UK — The Register

Anthropic's compute tab could top $1T as Q3 revenue nears $17B

AINovember 3, 20261 min read

Anthropic, the AI company, filed a public document saying its future payments for computing power could top $1 trillion. Its revenue is growing fast, but the growth is slowing down.

What happened

Anthropic published a regulatory filing ahead of a planned stock market listing (an IPO). The filing shows about $895 billion in promises to buy computing power and infrastructure. With other estimated payments, the total could go above $1 trillion. The company also warned that its revenue may not keep growing at this speed.

Key numbers

Revenue for the nine months ended September 30, 2026: $33.2 billion, up 1,206% from a year earlier.
Q3 2026 revenue: nearly $17 billion, up 46% from Q2.
Commitments: about $895 billion in total, with about $320 billion in existing contracts and $451 billion in leases that have not started yet.
Only about $213 billion of the $895 billion lets Anthropic avoid payments.
Deals with Microsoft, Google, Amazon, and Broadcom total $437 billion and cannot be cancelled.

Why the spending is so large

AI models need huge numbers of chips and data centers to run. Companies like Anthropic sign long deals to secure this computing power. These deals are like promises to pay for years, even if the company does not use all of the power.

Why it matters

If revenue does not grow faster than these costs, Anthropic could get into a bad cycle. It might need to borrow money or sell more shares. Selling more shares can lower the value for current owners. Investors will watch closely to see if the company can keep its money coming in.

What to watch

Whether Anthropic's revenue keeps growing as fast as its compute costs.
How the company pays for its huge commitments, such as debt or new shares.
Details about its customers in the updated filing.
How the stock market listing goes.

Glossary

IPO: Initial public offering: when a private company first sells shares to the public on a stock market.
Anthropic: An AI company that builds AI models and sells access to them.
Commitments: Promises to pay money in the future, often under contracts.
Broadcom: A company that makes computer chips and networking parts.

ITIS and NTL Healthcare Team Up to Pursue AI-Based Women's Cancer Diagnostics Market

ITIS signed an MOU with NTL Healthcare to jointly develop AI-based women's cancer diagnostic solutions and expand into overseas markets. The partnership combines ITIS's medical AI platform with NTL Healthcare's cervical cancer diagnostic expertise, and the companies plan joint R&D, precision medicine solution validation, and commercialization.

AIITmediaNov 3

Anthropic Amends IPO Filing: Q3 Revenue Near $17 Billion, Infrastructure Commitments About $895.4 Billion

Anthropic filed an amended registration statement (Form S-1/A) with the U.S. SEC on November 2 for its planned IPO, adding third-quarter 2026 results and future spending commitments for AI compute and infrastructure. Revenue for July–September was $16.956 billion, up about 47% from the previous quarter, but the company posted an operating loss of $567 million. The total infrastructure commitments of about $895.4 billion include roughly $213.8 billion that can be canceled under pre-start conditions, so the full amount is not unconditional debt.

AITechCrunchNov 3

Trajectory raises $45M to make AI model training accessible to nontechnical experts

Trajectory, a startup founded about a year ago by three Stanford classmates, raised $45 million in a Series A round at a $300 million valuation, led by Sequoia. The company's platform lets businesses use feedback to continuously retrain open source AI models for specialized tasks, an approach it calls continual learning, and it is now publicly available. Customers include Harvey, Rogo, Decagon, Mercor, and Nvidia.

AIThe RegisterNov 3

AI could break legal systems by finding previously hidden loopholes

Stanford Law School visiting fellow Felix Ringe and collaborator Linus Folkerts argue that large language models have become capable enough to find exploitable inconsistencies in written law at scale, much as they can find vulnerabilities in code. They cite several potential gaps in German law they uncovered with LLM help, while noting they are sharing only exploits they believe could not cause serious harm. The researchers warn that cheap automated legal reasoning could jeopardize the functioning of legal systems, and suggest lawmakers may need to harden legislation, possibly by using advanced AI themselves.