Days before what could be the largest IPO in history, Anthropic CEO Dario Amodei names the central weakness of his business model. Nobody knows how the company’s compute investments will pay off, he says. It isn’t the first time he’s sounded this warning.
“There’s a huge uncertainty in what’s going to happen next year.” Speaking at a JPMorgan Chase investor event in New York on Friday, Anthropic CEO Dario Amodei used that line to sum up the problem facing his company.
The admission came, of all times, in the last days before the company’s planned Nov. 19 IPO, and compute was one of several topics in the conversation. Speaking with JPMorgan CEO Jamie Dimon, he said, “Compute is the deepest business dilemma that the company faces.”
AI technology is advancing so fast that, Amodei said, no one yet knows how that progress will translate into economic value. But compute has to be bought one to two years ahead, before anyone knows whether revenue will justify the cost.
If Anthropic buys too little, he said, it won’t have the resources for customers and new technologies, and revenue takes a hit. Buying too much carries the opposite risk. “You can financially overextend yourself,” Amodei said.
$901 billion bet on an uncertain future
Anthropic has already committed nearly $901 billion, most of it to compute and chips, according to Bloomberg. Roughly $71 billion of that comes due by September 2027. Bloomberg notes that figure exceeds the market cap of about two-thirds of S&P 500 companies.
Back in February 2026, Amodei warned in much sharper terms that he could buy $1 trillion of compute starting at the end of 2027 and go bankrupt if revenue fell short. Being off by just one year could mean bankruptcy. He is not sure some competitors have done the math, and he said they don’t really understand the risks they’re taking. The stakes have kept rising since then.
Anthropic wants the IPO to cushion against the unknown
Amodei described two ways Anthropic softens the compute dilemma: gross margins and cash on the balance sheet. He said the company’s gross margins are “quite good,” especially compared with other players, and that Anthropic has a large cash balance and is seeking to raise more.
The company also buys relatively little compute upfront, then can turn to the spot market for additional capacity if business goes better than expected. Amodei compared that approach to buying insurance, saying it leaves Anthropic covered if results fall short while allowing it to invest more when things go well.
The compute bets were the biggest reason to go public: the more cash Anthropic holds, he said, the wider its margin for error between a world with too little compute and one with too much.
Anthropic plans to raise up to $64.4 billion through the IPO and another $26 billion through a private placement, Bloomberg reports. Combined, that would make it the largest IPO ever. The offering is reportedly already about six times oversubscribed.
Anthropic’s annualized revenue run rate hit $81 billion at the end of October, up $10 billion from late September, according to company filings. Bloomberg says the rising popularity of AI products, including its coding models, has helped fuel revenue growth.
Amodei struck a relaxed tone about the IPO’s timing, saying he wasn’t fixated on a specific month or particular market conditions. Anthropic, he noted, grew from nothing to its current size in just six years. The most common advice he got before the IPO, or the advice that resonated most, he said, was something like “never let your company be worth a single dollar more than it’s really worth.” That, he said, is very similar to a model Anthropic uses internally: “under-promise and over-deliver.”
In the same conversation, though, Amodei said he hopes AI will cure cancer and Alzheimer’s within five to 15 years, adding that he doesn’t know exactly. In Silicon Valley, that may already count as under-promising.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive “AI Radar” frontier report six times a year, full archive access, and access to our comment section.
