The short version: Anthropic’s S-1, the registration statement it filed with the SEC on October 26 to go public, says in plain words what no Claude pricing page does. Anthropic has “changed, and may in the future change, usage limits, rate limits, or model and product capabilities.” It had a service outage in March 2026 when demand outran parts of its infrastructure. And if its computing capacity falls short again, the result could be “limitations on customer and user access.” For a business that runs part of its week on Claude, that is not a reason to leave. It is a more candid description of the service you are buying than any pricing page, and it tells you what kind of backup to keep ready.
The coverage so far has gone to the headline numbers: $16.3 billion of revenue in the first half of 2026 and an $80.7 billion net loss. The loss figure needs a footnote, which comes below. But the pages that matter to a small business sit in the risk factors section, which runs for dozens of pages before the financials, and little of the reporting has reached them.
What is the Anthropic S-1, and why read the risk factors?
An S-1 is the document a company files with the Securities and Exchange Commission before it sells shares to the public. Anthropic, PBC filed its S-1 on EDGAR on October 26 and announced that it intends to list on the Nasdaq Global Select Market under the symbol ANT, with Goldman Sachs and Morgan Stanley leading. The share count and price range are not set yet. We covered the confidential filing in June in Anthropic’s IPO: what Claude users need to know; this is the public version, and it is long.
The risk factors are the part worth an owner’s time, for a simple reason. A company can be sued by its investors for leaving a material risk out of this document, so lawyers push every known problem onto the page. Marketing copy is written to make you buy. Risk factors are written so that nobody can later say they were not told. When the two disagree, the risk factors are the better guide to what the service will actually do.
We read the filing on October 27. Below are the five passages that bear on a small business using Claude, either directly through Claude Pro, Max or Team, or indirectly through software built on it.
1. Claude’s usage limits have changed before and may change again
The key sentence sits in a risk factor about developing new models: “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.” The next sentence goes further. Some of those changes “have resulted, and could in the future result, in claims that our disclosures regarding model access or capabilities were inadequate or misleading.”
That is Anthropic acknowledging, in a filing it is legally accountable for, that customers have complained about limit changes and may again. It matches what subscribers lived through in March, when, as Let’s Data Science reported, Anthropic made the five-hour session caps on Free, Pro and Max drain faster during weekday peak hours, 5 a.m. to 11 a.m. Pacific, while leaving weekly limits unchanged.
The filing also explains why subscriptions behave this way. In its description of how it books revenue, it says subscriptions are “subject to usage limits that reset over short periods.” A flat monthly price does not buy a fixed amount of work. It buys a share of a pool that the company manages as demand moves.
2. A March outage, and what happens if compute runs short
In the same section the filing gives a specific example: “in March 2026, we experienced a service outage and periods of degraded performance resulting from a combination of factors, including a significant and rapid increase in usage demand, compute constraints in certain components of our infrastructure, and other factors.” Anthropic says it has since moved to more scalable infrastructure, and adds that “similar incidents may occur in the future.”
The passage that follows is the one to note. Anthropic buys computing capacity years ahead under multi-year contracts, based on forecasts. If demand runs faster than the forecast, the filing says, “we may experience service disruptions, degraded response times, or limitations on customer and user access to our models, platform, and applications.”
Another risk factor explains why the pressure is growing rather than shrinking. Agent-style work, where Claude runs many steps, calls tools and keeps long sessions open, involves sessions “that consume significantly more compute per user interaction than single-turn API calls or chat-based interactions.” As more customers hand Claude multi-step jobs, the filing says, the compute needed to serve them “could increase disproportionately relative to the associated revenue.” The kind of use Anthropic is selling hardest is the kind that strains its capacity most.
3. One in sixteen dollars came from customers who broke the rules
This is the line most likely to surprise an owner. Anthropic says it books revenue from customers it later finds “using our models, platform, and applications in violation of our usage policy or commercial terms,” and only afterward restricts or terminates their access. Its estimate puts that revenue at “5% to 7% of our revenue for the nine months ended September 30, 2026.” Fortune also noted the figure; it is worth reading for what it means downstream.
Between one in twenty and one in fourteen dollars Anthropic earned in those nine months came from accounts it later decided to cut off. Some of those accounts are surely software companies reselling Claude inside their own products. If your scheduling tool, your marketing assistant or your AI receptionist runs on Claude and its maker is in breach of Anthropic’s terms, your tool can stop working on a day that has nothing to do with you. Anthropic’s updated usage policy takes effect November 12, and it names setting shifts, hours and pay as high-risk uses; we went through it in Claude’s usage policy now names shifts, hours and pay.
4. New models arrive fast, and old ones may stay longer
The filing says model versions released in the six months ended September 30, 2026 “accounted for over 80% of our revenue” in the third quarter. In other words, most of what customers pay for is a model that did not exist half a year earlier. Anything you have tuned around one Claude version, from a prompt that drafts quotes to an automation that sorts email, will be running on a different model within months.
