Zitron Argues Against Any AI Bailout as OpenAI Burn Hits $852B

Ed Zitron's argument is uncomplicated, and that is what makes it worth sitting with. The generative AI industry is not systemically important, its demand is largely manufactured, and when it breaks the correct response is to let it break. In his latest essay for Where's Your Ed At, he lines up the revenue and the capex side by side and asks whether the money coming in looks anything like the money going out.

The revenue side is thin. Zitron puts OpenAI at $5.7 billion and Anthropic at a little under $5 billion in the first quarter of this year, and by his reading those two firms represent 89% of the revenues of the largest AI companies, with the rest of the sector pulling in around $20 billion annually. The capex side is enormous. He points to more than $765 billion in capital expenditures in 2026 and a trillion more in 2027 across the hyperscalers, against NVIDIA's own claim of $1 trillion in GPU sales by the end of 2027. On his math, actually consuming that hardware would require the industry to spend $435 billion or more a year on compute, and he does not believe that demand exists at anything like current prices.

His core analogy is that GPUs are not fiber optic cable. Fiber sat cheaply once laid; GPUs depreciate fast, cost a lot to run, and get worse relative to the next generation. He treats several other patterns as evidence of subsidized demand: subscription customers, he writes, can get $40 of compute for a dollar, and NVIDIA has been promising to rent back any unused capacity from cloud providers buying its GPUs, which he reads as the vendor backstopping its own sales.

The pointed part is the politics. Zitron dismisses Sam Altman's floated idea of handing the US government a 5% stake in OpenAI, worth around $42 billion, as trivial against OpenAI's projected $852 billion burn through the end of 2030. His prescription is a single line: "No bailouts, no handouts, no special treatment, no tax breaks, no CHIPS act, and no sovereign wealth fund."

The honest caveat is that this is a bear case, not a forecast, and it rests on Zitron's reading of the demand curve. The piece does not model what enterprise usage looks like if subscription prices were reset, and it does not break out which private credit funds carry the heaviest data-center exposure. What it does give leaders is a clean check on the assumption that a policy backstop is coming for AI capex if the numbers roll over. If he is even directionally right about who gets hurt on the way down, the exposure is much wider than NVIDIA and the two frontier labs, and boards signing off on another year of infrastructure commitment should be asked how they would stand if he is.