Anthropic's IPO Prospectus Puts AWS Dependency at Center of Its Public Story

Anthropic filed a prospectus this week that tells two stories at once. The first is a growth chart that's hard to argue with. The second is a dependency map that any infrastructure-minded investor should read carefully.

According to Reuters, which reviewed the filing, Anthropic's revenue surged 12-fold in 2025, reaching nearly $4.6 billion. Usage-based billing drove most of that: <cite index="28-4">about $3.8 billion came from customers paying based on how much they use Claude, while subscription revenue came to $789 million.</cite> But <cite index="28-2">operating losses more than doubled to top $8 billion</cite> over the same period, which tells you where the money is going before it ever arrives.

Where it's going is compute. Specifically, Amazon's compute. <cite index="19-14">Anthropic committed to run Claude on nearly 1 gigawatt of Trainium2 and Trainium3 capacity by the end of 2026, with up to 5 gigawatts of total AWS capacity secured</cite> under a deal announced in April. That agreement, reported by both companies simultaneously, has Anthropic <cite index="21-3">spending more than $100 billion on AWS technologies over the next 10 years, securing current and future generations of Trainium and Graviton cores.</cite>

The architecture decision embedded in that commitment is worth naming clearly. Anthropic isn't just buying cloud time on a spot market. <cite index="11-11">The overall commitment includes Trainium2, Trainium3, Trainium4, and the ability to purchase future generations of Trainium as they become available,</cite> meaning Anthropic's training and inference stack is tied to Amazon's custom silicon roadmap for the better part of a decade. That's not a vendor relationship; it's a supply chain merger.

The prospectus, as described by CNBC citing Reuters, doesn't bury this. <cite index="33-1,33-2">Anthropic plans to warn investors that its AI models pose a 'catastrophic or existential risk to humanity,' and the filing dedicates 80 pages to the risks of AI, nearly double the 48 pages it spends discussing its business plans.</cite> The company also flags model dependency risk prominently. <cite index="28-6">In the prospectus, Anthropic framed its relationships with Amazon, Google, and Microsoft as an advantage, saying it can tap their vast sales networks to reach customers already using their services.</cite>

That framing works until it doesn't. <cite index="27-5">Claude services have recently experienced persistent downtime, rate limiting, and performance degradation, with some customers switching to other AI platforms</cite> due to compute constraints. The AWS and Google deals are, at least in part, a direct answer to that operational failure.

On the IPO itself: <cite index="29-2,29-3">Anthropic filed a confidential S-1 draft with the SEC on June 1, 2026, and is reportedly targeting a November 2026 listing at around a $2 trillion valuation.</cite> The Wall Street Journal and Reuters have both pointed to Nasdaq. <cite index="35-17,35-18">A roadshow is targeted for mid-October, with the listing itself targeted for November; the date was pushed back from an earlier October window so the offering can include Q3 results.</cite>

That means a public S-1 lands soon, probably within weeks. When it does, the number worth watching isn't the revenue run rate or the valuation ask. It's the line items for compute costs relative to gross margin. That's where you'll see whether Anthropic's model of locking in gigawatts of custom silicon at a fixed commitment actually produces better unit economics than buying Nvidia GPUs on the open market, or whether it just trades one kind of price risk for another.

Sources cited:
- Reuters (via The Star / CNBC) (https://www.cnbc.com/2026/09/29/anthropic-warns-ai-existential-risks-ipo-filing-reuters.html)
- Anthropic official announcement (https://www.anthropic.com/news/anthropic-amazon-compute)
- About Amazon (official press release) (https://www.aboutamazon.com/news/company-news/amazon-invests-additional-5-billion-anthropic-ai)
- The New Stack (https://thenewstack.io/anthropic-amazon-aws-investment/)
- Useluminix / Luminix AI Strategic Research (https://www.useluminix.com/reports/company-overviews/what-do-we-know-about-the-anthropic-ipo)
- TradingKey (https://www.tradingkey.com/analysis/stocks/us-stocks/261804457-anthropic-amazon-aws-trainium-openai-tradingkey)