is it safe to build your business on claude? what anthropic's own s-1 says
if your business runs on claude, the most useful document published this month is anthropic's own ipo filing. not for the valuation. for what it tells customers to expect.
yes, with your eyes open. anthropic reported about $131.9 billion in cash and short-term investments and $17.0 billion of revenue in a single quarter, so it is not about to vanish. but its s-1/a says plainly that three things will keep changing: usage and rate limits, which models are available, and who gets cut off for breaking the rules. build so those changes cost you an afternoon, not a client.
will anthropic still be here?
the filing describes a company with deep cash and fast growth, carrying very large long-term bills. as of september 30, 2026: $95.6 billion in cash and cash equivalents, $36.3 billion in short-term investments, $14.8 billion of undrawn credit, against about $69.8 billion of compute commitments due in the next 12 months (anthropic s-1/a, nov 2 2026). total commitments are about $895.4 billion through 2045. the full breakdown is in anthropic's $895 billion compute bill. this is not investment advice; it is what the document says about the near term.
for a business owner, the more useful question is not "will it survive" but "will it keep working the way it works for me today." the filing answers that one too.
limits will change, and anthropic says so
the s-1/a states that anthropic has changed, and
"may in the future change, usage limits, rate limits, or model and product capabilities"
anthropic, s-1/a, november 2, 2026
the reason is in the numbers. the filing says demand is limited principally by the availability of compute, and that in q3 2026 it spent $5.0 billion on short-term compute that costs significantly more than its multi-year deals (anthropic s-1/a, nov 2 2026). it also describes subscriptions as subject to usage limits that reset over short periods. when compute is the bottleneck, limits are the valve. if a workflow only works on a good day, it does not work.
older models will be retired
the model you use today is not permanent, and anthropic treats deprecation as a known tradeoff. the s-1/a says customers who built older models into their workflows may resist migrating, that an accelerated or forced deprecation could cost it customers, and that it may sometimes keep legacy versions running alongside newer ones for safety and other reasons (anthropic s-1/a, nov 2 2026). either way, a retirement date is coming for every model you use. the businesses that feel it least are the ones that can switch in one place and have already tested the next model on real work.
rule-breaking usage is being cut off
anthropic estimates 5% to 7% of its revenue in the first nine months of 2026 came from customers violating its usage policy or commercial terms, and it is actively removing that revenue. it stepped up detection and enforcement in august and september 2026, acknowledges that restricted customers may re-register under new accounts, and expects the share to fall (anthropic s-1/a, nov 2 2026).
anthropic's usage policy says that if it suspects a violation, it may warn you or throttle, limit, suspend, or terminate access. the current version on its site takes effect november 12, 2026. a company willing to tell the sec it is giving up 5% to 7% of revenue is not going to make an exception for your workaround. if any part of how you deliver depends on something you have not checked against the policy and your plan's terms, check it now.
five things to set up this week
each one turns a change anthropic has warned about into a small edit instead of an emergency.
- keep your business context in files you own. your offers, voice, clients, and processes belong in documents you control and load into claude, not scattered across chat history. a new plan, model, or tool can read the same files. the system i use is in files not chats: the 6-file context system, and the memory question is in claude memory or files.
- name your model in one place. one setting, one skill file, one line. when a model is retired, you change it there, after running the next model on a real task first.
- know your limits and your heavy-day plan. write down which plan you are on, what happens when you hit the limit, and what you do then. if you are on max or team, the monthly api credits are one fallback, with rules worth knowing first: claude max vs team api credits.
- read the usage policy once, against how you actually use claude. not the whole legal center. just the policy, with your real workflows next to it.
- cap any single dependency at about a third. no tool, client, or platform above roughly a third of revenue or delivery. spacex just showed what one customer at 34.5% looks like from the inside: spacex's customer b, read from both filings.
the takeaway
claude is a reasonable foundation for a small business, and anthropic's filing is unusually candid about what will move under you. limits, models, and enforcement are not surprises anymore; they are disclosed risks. the businesses that do well on claude will not be the ones that guessed right about which model lasts. they will be the ones whose context, prompts, and processes live in files they own, so any change is an afternoon of edits.
common questions
Is it safe to build a business on Claude?
Anthropic's S-1/A, filed November 2, 2026, shows about $131.9 billion in cash and short-term investments and $17.0 billion of revenue in Q3 2026, so the company is not about to disappear. The same filing says usage limits, rate limits, and model capabilities may change, older models get retired, and usage that breaks its policy is being cut off. It is safe if you build so that those changes cost you an afternoon, not a client: keep your context in your own files, name your model in one place, know your limits, and stay inside the usage policy.
Will Claude's usage limits change?
Anthropic says they may. Its S-1/A states that it has changed, and may in the future change, usage limits, rate limits, or model and product capabilities. The same filing says demand is limited principally by the availability of compute, and that $5.0 billion of short-term compute in Q3 2026 cost significantly more than its multi-year compute, which is the pressure behind limits.
Will Anthropic retire the Claude model I use?
Eventually, yes. The S-1/A discusses model deprecation as a known tradeoff: customers who built on older models may resist migrating, and Anthropic may sometimes keep legacy versions running longer for safety or other reasons. Plan for the model you use today to be replaced, and keep the model name in one place so switching is a one-line change you test first.
Why do Claude accounts get restricted or banned?
Anthropic's Usage Policy says that if it suspects a violation, it may warn you or throttle, limit, suspend, or terminate access. Its S-1/A estimates that 5% to 7% of revenue in the first nine months of 2026 came from customers violating its usage policy or commercial terms, and says enforcement was stepped up in August and September 2026. The current version of the Usage Policy on Anthropic's site takes effect November 12, 2026.
Is Anthropic going bankrupt?
Its own filing does not point that way in the near term. As of September 30, 2026 it reported $95.6 billion in cash, $36.3 billion in short-term investments, and $14.8 billion of undrawn credit, against about $69.8 billion of compute commitments due in the following 12 months. Its long-term commitments total about $895.4 billion through 2045, which depends on revenue continuing to grow. This is not investment advice.
how exposed is your own ai setup?
the free ai reliability scorecard takes 60 seconds: seven questions, a score out of 100, and where your setup breaks before a limit change or a model retirement finds it for you.
take the scorecardwant the version with people in the room? the ai builders lounge is where the weekly builds happen.
or just follow along. new field notes most weeks on x, instagram, and tiktok.