On October 26, Anthropic, the company behind Claude, filed to go public on Nasdaq under the ticker ANT. The filing runs more than 300 pages, and the risk factors alone take up close to a hundred of them. I read it so you do not have to. Here are the six risks that matter most, and why every executive buying AI should care whether or not you ever buy a share.
The headline numbers
Revenue went from $41 million in 2023 to $4.6 billion in 2025, then $16.3 billion in just the first half of 2026. The company holds about $131.7 billion in cash and short term investments. More than 11,900 customers spend $100,000 or more a year with it. Those are remarkable numbers. The risks are what sit underneath them.
1. Compute is the business, and it is expensive
In the first half of 2026, Anthropic spent $8.9 billion on compute and infrastructure against $16.3 billion of revenue. It has signed commitments for chips and data centers that run as far out as 2045, and that spending continues whether demand holds up or not. The company also admits it has asked its cloud partners for more capacity than they were willing to give.
2. Its suppliers are also its competitors
Amazon and Google provide chips and cloud capacity, sell Claude to their own customers and build rival models, and Amazon also holds an investment in Anthropic. The filing says plainly that these relationships could create conflicts of interest. Revenue is also concentrated in a limited number of customers and distribution channels, so losing one large account or channel would show up fast.
3. This growth cannot last at this pace
Net revenue retention is about 900%, meaning existing customers now spend roughly nine times what they did a year ago. No company keeps that up, and Anthropic itself says past growth is not a guide to the future. A stock priced on hypergrowth falls hard when growth turns merely strong.
4. Staying on top is never finished
The whole business depends on having one of the best models at all times. Training runs cost many millions of dollars, take weeks or months, and sometimes fail. Anthropic also holds itself to safety rules its rivals may not, and it has delayed releases, such as its Mythos model, to stay within them. The filing admits those choices could slow it against competitors, and it flags that using Claude to build the next Claude means a hidden flaw could carry forward from one version to the next.
5. Legal and regulatory exposure is already real
Anthropic paid a $1.5 billion settlement in 2025 in the Bartz copyright case over training data, one of the cases on our AI lawsuit tracker. It warns its contracts and insurance may not cover harms caused by its models, including models acting in ways no user directed. Regulation is not hypothetical either: in June 2026 it suspended access to two of its models for nearly three weeks to comply with U.S. export controls.
A trust elects most of the board. The seven cofounders, through one special share, hold 50.1% of the votes. As a public benefit corporation, the board must weigh the mission alongside shareholder returns. Public investors will own a slice of the economics with almost no say over strategy, management or a sale.
The number that looks worse than it is
The filing shows a net loss of $80.7 billion for the first half of 2026. Most of that is an accounting charge from convertible notes rising in value as the company's valuation climbed. On the company's adjusted measure, the same six months show $1.6 billion of operating income. The operating picture is close to breakeven, not an $80 billion hole.
What this means if you lead AI adoption
You do not need to buy the stock for these risks to touch you. If your workflows run on Claude, or on any single frontier model, the same dependencies reach you: compute shortages turn into rate limits, export rules turn into access gaps, and a vendor's safety choices can change what a model will do. Treat your AI provider like any critical supplier. Know your fallback model, test it, and make sure your contracts say what happens when access changes.
The leaders who build that discipline now will not be scrambling when the market shifts. That is governance, not paperwork, and it is the difference between adopting AI and depending on it.
Further reading on The World of AI
This is a summary of a public filing, not investment advice. Read the full S-1 at sec.gov before making any investment decision.
