Anthropic’s IPO Filing Exposes a Cloud Capacity Crunch

Threat assessment: Anthropic’s biggest suppliers are now its rivals.

Anthropic can’t get enough cloud capacity from AWS, Google, and Microsoft. Those companies increasingly treat the Claude maker as a competitor, not a customer. That’s according to an S-1 filing leaked to The Wall Street Journal and reported by TechCrunch AI. The S-1 is the document a company files before going public, and Anthropic is reportedly aiming for a November IPO.

The filing adds detail to figures that Reuters and the Financial Times reported in late September. It also shows how much this company needs compute, and how much it’s spending to get it.

📋 Situation Report

Here’s what the leaked filing shows, as detailed in TechCrunch AI:

1. Cloud spend could top $750 billion. Anthropic believes it could spend more than that on cloud contracts, and some of those deals are still being negotiated.
2. It’s still not enough. Anthropic is leasing its own data centers and working with third-party operators to fill the gap.
3. The backup plan has weak spots. The company worries those operators have a “limited operating history,” so they could prove less reliable than the hyperscalers.
4. Losses are huge. Anthropic posted an $80.7 billion net loss in the first half of the year.
5. Revenue is growing fast. It brought in $16.3 billion in the first half. Last week it told prospective investors that third-quarter revenue alone hit $16.8 billion, up 45% from the previous quarter.
6. A few customers carry a lot of the business. Two customers account for about a fifth of sales.
7. Liability could be bigger than its insurance. The filing warns Anthropic could be held liable for damage caused by harmful AI behavior, and those costs could exceed its available insurance coverage.

🎯 Why This Matters

What stands out here is the position Anthropic is in. Amazon and Google are major investors and its main infrastructure providers. Microsoft has become a partner too. All three also build and sell their own AI models. So every GPU they rent to Anthropic is one they can’t give their own teams or other customers.

For years that tension stayed quiet. Hyperscalers were happy to collect cloud revenue while AI labs trained frontier models on their hardware. Now that Anthropic’s quarterly revenue rivals some of their cloud divisions’ growth, the relationship looks less like a partnership and more like a negotiation between rivals.

The customer concentration matters too. When two buyers make up roughly 20% of revenue, losing or shrinking either one would hit hard. Public market investors will watch that number closely.

⚠️ The Compute Problem

Compute is the oxygen of frontier AI. You can’t train bigger models or serve millions of Claude users without it. Anthropic’s move into leased data centers and newer operators shows it’s diversifying out of necessity, not preference.

That comes with risk. Newer data center firms may not have the track record, power contracts, or operational depth that AWS and Google built over two decades. If one of them stumbles, Claude’s uptime and Anthropic’s training roadmap could suffer.

🔭 What to Watch

1. The IPO timeline. A November listing would make Anthropic one of the most closely watched public offerings in tech history. Expect more disclosures as the official filing goes public.
2. Hyperscaler moves. Watch whether AWS, Google, or Microsoft restructure their deals or prioritize their own models when capacity gets tight.
3. Reliability for Claude users. If you build on Claude, keep an eye on rate limits, capacity constraints, and pricing changes. Supply pressure upstream often reaches developers.
4. AI liability disclosures. Anthropic is openly telling investors that AI harms could cost more than insurance covers. Other labs heading toward public markets may have to make similar disclosures.

🧭 Assessment

Anthropic is growing faster than almost any company in the industry, and it’s also burning cash at historic levels. The filing shows a company boxed in by its own success: it needs more compute than its partners want to sell it, and the competitors who own that compute have little reason to help.

The November IPO will test whether public investors buy the growth story or focus on the losses and dependencies behind it. Full details are available in the original TechCrunch AI report.