Investors borrow $36 billion for Google chips destined for Anthropic

Anthropic is financing the purchase of extremely expensive servers through third-party investors, who have raised $36 billion in borrowed funds. It shows how AI companies competing with OpenAI cope with astronomical infrastructure costs.

A circular financing scheme

According to Bloomberg, a group of investors is taking out a $36 billion loan to buy Google TPUs (Tensor Processing Units), specialized chips that train and run modern AI models. The chips are then handed over to Anthropic.

This approach means Anthropic itself does not act as the borrower and does not bear direct legal responsibility for how the financing is used.

The structure resembles those used at OpenAI. Money changes hands many times, the companies get the hardware they need, but they formally keep their distance from the debt obligations.

Why investors are taking this risk

AI infrastructure requires capital spending that goes beyond traditional financing:

• A single data center worldwide costs $10–20 billion
• Google TPUs are supplied in limited volumes and only under long-term contracts
• Training a modern LLM takes millions of compute hours
• Competition for chips has intensified, so capacity must be reserved years ahead

For investors, this is a risky bet: if Anthropic grows in value and captures a significant share of the AI market, their investment will pay off many times over. If not, they are left holding expensive hardware.

What this means

Fintech and traditional financiers are weaving themselves into the core of the AI industry. The race for computing power may soon hit a borrowing limit, with banks setting strict conditions or refusing to lend altogether. That could slow the growth of new AI startups and concentrate power in the hands of large companies with their own cash flows.