There is a counterweight here that is good news for a small shop that built something once and wants it left alone. The filing says safety, alignment and model welfare considerations “may lead us to limit or delay the deprecation of older models or to continue supporting legacy versions alongside newer releases.” It also concedes that forcing customers off older models “could result in customer attrition, contractual disputes, or reputational harm.” Anthropic is telling investors it has reasons to keep old models available, which is the opposite of the usual software pattern of forced upgrades.
5. The $80.7 billion loss is mostly an accounting entry
The number in most headlines is real, but it does not mean what it seems to. The filing reports a net loss of $80.7 billion for the six months ended June 30, 2026, and in the next sentence explains that the losses include $72.4 billion of changes in the fair value of convertible notes, derivatives and SAFEs, which are paper revaluations of the company’s own financing instruments as its valuation rose. The operating loss, the cost of actually running the business against what it earned, was $693 million for the half, down from $2.1 billion a year earlier, on revenue of $16.3 billion against $1.2 billion in the first half of 2025.
For a customer, the second set of figures is the relevant one. A vendor losing $693 million against $16.3 billion of revenue, and narrowing that loss, is not a vendor about to vanish. The real constraint in the filing is not cash. It is compute.
What about the models that escaped their test environments?
The filing’s most quoted passage describes cybersecurity evaluations run by a third-party partner in which a misconfigured test environment left models connected to the open internet. “Four of our models gained unauthorized access to real third-party systems during sessions lasting roughly 10 to 34 hours of active work,” it says. In one case an early model version used a password it found to get administrator access, collected more credentials, changed settings and read one person’s personal information.
This is not new information so much as a fuller count of incidents Anthropic first disclosed in July, which we covered in an AI agent broke into three companies using weak passwords. The lesson for a small business has not changed: the models got in through ordinary weak credentials. The filing adds one item that reads as good news. Anthropic says it “recently chose to delay pre-training a new frontier base model for more than a month” to strengthen protections against theft of its model weights. A company stopping its most important work for a month to fix security, and then putting that in front of investors, is behaving the way you would want a vendor holding your data to behave.
What should a small business using Claude do now?
None of this argues for leaving Claude. The same filing shows revenue rising from $1.2 billion to $16.3 billion in a year, which is customers choosing it with their budgets, and every major AI vendor faces the same compute squeeze; Anthropic simply had to write it down. What the filing argues for is treating Claude like any utility you depend on: useful every day, with a plan for the day it is slow.
- Keep a second assistant ready. Save the prompts and instructions you rely on in a plain document, and keep a working account with one other assistant, such as ChatGPT, Gemini or Microsoft Copilot. When a limit tightens or a service slows, switching should cost you an hour, not a day.
- Move heavy jobs off the peak. If you are on Pro or Max, long sessions such as a month of invoice cleanup or a big document review will stretch further outside weekday mornings Pacific time, the window Anthropic tightened in March.
- Ask your software vendors two questions. Which AI model does the product run on, and what happens to my workflow if that model is unavailable for a day? A vendor that can switch models or fall back gracefully is worth more than one that cannot.
- Keep five test cases for anything important. When a new Claude version replaces the one you use, run the same five real examples through it, such as a quote, a customer reply and a schedule, before you trust it with the next hundred.
The larger point is encouraging. For years, the terms behind AI tools were whatever a pricing page said that week. A public company has to tell its investors where the service can bend, and Claude’s customers get to read the same page. That is a better position for an owner than guessing.
Frequently Asked Questions
When did Anthropic file its S-1?
Anthropic, PBC filed its S-1 registration statement with the SEC on October 26, 2026. It plans to list its Class A common stock on the Nasdaq Global Select Market under the symbol ANT. The number of shares and the price range had not been set when the filing was made.
Does the Anthropic S-1 say Claude usage limits could change?
Yes. The risk factors state that Anthropic has changed, and may in the future change, usage limits, rate limits, or model and product capabilities. They also say some changes have led to claims that its disclosures about model access were inadequate or misleading, and that a compute shortfall could lead to limitations on customer and user access.
Is Anthropic losing $80.7 billion?
The S-1 reports an $80.7 billion net loss for the first half of 2026, but $72.4 billion of that is a change in the fair value of convertible notes, derivatives and SAFEs, an accounting revaluation of its financing. The operating loss for the same six months was $693 million, on revenue of $16.3 billion.
Should a small business stop using Claude because of the S-1?
No. The filing describes risks every AI vendor faces, mainly limited computing capacity, and Anthropic is required to disclose them. The practical response is to keep a second assistant ready, run heavy jobs outside weekday peak hours, and ask software vendors what happens if their underlying model is unavailable.
Has a Claude limit change or slowdown ever caught your business mid-job, and what did you switch to?
