Anthropic CEO Dario Amodei used his last big public appearance before the company’s planned Nov. 19 IPO to talk about the biggest risk in his business. Speaking with JPMorgan CEO Jamie Dimon at a JPMorgan Chase investor event in New York on Friday, he said: “There’s a huge uncertainty in what’s going to happen next year.” According to The Decoder, he called compute “the deepest business dilemma that the company faces.”
CEOs don’t usually say things like that days before what could be the largest IPO ever. It’s still worth listening to, because the problem he described affects every frontier AI lab, not only Anthropic.
Key takeaways
- The core risk: Anthropic has to buy compute one to two years ahead, before anyone knows whether revenue will cover the cost.
- The scale: Bloomberg reports Anthropic has committed nearly $901 billion, mostly to compute and chips. About $71 billion of that comes due by September 2027.
- The cushion: The IPO aims to raise up to $64.4 billion, plus another $26 billion through a private placement. Together, that would make it the largest IPO in history.
- The demand: The offering is reportedly about six times oversubscribed.
- The growth: Company filings show Anthropic’s annualized revenue run rate hit $81 billion at the end of October, up $10 billion from late September.
Why compute is such a hard bet
Amodei laid out the trap plainly. AI is improving so fast that nobody knows yet how that progress will turn into economic value. Chips and data center capacity still have to be locked in long before that answer arrives.
If Anthropic buys too little, it can’t serve customers or ship new technology, and revenue suffers. If it buys too much, “you can financially overextend yourself,” Amodei said.
He’s said this before, and more bluntly. Back in February 2026, he warned that Anthropic could buy $1 trillion of compute starting at the end of 2027 and go bankrupt if revenue came in short. Being off by a single year could be enough. He also suggested some competitors hadn’t done that math and didn’t fully understand the risks they were taking.
The numbers have grown a lot since then. Bloomberg notes that the $71 billion due by September 2027 is more than the market cap of about two-thirds of S&P 500 companies.
How Anthropic is hedging
Amodei named two main buffers: gross margins and cash. He described the margins as “quite good,” especially next to other players, and said Anthropic has a large cash balance and wants more.
The company also uses a hybrid buying strategy:
- Commit conservatively. Buy relatively little compute upfront.
- Top up on the spot market. If business beats expectations, rent extra capacity as needed.
Amodei compared this to buying insurance. Anthropic stays covered if results disappoint, and it can still spend more when things go well.
That’s why the company is going public. More cash on the balance sheet, he said, gives Anthropic more room for error between having too little compute and having too much.
What this means for the industry
This is significant because it’s the clearest public statement yet of the economics underneath the AI boom. Revenue is climbing fast, and Bloomberg credits much of Anthropic’s growth to the popularity of its coding models. But the cost side is fixed years ahead, while the revenue side is a forecast.
For practitioners and enterprise buyers, here’s what to watch:
- Capacity and pricing. Labs that buy compute cautiously may ration access or adjust prices when demand spikes.
- Vendor stability. Huge compute commitments make a lab’s balance sheet part of your procurement risk.
- Competitive pressure. Amodei’s earlier jab suggests he thinks some rivals are overextended. If he’s right, the next downturn in AI demand will hit unevenly.
A careful pitch, mostly
Amodei sounded relaxed about the IPO’s timing and said he wasn’t fixated on a particular month or market conditions. He pointed out that Anthropic grew from nothing to its current size in six years.
The pre-IPO advice that stuck with him most was to “never let your company be worth a single dollar more than it’s really worth.” He said that matches a principle Anthropic uses internally: “under-promise and over-deliver.”
In the same conversation, he also said he hopes AI will cure cancer and Alzheimer’s within five to 15 years, while admitting he doesn’t know exactly. By Silicon Valley standards, that may still count as modest.
The real test starts after Nov. 19. Public markets will now be watching each quarter to see whether Anthropic’s revenue keeps pace with its compute bills. The Decoder has the full report.