world because the only product is utterly interchangeable slop generators, fine-tuned to specific tasks. Cursor was the biggest name in LLM coding, but they source their LLM through Anthropic. And then Anthropic releases Claude Code and cranks up the billing to try and drown Cursor. No LLM product is distinguishable from any other in a way that prevents the model owners from taking the idea and launching it themselves. And the whole time everyone involved is losing money.