Anthropic with a BB Rating Isn't Bankrupt: Google and Amazon Pay for 85% of Its Compute

A popular thesis: Anthropic is about to get a BB rating, won't be able to afford loans for data centers, and will go bankrupt. But the company mostly doesn't build its own data centers. Instead it signs long-term leases with Google and Amazon and pays from revenue. Anthropic's own credit quality is secondary here: the money for construction is raised by suppliers with their own high ratings.

AI-processed from TG: @seeallochnaya (Сиолошная); edited by Hamidun News
Anthropic with a BB Rating Isn't Bankrupt: Google and Amazon Pay for 85% of Its Compute
Source: TG: @seeallochnaya (Сиолошная). Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceTG: @seeallochnaya (Сиолошная) ↗
Editor-in-Chief
Key metric #185% of the 8 GW of new compute is Google TPU and AWS Trainium
BB per S&P statisticsroughly 5% default over 5 years and ~10% over 10
Key metric #3The main risk is the pace of revenue growth, not the cost of debt

In discussions of Anthropic's finances, a confident scenario often comes up: agencies are about to hand the company a "junk" BB-level rating, borrowing rates will soar, and the $900 billion in compute commitments will become unbearable. The conclusion: bankruptcy is inevitable. It sounds coherent, but it rests on a wrong premise.

Where the Logic Breaks Down

The thesis only works if Anthropic builds data centers itself. It would then have to first borrow huge sums for chips and construction, and then spend years paying them back with customer revenue. In that model, the cost of credit decides everything.

In practice, Dario Amodei does it differently: he signs long-term lease agreements with reliable providers. They are the ones who raise money for construction, while Anthropic pays rent directly out of revenue. For the company itself, the rating stops being the bottleneck. What matters more is that Google, Amazon and others can finance their buildouts cheaply.

Even where facilities are leased through Fluidstack, the payments are guaranteed by Google (rated AA) and Broadcom (A). In other words, credit risk largely shifts from Anthropic to giants with strong balance sheets.

Why Google and Amazon Are on Board

Of the 8 GW of expansion at identified sites, only 1 GW is Nvidia chips. The remaining roughly 85% is Google's own TPUs and AWS Trainium. For these companies, Anthropic is an anchor customer that helps them enter the AI chip market and take share from Nvidia, whose market capitalization shows how lucrative this market is.

As a result, the financing cost of about 85% of the commitments depends on the resilience of Google, Amazon, SpaceX and Broadcom. Only a small part relies on Anthropic's own credit quality. In its S-1, the company also writes that it may partly pay for future leases with raised capital.

What BB Really Means

The word "junk" is scarier than the numbers. According to S&P's long-term statistics, BB-rated companies default with a probability of about 5% over five years and about 10% over ten. The same news can be rewritten this way: rating agencies expect Anthropic to meet its obligations over the next 10 years with 90% probability. That no longer looks like a death sentence.

Limitations and Practical Takeaway

This doesn't mean there are no risks. The non-cancellable lease commitments had to be fitted to a revenue growth schedule. If Dario got the growth forecast wrong, the burden will become heavy, regardless of the rating. The author of the analysis openly calls the risk of collapse far from small.

For developers and founders building products on Claude, the takeaway is practical. When assessing a vendor's resilience, look not at the rating headline but at the structure of obligations: who builds, who pays, who guarantees. The key variable here is revenue growth. As long as it stays on plan, the scenario of "the bubble will burn out within a couple of years with a hundred percent guarantee" is not supported by the documents. But it is still sensible to have a plan B in case growth slows, as with any critical supplier.

There is also another side. Anthropic's bet on TPU and Trainium intensifies competition with Nvidia and may over time affect inference prices and availability. Watch how these sites expand: both API limits and costs depend on them.

Story

Key Questions Answered

Where the Logic Breaks Down?

The thesis only works if Anthropic builds data centers itself. It would then have to first borrow huge sums for chips and construction, and then spend years paying them back with customer revenue. In that model, the cost of credit decides everything. In practice, Dario Amodei does it differently: he signs long-term lease agreements with reliable providers. They are the ones who raise money for construction, while Anthropic pays rent directly out of revenue.

Why Google and Amazon Are on Board?

Of the 8 GW of expansion at identified sites, only 1 GW is Nvidia chips. The remaining roughly 85% is Google's own TPUs and AWS Trainium. For these companies, Anthropic is an anchor customer that helps them enter the AI chip market and take share from Nvidia, whose market capitalization shows how lucrative this market is. As a result, the financing cost of about 85% of the commitments depends on the resilience of Google, Amazon, SpaceX and Broadcom. Only a small part relies on Anthropic's own credit quality.

What BB Really Means?

The word "junk" is scarier than the numbers. According to S&P's long-term statistics, BB-rated companies default with a probability of about 5% over five years and about 10% over ten. The same news can be rewritten this way: rating agencies expect Anthropic to meet its obligations over the next 10 years with 90% probability. That no longer looks like a death sentence.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Unified AI Workspace · Nomad

Test Frontier LLMs & Autonomous Agents in One Workspace

Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.

Try Nomad AI →Launch Nomad Agent
What do you think?
Loading comments…