Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
IndustryInsight · On RadarThe Decoder→ original

Ahead of IPO, Amodei Admits Anthropic Committed to Spending Nearly $901 Billion on Compute

A few days before the IPO, Anthropic CEO Dario Amodei acknowledged that buying compute, meaning servers and chips, is the hardest part of the business. Capacity is ordered one to two years in advance, and the company's commitments total nearly $901 billion, of which about $71 billion is due by September 2027. It's like ordering ovens for a bakery without knowing demand: order too few and you lose customers, order too many and you drown in debt. The IPO is meant to serve as a cushion in case revenue falls behind spending.

AI-processed from The Decoder; edited by Hamidun News
Ahead of IPO, Amodei Admits Anthropic Committed to Spending Nearly $901 Billion on Compute
Source: The Decoder. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Decoder ↗
Editor-in-Chief
IPO on November 19up to $64.4 billion on the exchange and another $26 billion in a private placement
Key metric #2Orders for shares are roughly six times the amount on offer
Key metric #3Annualized revenue of $81 billion, $10 billion higher than in September

Picture a bakery that has to order and pay for its ovens two years in advance, without knowing how much bread people will buy. Order too few and customers go to the neighbors. Order too many and the ovens sit idle while the loans still have to be repaid. Anthropic finds itself in a similar situation, except its ovens are servers and chips, and its obligations total nearly $901 billion. CEO Dario Amodei acknowledged this a few days before the company's IPO, scheduled for November 19, which is its first public stock sale.

At a JPMorgan Chase investor conference in New York, he called buying compute "the deepest business dilemma" facing Anthropic and added: "There is enormous uncertainty about what will happen next year."

Why buying compute has become a dilemma

Compute is the capacity used to train and run models such as Claude. According to Amodei, it has to be reserved one to two years ahead, before it is clear whether it will pay off. The technology is moving so fast that no one knows exactly how progress will turn into money.

If the company buys too little, it won't have enough resources for customers and new development, and revenue will suffer. If it buys too much, the company may, in Amodei's words, "overextend itself financially."

The numbers are large. According to Bloomberg, Anthropic's obligations have reached nearly $901 billion, mostly for compute and chips. About $71 billion must be paid by September 2027. Bloomberg notes that this sum exceeds the market capitalization of roughly two-thirds of the companies in the S&P 500, which are the largest publicly traded U.S. companies.

Amodei warned about this earlier. In February 2026 he spoke far more bluntly: one could buy $1 trillion worth of compute starting in late 2027 and go bankrupt if revenue fell short. An error of even one year could prove fatal. He is not sure that some competitors considered such scenarios at all.

How Anthropic is cushioning the risk

Amodei named two shock absorbers: gross margin and cash on the balance sheet. He said the margin is "fairly good," especially compared with other players. In addition, the company buys relatively little capacity in advance and, if the business develops well, buys more on the spot market, meaning at current prices and without long contracts. He likened this to insurance: if results come in below expectations, the company is covered, and if they come in above, it can scale up investment.

The IPO plays the role of a cushion in this scheme. The more money Anthropic has, the wider the margin of safety between the scenarios of "too little compute" and "too much compute." According to Bloomberg, the company plans to raise up to $64.4 billion through the IPO and another $26 billion through a private placement. Together, this would be the largest IPO in history. The same report says orders for shares already total about six times the number of shares on offer.

Revenue is still growing fast. Annualized revenue (run rate: current sales extrapolated to a full year) reached $81 billion at the end of October, $10 billion more than at the end of September, according to company documents. Bloomberg links the growth to demand for AI products, including coding models.

What this means for developers and product managers

The first takeaway: prices, rate limits and API speed depend on capacity purchases made years before a release. If you build products on Anthropic's models, it matters how accurately the company has predicted demand. It is therefore sensible not to rely on a single provider and to keep a fallback option.

The second takeaway: after the IPO, Anthropic will face public reporting and investor pressure. Margins, compute spending and growth rates will become visible, which will give customers more data to judge the provider's stability.

There are caveats. The figures on obligations, offering size and oversubscription come from Bloomberg, not from an official prospectus. The IPO has not yet taken place, and its terms could change. Amodei did not back up his claim of a "fairly good" margin with specific numbers.

He himself stays calm: he said he is not tied to a particular month or market conditions, and that the company grew from zero to its current size in six years. The best advice before an IPO, he said, sounded roughly like this: "Don't let the company be worth a dollar more than it really is." This echoes Anthropic's internal principle of "promise less, do more." In the same conversation he expressed hope that AI would cure cancer and Alzheimer's within 5 to 15 years, adding that he does not know for certain.

Story

Key Questions Answered

Why buying compute has become a dilemma?

Compute is the capacity used to train and run models such as Claude. According to Amodei, it has to be reserved one to two years ahead, before it is clear whether it will pay off. The technology is moving so fast that no one knows exactly how progress will turn into money. If the company buys too little, it won't have enough resources for customers and new development, and revenue will suffer. If it buys too much, the company may, in Amodei's words, "overextend itself financially." The numbers are large.

How Anthropic is cushioning the risk?

Amodei named two shock absorbers: gross margin and cash on the balance sheet. He said the margin is "fairly good," especially compared with other players. In addition, the company buys relatively little capacity in advance and, if the business develops well, buys more on the spot market, meaning at current prices and without long contracts. He likened this to insurance: if results come in below expectations, the company is covered, and if they come in above, it can scale up investment.

What this means for developers and product managers?

The first takeaway: prices, rate limits and API speed depend on capacity purchases made years before a release. If you build products on Anthropic's models, it matters how accurately the company has predicted demand. It is therefore sensible not to rely on a single provider and to keep a fallback option. The second takeaway: after the IPO, Anthropic will face public reporting and investor pressure.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…