Anthropic's leaked S-1 is a vendor risk story, not just IPO news

Anthropic's leaked S-1 landed in The Wall Street Journal this week, and the headline numbers are brutal: an $80.7 billion net loss in the first half of this year against $16.3 billion in revenue, two unnamed customers accounting for roughly a fifth of all sales, and AWS, Google, and Microsoft — the three hyperscalers Anthropic depends on for compute — increasingly treating the Claude-maker as a competitor. If you're running Claude in production, this isn't an IPO story. It's a vendor risk story, and it landed on your desk whether you wanted it or not.

TechCrunch reported Tuesday on the leaked filing, noting that Anthropic is having trouble securing enough cloud capacity from those same hyperscalers, which now see it as a rival. To fill the gap, Anthropic is leasing its own data centers and working with third-party operators — but the S-1 itself warns that those operators have a 'limited operating history,' which is a polite way of saying the backup plan is unproven. The filing also discloses that Anthropic believes it could spend more than $750 billion on cloud contracts, some of which covers deals still in negotiation. Anthropic reportedly told prospective investors last week that Q3 revenue alone hit $16.8 billion, up 45% from the prior quarter. The company did not respond to TechCrunch's request for comment, and a November IPO is reportedly in the works, though TechCrunch's framing is 'reportedly' — not confirmed.

The structural problem hiding inside the growth story

The revenue growth is real and it's fast. Going from $16.3 billion across the first half of the year to $16.8 billion in Q3 alone is a sharp acceleration. Anthropic will use that number in every roadshow conversation, and it deserves credit. But growth doesn't fix structure.

Think about what the S-1 actually describes. Your primary infrastructure providers — the companies whose data centers your product runs on — view you as a competitive threat. You're leasing alternative capacity from operators with limited track records. You believe your cloud spend could exceed $750 billion, much of it not yet locked in. And two customers you can't name represent about 20% of your revenue. Lose one, and you've got a material revenue event. Lose cloud priority at AWS or Google during a capacity crunch, and you've got a service event. Both at once, and you've got a crisis.

The liability disclosure makes it stranger still. The S-1 warns that Anthropic could face damages from harmful AI behavior that would exceed its available insurance coverage. That's a standard risk-factor boilerplate move, yes — but it's also the company putting in writing that the downside scenarios are, at some scale, uninsurable. Enterprise buyers who've embedded Claude into customer-facing workflows should read that sentence twice.

The fair counterargument is that every pre-IPO S-1 is engineered to disclose every conceivable risk, and Anthropic's revenue trajectory is genuinely impressive. A company growing 45% quarter over quarter is not obviously fragile. And hyperscaler tension is an industry-wide dynamic, not unique to Anthropic.

That's true, and I'd take it seriously — if the concentration risks weren't so specific. Two customers at a fifth of revenue isn't just a risk-factor boilerplate. It's a real cliff. And 'our cloud providers see us as a competitor' isn't a hypothetical; it's an active negotiation problem the company is disclosing under oath.

What this means if Claude is in your stack

The practical question isn't whether Anthropic survives. It probably does, especially if the IPO closes at a reasonable valuation. The question is what happens to your workflows during the messy middle — the period between now and whatever stable state the company reaches post-IPO, post-capacity negotiations, post-customer-concentration fix.

A few things worth doing this week. First, pull your Anthropic contract and check what SLA commitments exist around uptime and capacity, and what remedies you have if those aren't met. If the answer is 'not much,' that's useful to know now rather than during an outage. Second, ask your team which internal processes would break, and how badly, if Claude API access degraded for 48 to 72 hours. If the answer is 'a lot,' that's a single-vendor dependency worth addressing before it's urgent. Third, if you're in procurement conversations with Anthropic right now, the S-1 disclosures are legitimate negotiating context — particularly around service continuity guarantees and what happens if capacity constraints affect response times or availability.

None of this requires abandoning Claude. It requires treating Anthropic the way you'd treat any vendor whose S-1 disclosed this combination of concentration risk, infrastructure uncertainty, and uninsurable liability scenarios.

What to watch

The November IPO timeline is the first real test. A successful offering at a strong valuation gives Anthropic capital to lock in its own infrastructure and reduce hyperscaler dependency. A delayed or repriced offering tells you the market read the same S-1 you just did and had questions. Watch also for any public signal from AWS, Google, or Microsoft about the terms of their renewed or renegotiated capacity deals — that's the structural risk that the IPO alone won't resolve. My bet is the IPO gets done, but the cloud capacity tension takes at least another two or three quarters to stabilize, and enterprise buyers who plan for that window will be glad they did.