On October 26, 2026, Anthropic — the company behind Claude — filed to sell shares on the stock market. Most coverage is about valuation. If your business uses Claude, directly or inside another tool, the more useful part is what the filing says could change for customers: prices, usage limits and which models stay available.
What did Anthropic file on October 26?
Anthropic filed a registration statement on Form S-1 with the US Securities and Exchange Commission on October 26, 2026, for a proposed initial public offering (IPO) of its Class A common stock. An S-1 is the long disclosure document a company must publish before selling shares to the public.
The basics, from Anthropic's announcement the same day:
- Exchange and ticker: Anthropic intends to list on the Nasdaq Global Select Market under the symbol "ANT."
- Size and price: The number of shares and the price range had not been set.
- Status: The registration statement had not yet become effective, so no shares can be sold yet.
This article is about what the filing means for businesses that use Claude. It is not investment advice.
Does anything change for your Claude account today?
No. An S-1 is a disclosure document; it does not change any product, price or contract. Your Claude plan, or the terms of any software that uses Claude behind the scenes, still governs what you get and what you pay.
What the filing does give you is something you rarely get from a software vendor: a detailed, legally required list of what the company itself says could change. Risk factors in an S-1 are written to be cautious, so they are not predictions. But they are a useful checklist for any business that has built daily work around one AI vendor — Anthropic, OpenAI, Google or anyone else.
How big is business use of Claude, according to the filing?
Large and growing fast. The figures below come from the S-1 as filed on October 26, 2026.
| Measure | Figure in the S-1 | Period |
|---|---|---|
| Revenue | $16.3 billion | Six months ended June 30, 2026 (up from $1.2 billion a year earlier) |
| Share of revenue from pay-per-use | 80% ($13.1 billion) | Six months ended June 30, 2026 |
| Customers bringing in $100,000+ over the past 12 months | More than 11,900 (up from more than 950) | September 30, 2025 to September 30, 2026 |
| Revenue through cloud marketplaces | About 48% | Six months ended June 30, 2026 |
"Pay-per-use" means customers pay for the text the models read and write, measured in tokens. "Cloud marketplaces" means customers buying Claude through the big cloud platforms — the S-1 names Amazon Bedrock, Google's Gemini Enterprise Agent Platform and Microsoft Foundry — rather than from Anthropic directly.
What does Anthropic say could change on pricing?
Anthropic says its pricing models may keep changing, and the filing points in both directions. The S-1 states: "We have in the past changed, and we expect that in the future we may change, our pricing models." It adds that the company "may be required to reduce our prices, offer shorter contract durations or offer alternative pricing models, such as outcome or task-based pricing."
The filing also flags that "agentic offerings" — AI that carries out multi-step tasks on its own — are "structurally more compute-intensive," meaning they cost more to run. It warns that if it cannot price those products to reflect that cost, its profitability could suffer.
For an owner, the practical reading is to budget for movement: prices, plan structures and contract lengths can all change, and the filing says so. Work done by AI agents may end up priced differently from chat — per task or per outcome rather than per token. If you have an annual commitment with any AI vendor, the renewal is where a change will show up first, so put the date in your calendar.
What about usage limits and model changes?
Anthropic is explicit that limits can move. The filing says: "We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities." A rate limit is a cap on how much you can use in a set period.
Older models are also retired on a schedule. Anthropic's documentation (as of October 29, 2026) commits to at least 60 days' notice before retiring a publicly released model. One retirement is close:
- Claude Sonnet 4.5 was deprecated on September 30, 2026 and is due to be retired on November 30, 2026. Anthropic's recommended replacement is Claude Sonnet 5.5.
- Requests to a retired model fail, so any tool still built on it will stop working until it is updated.
- These dates apply to Anthropic's own platforms; Amazon Bedrock and Google-managed Claude set their own schedules.
Should a small business worry about relying on one AI vendor?
Worry is too strong; planning is not. AI vendors change prices, limits and models often — OpenAI replaced its mid-priced GPT-5.6 Terra model with a new one at a new price on the same day this filing went in — and the S-1 simply puts Anthropic's version in writing. The risk for a small business is not the IPO. It is having a workflow that only one person understands, running on one model nobody is watching.
Four habits reduce that risk:
- Keep your instructions in your own files. The prompts, rules and examples that make a workflow work should live in your documents, not only inside a vendor's chat history.
- Know which model sits under each tool. Software with AI features does not always say whose model it runs on; ask.
- Test before deadlines. When a retirement is announced, run your real tasks on the replacement weeks before the cut-off.
- Own what matters. For core operations, software you own — rather than rent — keeps your data and logic portable. That is the idea behind Systems.
What should you do this week?
Five checks, about an hour in total:
- List every AI tool your team uses and the vendor behind each.
- Ask each vendor which model they run on, and whether anything uses Claude Sonnet 4.5.
- Note your current plan, usage limits and renewal dates in one place.
- Save the instructions and templates behind any AI workflow to a shared folder.
- Decide who on your team watches for vendor change notices.
If running models, retirements and limit changes sounds like a job you do not want, that is the case for handing the workflow to someone who watches it for you. We compare the options in managed AI vs DIY AI, and explain what an AI operator is: one workflow, run and watched for you.
For a step-by-step way to choose and protect your first AI workflow, start with the AI handoff playbook — or book a consultation and we will look at your setup with you.