Anthropic Spent Under a Third of Its Compute Serving Customers, Filing Shows

Oct 29, 2026
Anthropic spent less than a third of its $8.9 billion first half compute bill on serving customers, while most went to research, training and internal use. Its compute agreements could total about $800 billion.

Anthropic spent less than a third of its $8.9 billion first half compute and infrastructure bill on serving customers, according to an analysis by Tom's Hardware of its IPO filing. The company spent $2.4 billion serving customers while generating $16.3 billion in revenue, with at least 66% of compute going to research, training and internal use.

The analysis estimates that Anthropic's TPU and GPU agreements could total about $800 billion. The commitments cover capacity from Broadcom, Google, Amazon, SpaceXAI, Microsoft, Nscale, Lambda and AMD, with delivery schedules extending into the 2030s.

The five largest agreements would cost an estimated average of $103 billion a year once delivered. The filing warns that margins could narrow when new hardware arrives before it supports additional revenue.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Silicon Brief, AI Funding Brief or Daily AI Brief.

Also, consider following us on social media:

AI Chips & Datacenters AI Funding News AI Brief AI Brief (X)
Free newsletter

AI Funding Brief

Guide
AI app builders

AI app builders

Tools that turn a description into a working web or mobile app, and the platforms companies use to build internal tools with AI

AI Token Webinar

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.