Daily Issue #3 · 7 min read

What Anthropic's IPO filing tells you if you build on Claude

Anthropic's S-1 is being read for its valuation. I read the risk factors instead, because they describe the risks every team building on Claude already carries.

The short answer

What does Anthropic's S-1 mean for teams that build on Claude? On October 26, Anthropic filed to list on Nasdaq under the ticker ANT, and the prospectus it filed with the SEC gives a detailed look at how a frontier AI lab runs. For builders, the useful parts are in the risk factors. Anthropic says in writing that it has changed rate limits and usage limits before and may again, that forced model deprecations are a real tension, that it buys compute 12 to 18 months ahead and pays more when demand runs past it, and that it delayed training a new frontier model for more than a month over security. I would keep building on Claude. I would also design as if any model provider can change its terms, limits, and timelines on short notice.

Why I picked this story

Most coverage of the filing is about valuation, the size of the loss, and the share structure. That matters to investors. I read it from a different seat: as someone who would ship features on top of these models and then has to keep them running.

A prospectus is unusual because lawyers make the company list everything that could go wrong. Product blog posts never do that. So for once we get the vendor's own list of what could go wrong, in plain sentences, with legal liability attached. Many of those items land on customers too.

The numbers, briefly

Here is what the filing reports, in Anthropic's own figures:

• Revenue: $16.3 billion in the first half of 2026, up from $1.2 billion in the first half of 2025. Full year 2025 revenue was about $4.6 billion.
• A profitable quarter: the quarterly table shows operating income of $974 million for the three months to June 30, 2026, after an operating loss of about $1.7 billion in the quarter before.
• The headline loss: a net loss of $80.7 billion for the half year, of which $72.4 billion is a change in the fair value of convertible notes and similar instruments. The operating loss for the half was $693 million.
• How it earns: consumption based revenue, mostly API and usage billing, was $13.1 billion or 80% of the total. Subscriptions were $3.2 billion or 20%.
• Who pays: more than 11,900 customers paid $100,000 or more over the trailing twelve months to September 30, up from more than 950 a year earlier. About 250 paid $10 million or more.

The growth is striking. What I find more useful is how much of this business is metered usage from companies wiring Claude into their own products. That is us. And the risk section reads differently once you know you are the revenue line.

Five passages I would underline as a builder

1. Rate limits and model behavior can change

This is the sentence I would pin above every AI roadmap: "We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities, which could adversely affect our reputation and the perceived value of our offerings."

Every provider does this. It is still rare to see one say it so plainly. If your feature only works at today's limits and today's model behavior, you have a dependency you have not written down.

2. Deprecation cuts both ways

Anthropic writes that safety and welfare considerations may lead it to "limit or delay the deprecation of older models," and also that "customers who have integrated older models into their workflows may resist migration to newer versions."

I read this as a warning about my own habits. Most teams pin a model, tune prompts against it, and only think about migration when a retirement email arrives. A prompt that scores well on one model version can behave differently on the next. The fix is boring: keep an evaluation set for each AI feature so that moving models is a test run, not a rewrite.

3. Compute is bought a year or more ahead

The filing says most of Anthropic's compute is contracted under long commitments it "typically" procures "12 to 18 months in advance of availability." When demand runs past that, it buys short term compute, which "typically carries a higher cost per unit." It expects to use more short term compute in the near term. As of June 30, it had about $265.6 billion in fixed or minimum compute commitments, with terms of one to 20 years.

For a customer, I think this explains a lot of past friction. Capacity is planned long before demand shows up. When a new model is popular, there is a gap between what people want and what has been racked. That gap shows up as tight limits, slow responses at peak, or price and plan changes. I would plan for those moments rather than be surprised by them.

4. Half the revenue flows through the clouds

Revenue from customers buying through cloud marketplaces was about 48% of the total in the first half of 2026, according to the filing. Anthropic also notes that it relies on a small number of these cloud providers for compute, and that some of them invest in it or are its customers, which it says could create conflicts of interest.

The practical point for builders is that the same model is often reachable through more than one door: the direct API and one or more cloud platforms. Those doors can differ in quotas, regions, feature timing, and billing. An ecommerce search team that can route a request to the same model through a second endpoint has a cheap form of resilience that many teams skip.

5. Security can move release dates

The filing says Anthropic "recently chose to delay pre-training a new frontier base model for more than a month to allow time to further enhance our security measures related to preventing model weight exfiltration and protecting our intellectual property."

I think this is the right call by the lab. It also means release dates for frontier models are less certain than launch hype suggests. If a product plan depends on "the next model" landing in a given quarter, it rests on a date that the vendor itself says can slip for safety or security reasons.

What I would not read into it

I would not treat the $80.7 billion net loss as a sign that Claude is going away. Most of it is an accounting charge tied to a rising valuation, and the operating numbers tell a different story. I would also not treat the risk factors as a list of things that will happen. Every prospectus lists worst cases.

The value is that these risks are now stated by the vendor, in one place, in language a legal team signed off on. Most of them apply to every frontier lab. Anthropic simply had to write them down first.

My takeaway

The filing shows a company whose revenue now comes mostly from other businesses building on its models, and which openly says that limits, model versions, capacity, and timelines can move. If you build on Claude, or on any frontier model, design your product so that a change in any one of those is an inconvenience rather than an outage.

Quick notes

• OpenAI launches GPT-6.1 Terra for computer use: OpenAI introduced GPT-6.1 Terra on October 26, a fast model aimed at computer use tasks like filling in forms and reading screens. OpenAI says its standard API prices are 50% lower than GPT-5.6 Terra and that at medium reasoning effort it performs on par with GPT-6 Astra on its Agents' Last Exam benchmark at one tenth the price per token. These are OpenAI's own figures.

Common questions

When did Anthropic file for its IPO?

Anthropic filed a registration statement on Form S-1 with the US Securities and Exchange Commission on October 26, 2026. It plans to list its Class A common stock on the Nasdaq Global Select Market under the ticker ANT. The number of shares and the price range had not been set at filing.

How much revenue does Anthropic make?

According to the filing, Anthropic reported $16.3 billion in revenue for the first half of 2026, up from $1.2 billion in the first half of 2025. Revenue for full year 2025 was about $4.6 billion. About 80% of first half 2026 revenue was consumption based, mostly usage billing.

Why did Anthropic report an $80.7 billion loss?

Most of the net loss for the first half of 2026, about $72.4 billion, came from a change in the fair value of convertible notes and similar instruments as the company's valuation rose. The operating loss for the same period was $693 million.

Does the S-1 change anything for developers using the Claude API?

Nothing changes overnight. The filing does confirm risks builders should plan for: Anthropic says it has changed usage limits and rate limits before and may again, that model deprecation creates tension for customers, and that compute capacity is planned 12 to 18 months ahead. Keeping an evaluation set and a fallback path makes these changes easier to absorb.