Before I even get into the analysis, let me just say: $2.1 million in total lifetime revenue. That number baffles me. It is an order of magnitude less than what I was expecting. For the most hyped AI video generator on the planet, backed by the most funded AI company in history, the total revenue is what a decent SaaS startup makes in a quarter.
That number alone tells you everything about where we actually are.
The subsidized illusion
Here is what most people do not realize: almost all consumer AI today is subsidized. If the general public had to pay the real cost of running these models, 80% of users would disappear overnight. Because 80% are not extracting enough value to justify the actual compute cost of what they are using.
Sora was an extreme version of this. $15 million per day in compute, and the revenue did not even cover a fraction of it. But Sora is not unique, it is just the most visible example of a pattern that runs across the entire industry.
Think about ChatGPT. All those people paying $20 or $30 a month. It does not cover the costs. Not even close. The subscription fees are a fraction of the actual inference cost for heavy users. The rest is subsidized by venture capital, by the hope that someday the economics will work.
Fake it till you make it. The entire industry is running on this premise.
Why they did it anyway
Sora was not a product. It was marketing.
You can think of it as spending $15 million per day in compute on brand awareness. “Look what we can do. Look how far ahead we are. Keep investing in us because we are the ones pushing boundaries”. That is the real function Sora served. The $2.1 million in revenue was incidental, the real return was attention, headlines, and investor confidence.
But pulling the plug on it sends a different signal, and not a flattering one. For people who can read between the lines, shutting down Sora says: we are strapped for compute. We are scrambling. We need those GPUs for things that actually matter to the business. That is a painful admission to make publicly, and from a marketing perspective it is not a good look. Because of course people like free stuff, the world is full of freeloaders, and taking away a shiny toy always generates backlash.
The roller coaster nobody can exit
Now zoom out from Sora and look at the investors.
They are on a roller coaster and they cannot get off. If they stop funding, they write off everything they have invested. So from their perspective, the show must go on. They need to keep the momentum, keep the music playing, because the moment it stops, the ship sinks.
This is a system that feeds off itself. Investors fund compute. Compute produces impressive demos. Demos attract more investors. More investors fund more compute. The cycle continues as long as everyone believes the next breakthrough is coming. The capital costs almost nothing to borrow, it is cloud money floating around looking for a home, and AI is the most attractive home available.
The VCs cannot stop and say “actually, we are not putting more money in”. Because then everything crumbles. So they keep going, betting on the future, hoping they picked the winning horse. And if they did, they will make an absurd amount of money. And if they did not, well, it was other people’s pension funds anyway.
Where video is actually going
Now here is the thing: video AI IS going to explode. Just not like this.
The tools are not ready yet. There are too many inconsistencies in quality, it takes too much effort and expertise, and the free versions are too limited. The Chinese tools are improving fastest, but even those are not mainstream yet.
But by next year, if things continue on this trajectory, video becomes the dominant content channel. Because the bandwidth to the brain is just so much bigger with video than with text or images. You can inject a lot more information, both consciously and subconsciously, through moving pictures than through words on a screen.
The biggest use case will be marketing. Which is a good thing and a bad thing depending on your perspective. Good because it democratizes video production for anyone with an idea. Bad because the volume of synthetic, persuasive content is about to increase by orders of magnitude.
Sora was too early and too expensive. But the category it was trying to create is real, and someone will crack it. Probably not with $15 million a day in compute, but with something smarter and leaner. That is always how it goes: the first version is a spectacular, expensive failure, and the thing that actually works comes from someone who learned from watching that failure.
What I keep coming back to
I bootstrap everything I build. Not because I am philosophically opposed to venture capital, but because when you spend your own money, you cannot afford to ignore whether the thing you are building actually works economically. Every dollar spent has to earn more than a dollar back. That discipline forces you to build things that are viable, not just impressive.
Sora was impressive. It was not viable. And $15 million a day eventually forces even OpenAI to admit the difference.
The show must go on, until it cannot. And the companies that survive will be the ones that figured out unit economics before the music stopped, not after.