
[Adobe Stock]
The deal extends a pattern skeptics of the AI boom call circular financing. Hyperscalers such as Microsoft, Amazon and Google invest in frontier labs like OpenAI and Anthropic, sell them compute, and then book them as core customers. Chipmakers are following a similar playbook. Nvidia has taken stakes in startups that buy its GPUs, and in August it announced a partnership with Blackstone and other firms to mobilize more than $500 billion for AI, including financing for customers buying its chips.
“Nvidia is putting in place a massive amount of its balance sheet, and Broadcom is having to follow suit,” Seaport Research analyst Jay Goldberg told Reuters.
The arrangement also strengthens an alternative to Nvidia’s GPUs. Anthropic says it trains and runs Claude on Google TPUs, Amazon’s Trainium chips and Nvidia GPUs, matching workloads to the hardware best suited to them. Financing the TPU expansion could give Anthropic more options when allocating future workloads and negotiating computing capacity. The companies have yet to quantify any displacement of Nvidia purchases.
The obligation behind Broadcom’s financing is significant. Anthropic has committed $125.2 billion to a five-year lease of TPU capacity, according to the prospectus. That sits alongside other large compute commitments, including OpenAI’s deals with Oracle and Microsoft, worth $300 billion and $250 billion, respectively. Anthropic has placed cash in a restricted account for Broadcom’s benefit and may have to contribute more under certain circumstances. It warns that payment or performance defaults could cause a substantial portion of its lease obligations to become immediately due while limiting access to the financing facility. The filing also flags a potential conflict of interest, since Broadcom is both Anthropic’s hardware supplier and its financing partner.
That raises the question of how much revenue AI developers must generate to support their computing commitments. “Where’s the return on capital?” Yahoo Finance breaking business news reporter Jake Conley asked in an Oct. 1 discussion of the Broadcom arrangement. The discussion focused on the gap between the developers’ infrastructure spending and the revenue needed to repay suppliers and financiers.
Supporters argue that supplier financing helps companies secure scarce computing capacity and accelerate the AI buildout. “Vendor-backed financing brings capacity online faster than if every AI developer had to fund it themselves,” Anthony Saglimbene, chief market strategist at Ameriprise Financial, wrote in a September commentary. He also warned that financing customers can make reported demand appear stronger and concentrate risk among interconnected companies.
Critics see echoes of the late-1990s telecom boom, when equipment makers such as Lucent and Nortel lent upstart carriers the money to buy their gear. In 1998, Lucent extended Winstar a $2 billion credit line, with agreements requiring it to buy most of its equipment and services from Lucent. In 1999, Nortel agreed to lend NET-tel Communications, a Washington, D.C.-based carrier, $100 million for Nortel switches and other equipment. Such arrangements left suppliers exposed when their customers failed.
Other AI investments put the borrowing at a different point in the chain. On Oct. 1, SoftBank, one of OpenAI’s largest backers, completed a $10 billion equity investment in the company. It was the final tranche of a $30 billion follow-on commitment and brought SoftBank’s total to $64.6 billion for about 13% of OpenAI. SoftBank said it funded the tranche with proceeds from its own senior notes, after fully repaying a $40 billion bridge loan taken out mainly for its OpenAI investments. In that structure, the leverage sits with the investor, and OpenAI receives equity capital without a loan repayment obligation.
Across these structures, the payoff depends on AI developers turning computing capacity into revenue sufficient to sustain their commitments. “Universal buzz around AI may be a sign that the hype has moved faster than the payoff,” Apollo chief economist Torsten Slok wrote in an Oct. 1 note. His commentary examined whether growing mentions of AI on earnings calls translate into spending and measurable productivity gains, the outcomes that would help support the infrastructure being financed today.




Tell Us What You Think!
You must be logged in to post a comment.