Anthropic reportedly pursuing additional $36 billion funding round for AI infrastructure capex.

Anthropic, a leading artificial intelligence company, is preparing for another massive funding round to expand its computing capacity. According to Bloomberg reporting, investment firm Blackstone has begun preliminary discussions to arrange a second $36 billion debt package to finance Google Tensor processors. The deal underscores the enormous scale of spending required for AI technology development.

As talks remain in early stages, the final size, structure, and participants may yet change. However, the new funding is expected to follow an approximately $35 billion deal announced two months prior, under which funds were directed toward long-term leases of Google computing infrastructure across five large data centers operated by Anthropic.

The previous deal was structured through a special-purpose vehicle (SPV) that purchased computing equipment based on Google TPUs with a capacity of approximately one gigavatt. Anthropic obtained this capacity through a long-term lease rather than an outright purchase—an approach that allows the software developer to avoid substantial upfront capital expenditure. According to the Financial Times, this model effectively finances AI infrastructure as if it were a major power facility.

Experts characterize this structure as an emerging financial mechanism for the high-tech sector. The vast computing power required to train and operate AI models is now being financed through long-term debt instruments and dedicated asset classes, mirroring the financing structures of large-scale power generation.

At present, the parties are discussing only preliminary matters, with no indication that a final agreement has been reached or new chips ordered. Representatives from Blackstone, Anthropic, Google, and Apollo declined to comment on the reported financial plans.

If successfully completed, the $36 billion deal would rank among the largest infrastructure transactions in AI industry history and would significantly shape how computing capacity supply and financing mechanisms develop going forward.