Anthropic's S-1: five things that change for UK firms paying for Claude

What changed

On 26 October 2026 Anthropic, PBC filed a preliminary Form S-1 with the US Securities and Exchange Commission, applying to list Class A shares on the Nasdaq Global Select Market under the symbol ANT. The price range, the number of shares and the offering size are blank, which is normal at this stage. What is not normal is the scale of the disclosure. Revenue was $16.3bn for the six months to 30 June 2026, against $4.6bn for the whole of 2025. The second quarter of 2026 produced $974m of GAAP operating income on $11.6bn of revenue, the first quarter in the company's history to do so, although the six months as a whole were still a $693m operating loss. The net loss for the half was $80.7bn, of which $72.4bn is the accounting remeasurement of convertible notes, derivatives and SAFEs as the company's valuation rose. Six-month and quarterly figures are unaudited.

We have put the full set of figures, with the printed page each came from, on our Anthropic IPO tracker. This briefing is about what the filing changes for a UK business that pays for Claude. It is not about whether anyone should buy the shares, and The AI Consultancy, which is Certified in the Claude Partner Network and takes no vendor income (see our independence page), is not authorised to offer that view.

Why it matters for UK business

Five things in the filing change how a buyer should think.

1. Consumption is the business, so your usage is the vendor's revenue. About 80% of first-half revenue was consumption-based: token and usage charges on the Claude Platform and through AWS, Google Cloud and Microsoft Azure. Subscriptions to Pro, Max, Team and Enterprise were about 20%. The filing also describes a growing use of annual and multi-year non-cancellable minimum-spend contracts. For a buyer, every design decision that affects prompt length, retrieval volume and the number of agent steps is a financial decision. Quote implementation as a fixed scope. Put model usage in a separate line with a budget, alert limits and a defined trigger for a scope conversation. Do not let a proposal blend the two.

2. The headline numbers need three different readings. A positive adjusted operating result is not GAAP net profit: the adjusted figure adds back $2.2bn of stock-based compensation and $126m of non-cash charitable contributions, both of which dilute shareholders. The $80.7bn net loss is not cash burn: operating cash flow for the half was a $3.9bn inflow. The 900% net revenue retention figure compares a prior-year cohort's spend with its spend a year earlier, excludes Pro and Max, and will fall as the base grows; it is not a measure of how many customers stayed. And the compute arrangements cannot be added into one total, because they sit on different bases, from a Microsoft minimum of at least $44.9bn to a SpaceX.AI arrangement of up to $120.5bn that either side can cancel on 90 days' notice after early 2027. If a supplier, a board paper or a journalist quotes a single total, ask which rows were added and on what basis.

3. Capacity, not cash, is the dependency. The company held $112.4bn of cash and $19.3bn of short-term investments at 30 June 2026, almost all of it financed. A supplier can have abundant cash and still face operational disruption from equipment delivery, data-centre readiness, partner disputes or conditional financing, and the filing's own risk factors say so. The filing also records that June 2026 export restrictions affected access to the Fable 5 and Mythos 5 models worldwide until lifted, and that litigation over government access produced a favourable district court ruling in August and a split appeals decision in September. For a UK deployment, the protections that matter are at the application layer: a second configuration that has actually been tested against your quality bar and your privacy requirements, documented migration steps, and data you can take with you. A fallback nobody has run is not a continuity plan.

4. The vendor is moving into the work consultancies sell. The filing describes Anthropic's own applications (Claude Code, Cowork and Design), an Applied AI function that embeds engineers in customer deployments, a partner network, a $100m Claude Frontier Academy that aims to train 10,000 engineers by the end of 2027, and Ode with Anthropic, an AI services company launched with private equity partners to serve mid-sized companies. We covered the consulting ventures in last week's briefing. The S-1 confirms the direction and adds the scale. For a buyer this is mostly good news: more capable delivery capacity is coming to the market. The question to ask any Claude adviser, including us, is what they bring that the vendor's own channel does not. The honest answers are sector knowledge, accountability for an outcome, continuity of service, and a willingness to say when Claude is the wrong tool.

5. Model eligibility is a renewal item. The filing is explicit that safety decisions can delay releases, constrain use or interrupt revenue, and that access to the most capable models can be gated. Mythos access runs through the Project Glasswing programme; Fable has a different access path and safeguards. Geography, use case and the specific model all matter. Write the model and its access terms into your deployment record, and check them when the contract renews rather than discovering a change in production.

What to do, and what not to do

Do:

• Separate implementation fees from usage in every proposal you receive, with volume assumptions, a usage budget and alert limits written down.
• Pick one workflow that would hurt if Claude were unavailable for a week, and test a second configuration against it, recording the quality gap and the migration steps.
• Record the approved model, settings, data handling and access terms for each deployment, and review them at renewal.
• Keep a baseline and a post-launch measurement for each pilot, so that the business case rests on your results rather than on the vendor's case studies.

Do not:

• Quote a single compute commitment total, a gross margin or a cash runway from this filing. The filing does not support any of them.
• Treat the 900% figure as customer retention, or the adjusted result as profit.
• Promise your own stakeholders uninterrupted access to a specific model on the strength of the vendor's cash balance.
• Read anything here as a view on the shares. It is not one.

Where The AI Consultancy fits

The full figures, charts and a watchlist of what the amended filing needs to answer are on the Anthropic IPO tracker, which we update on each filing. Structuring usage budgets, testing fallbacks and setting deployment records is the core of our Claude consulting work; the seat and token prices in sterling are in what Claude costs in the UK, and the plan comparison is in the Claude Enterprise buyer's guide.

Verified on 28 October 2026 against Anthropic, PBC's Form S-1 (accession 0001628280-26-067168) as published on sec.gov. All currency is US dollars as filed; six-month and quarterly figures are unaudited. This briefing is business analysis for organisations that use or evaluate Claude. It is not investment research or advice, and The AI Consultancy is not authorised by the Financial Conduct Authority.