Anthropic CEO Dario Amodei's message to Sam Altman and Jensen Huang: You have got your calculations all wrong, and this may lead to...

Dario Amodei, the CEO of Anthropic, has a blunt warning for the AI industry's biggest spenders: if your revenue projections are off by even 20%, no amount of deal-making will save you from bankruptcy. And he thinks some of his rivals haven't even done the basic math.In an interview on the Dwarkesh Podcast, Amodei laid out what he called the "spreadsheet dilemma" facing companies that are betting trillions on AI compute infrastructure. While he didn't name OpenAI, Sam Altman, or Jensen Huang directly, the target was hard to miss—OpenAI's $300 billion cloud deal with Oracle, Nvidia's reported $100 billion investment interest in OpenAI, and the broader Stargate project that has already crossed $400 billion in commitments. Amodei's core argument is simple: the AI industry is locking itself into massive, non-cancellable compute orders based on revenue projections that assume near-perfect execution. And in his view, some companies are "YOLOing" their way through trillion-dollar bets without fully understanding what happens if growth slows even slightly.Amodei used a concrete hypothetical to make his point. If a company commits to $1 trillion a year in compute spending—say, $5 trillion over five years starting 2027—it needs revenue to match. If it lands at $800 billion instead of $1 trillion, "there's no force on earth, there's no hedge on earth, that could stop me from going bankrupt," he said.The math is unforgiving. Data centres take one to two years to build. Chip orders are locked in well in advance. And if growth slows from 10x a year to even 5x, or arrives just 12 months late, the entire financial model collapses. "If I'm just off by a year in that rate of growth… you go bankrupt," Amodei added.The timing of these comments is hard to ignore. OpenAI signed a five-year, $300 billion cloud deal with Oracle in September 2025, covering roughly 4.5 GW of AI-optimised capacity per year. That single contract ballooned Oracle's remaining performance obligations by $317 billion—meaning roughly 58% of Oracle's backlog now hinges on OpenAI delivering the revenue it's projecting. Meanwhile, OpenAI reported losses of $13.5 billion on just $4.3 billion in revenue in the first half of 2025.Add Nvidia's reported $100 billion investment interest in OpenAI—where OpenAI buys Nvidia chips, Oracle hosts them, and revenue flows in a circle—and the picture starts to look fragile.Amodei positioned Anthropic as the cautious alternative. His company, he said, commits to "hundreds of billions" in compute, not trillions, and has actually written down the spreadsheet. "I get the impression that some of the other companies have not written down the spreadsheet, that they don't really understand the risks they're taking," he said.Anthropic's own revenue trajectory—from essentially zero in 2023 to a $9-10 billion run rate by early 2026—gives it room to be selective. Whether that restraint looks smart or timid will depend entirely on how fast AI revenue actually materialises across the industry.