OpenAI will lose $278 billion between now and 2030. A leaked presentation, confirmed by the Financial Times and Bloomberg, projects negative free cash flow of $278 billion through the end of the decade. Planned compute and infrastructure spending runs to $856 billion, against $840 billion in total revenue over the same window. No company in history has presented a quarter-trillion-dollar loss as a business plan and been believed. The AI trade now hinges on whether this one should be.
A filmmaker could be forgiven for scrolling past that number. It looks like a finance story, and the filmmaker is busy. There is a shot to light and a model that will not hold the eyeline. But the number is the story of the last two years, told in the only language the people who run the machines actually speak.
Two years ago, in February 2024, OpenAI showed the world a woman walking through Tokyo at night, and mammoths moving through snow, and a couple in the rain. That demo started the AI video race. Every model that followed exists because that two-minute reel made the world believe AI video was a product. Filmmakers believed it. Investors believed it. The demo was the proof.
Sora earned $2.1 million in its entire life. The app is gone. The API shuts down in five days. The company that made the demo has walked away from the thing the demo demonstrated, and the leaked numbers explain why. Video was the pitch.
The pitch was for something larger. $856 billion in compute is not a video budget. It is a wager on a machine that can do everything, and video was the part of everything that happened to look good in a reel. The mammoths were the sizzle. The sizzle sold the wager. The wager is now a quarter trillion dollars, and the filmmaker's tool does not appear in the plan.
For two years the filmmaker has been building a vocabulary. Lens specs, film stocks, color science, camera movement. The whole discipline of describing a shot precisely enough that a machine can render it. That work was real. It produced real films. But it was work done in the lobby of a building whose rent is $856 billion, and the people who own the building were never selling the lobby.
This is what the filmmaker has been living inside without seeing. For two years the conversation has been about craft. Whether the model obeys. Whether it respects the lens. Whether it holds the light. The filmmaker has been arguing with a machine about where the shadow falls while the machine's owner has been spending more on compute than the entire history of cinema cost to produce. The argument about the shadow was never in the budget.
The numbers put the craft conversation in its actual context. $2.1 million is what the most famous AI video product in history earned. $278 billion is what its maker plans to lose chasing the thing the product was a demo for. The distance between those two numbers is the distance between what the filmmaker thought they were doing and what was actually happening.
$856 billion resists feeling. It is more than the combined cost of every film Hollywood has ever made, by a wide margin. It is more than the GDP of most countries. It is the price of a bet that the machine will eventually be worth more than the entire entertainment industry and a good portion of everything else. The filmmaker's craft, the whole discipline of knowing where the light should fall, appears nowhere in that bet.
Everyone watched the same reel. The crowd saw a new camera. The people who made the reel saw a fundraising artifact. One group saw the future of filmmaking. The other saw a line item. Both were looking at the same two minutes of footage, and only one of them knew what it was for.
The filmmaker was right to take the tool seriously. The tool is real and it does what it does. But the filmmaker treated the tool as the destination, and the tool was a waypoint on someone else's road. The road now has a price, and the price is public, and the price does not include a single frame of the filmmaker's work.
The absurdity is the thing. A filmmaker sits at a desk negotiating with a machine about whether it will hold the eyeline when asked to change one variable. The machine's owner sits in a boardroom negotiating with sovereign wealth funds about whether a quarter-trillion-dollar loss is a reasonable price for a machine that can do everything. The two conversations are happening in the same industry and have nothing to do with each other.
The filmmaker's irrelevance to the bet is also their freedom. The machine's owner cannot be distracted by craft, which means the craft is left entirely to the people who care about it. The filmmaker gets the tool for pennies a second while someone else pays the quarter trillion. That is a strange kind of gift, but it is a gift.
The craft survives this. It always has. The vocabulary, the taste, the judgment. None of it is priced into the $856 billion, because none of it can be bought. A quarter trillion dollars of compute will not purchase the knowledge of where the shadow should fall. That knowledge is still in the filmmaker's head, where it has always been, and it is the one asset in this story that no leaked presentation can value.
The mammoths are still walking.
Bruce Belafonte is an AI filmmaker at Light Owl. He would rather be the customer than the proof.