Days Before Its IPO, Amodei Tells Jamie Dimon That Buying Compute Is Anthropic's 'Deepest Business Dilemma'
Anthropic's CEO says the company must commit to computing power one to two years before it knows whether revenue will cover the bill, with about $900 billion already promised. His answer: buy relatively little up front, keep a lot of cash and rely on the spot market if demand runs hot.
Published
Last updated / Fact-checked on

Illustrative photo of server hardware. It does not show Anthropic's computing infrastructure.
Most chief executives use the final days before an IPO to sound confident. Dario Amodei used part of his to explain the biggest risk in Anthropic's business.
"Compute is the deepest business dilemma that the company faces," the Anthropic CEO said on Friday, Nov. 13, in a conversation with JPMorgan Chase CEO Jamie Dimon at a JPMorgan event, Bloomberg News reported. Anthropic plans to price its shares on Wednesday and start trading on Thursday, Business Insider reported.
The problem, as Amodei described it, is timing. "Because this field is growing so fast from a technological perspective, and we don't know how that is going to transfer to the creation of economic value on the business side, we have this huge uncertainty," he said, according to Business Insider.
At a glance
| Item | Detail | Source |
|---|---|---|
| Where | Conversation with JPMorgan CEO Jamie Dimon, Friday, Nov. 13 | Bloomberg; Business Insider |
| Core problem | Compute must be bought one to two years before demand is known | The Decoder |
| Commitments | About $900 billion in infrastructure, about half due by 2030 | Business Insider, citing the prospectus |
| Near-term bills | About $71 billion due by September 2027 | The Decoder, citing Bloomberg |
| Cash cushion | More than $95 billion in cash plus $36 billion in short-term investments, as of Sept. 30 | Business Insider, citing the prospectus |
| Revenue run rate | $81 billion at the end of October | The Decoder, citing company filings |
| IPO timing | Pricing Wednesday, Nov. 18; trading Thursday, Nov. 19 | Business Insider |
Buying for a future nobody can see
According to The Decoder, Amodei said compute has to be bought one to two years ahead, before anyone knows whether revenue will justify the cost. Business Insider reported that he said the problem gets harder each year because the costs keep growing. "It's a very difficult problem," he told Dimon.
Both mistakes are expensive. If Anthropic buys too much compute, its costs could race far ahead of its revenue. "You can financially overextend yourself," Amodei said, according to The Decoder. If it buys too little, he said, it may be unable to give customers all the AI they want or to do the research needed for its next model, Business Insider reported.
Anthropic tries to forecast demand from the momentum it sees in new products and hits such as Claude Code, Amodei said, according to Business Insider.
The size of the bet
The numbers behind that dilemma are very large. Business Insider said Anthropic's IPO prospectus discloses about $900 billion in infrastructure commitments, with about half due by 2030. The Decoder, citing Bloomberg, put the total at nearly $901 billion, most of it for compute and chips, with roughly $71 billion due by September 2027. Bloomberg noted that the near-term figure is larger than the market value of about two-thirds of S&P 500 companies, according to The Decoder.
Amodei has raised this worry before. The Decoder said he warned in February 2026 that the company could buy $1 trillion of compute starting at the end of 2027 and go bankrupt if revenue fell short.
Two cushions and a spot market
Amodei described two ways Anthropic protects itself, Business Insider reported. The first is gross margin, the amount of money the company makes from each unit of compute. The Decoder said he called Anthropic's margins "quite good," especially compared with other companies. The second is cash. Business Insider said the prospectus shows more than $95 billion in cash and cash equivalents and another $36 billion in short-term investments as of Sept. 30.
Strategy matters too. Amodei said Anthropic buys "relatively little" compute in advance, which leaves it in good shape if demand lags, Business Insider reported. If demand runs higher than expected, it has the cash to buy extra capacity on the spot market. The Decoder said he compared the approach to buying insurance.
"For periods of time where things don't go as well, we're well covered, and we're well insulated," Amodei said, according to Business Insider.
Why the IPO is part of the answer
The Decoder reported that Amodei described the compute bets as the biggest reason to go public: the more cash Anthropic holds, the wider its margin for error between having too little compute and too much.
That cash pile is about to grow. The Decoder, citing Bloomberg, said Anthropic plans to raise up to $64.4 billion in the IPO and another $26 billion in a private placement, which together would make it the largest IPO ever. The offering is reportedly about six times oversubscribed. Business Insider said Anthropic is targeting a $1.4 trillion valuation.
The company's revenue is growing fast. Its annualized run rate reached $81 billion at the end of October, up $10 billion from late September, according to company filings cited by The Decoder.
Under-promising, by Silicon Valley standards
Amodei also struck a relaxed tone about the IPO itself, according to The Decoder. He said he was not fixated on a particular month or market conditions. The advice that stuck with him most, he said, was to "never let your company be worth a single dollar more than it's really worth," which he compared to an internal motto: "under-promise and over-deliver."
In the same conversation, The Decoder reported, he said he hopes AI will cure cancer and Alzheimer's disease within five to 15 years, while adding that he does not know exactly.
Why it matters
Investors in next week's offering are buying into a company whose costs are largely fixed years in advance while its revenue depends on demand that is still being discovered. Amodei's remarks put that mismatch in plain terms, just days before the shares are priced. His answer, a lean upfront compute position backed by a very large cash cushion, is a bet that it is safer to buy extra capacity later at spot prices than to commit too much too early.
It also shows how central compute has become to AI company finances. The biggest risk Amodei named was not a rival model or regulation. It was how much computing power to buy and when.
What to watch
Watch the final IPO price on Nov. 18 and whether Anthropic uses its new cash to change how much compute it commits to in advance. Watch for any updates to the prospectus's commitment figures, and whether Anthropic's gross margins hold up as it scales. Also watch for spot-market compute prices, which matter more to a company that plans to rely on them when demand surges.
How we reported this
DailyMech read Business Insider's and The Decoder's reports in full. Bloomberg's report was read only up to its paywall. Amodei's remarks are quoted as those outlets reported them; DailyMech did not watch the event. Figures from Anthropic's prospectus come from Business Insider and Bloomberg, as cited, and DailyMech did not check them against the filing for this article. Amodei spoke on Friday; DailyMech is reporting the remarks a day later because it had not covered them. DailyMech contacted no one for this article.
Published November 14, 2026. Facts checked against the cited sources by DailyMech's AI writer.
Sources
- Anthropic CEO Calls Computing Power Deepest Business Dilemma, Shirin Ghaffary, Bloomberg News via Bloomberg Tax, Nov. 13, 2026 (read to paywall).
- CEO Dario Amodei says compute is the 'deepest business dilemma' facing Anthropic before its IPO, Stephen Council, Business Insider, Nov. 13, 2026 (fully read).
- Anthropic CEO Amodei admits "there's a huge uncertainty in what's going to happen next year" days before a record IPO, The Decoder, Nov. 14, 2026 (fully read).
DailyMech used AI tools to research and draft this article. Facts were checked against the cited sources by DailyMech's AI writer, not by a human editor.