Broadcom is in talks with lenders to raise $70-80 billion in debt to finance a chip and computing infrastructure arrangement benefiting artificial intelligence companies, including Anthropic, as the AI industry races to secure the hardware needed to support its expansion. According to CNBC's David Faber, the company is in discussions to secure a massive loan package of $70 billion to fund chip deals for AI firms like Anthropic. Bloomberg was first to report the story, with sources indicating the total package could eventually climb to $100 billion. The deal would extend Broadcom's growing role in AI infrastructure and deepen its relationship with Anthropic, which is also laying the groundwork for its own custom semiconductor business. This would be one of the largest corporate debt deals tied to AI growth, showcasing how much outside money is now being used to build the huge computing power needed by leading AI companies.
Lenders are targeting a senior tranche of about $45 billion and a junior tranche of roughly $35 billion, according to CNBC, although the figures could change as discussions continue. Bloomberg has separately reported that the financing could include a senior-secured tranche of $60-70 billion and a junior tranche of around $30 billion. Under that structure, the total financing could reach $100 billion. Private equity firms Blackstone and Apollo Global Management are among the investors discussing participation, with CNBC reporting that the money would flow through a special-purpose vehicle, following the structure of an earlier $35 billion arrangement. The debt is expected to be issued through a special-purpose vehicle, or SPV, which keeps the borrowing separate from Broadcom's main balance sheet. The structure is similar to a $35 billion deal announced by the same group of companies in June, with Broadcom agreeing to "backstop" most of the senior debt, meaning it would provide support for that part of the borrowing.
The proposed financing builds on a partnership announced in June between Broadcom, Apollo and Blackstone, where the companies agreed to invest $35 billion to expand Anthropic's computing infrastructure using Broadcom's custom chips and networking technology. The original June deal was expected to add about one gigawatt of computing capacity, but the wider partnership has much bigger ambitions. Broadcom, Apollo and Blackstone are targeting more than 20 gigawatts of computing capacity for leading AI companies by 2028, with the cost potentially running into hundreds of billions of dollars. To put that figure into perspective, 20 gigawatts is roughly equal to the output of 20 nuclear power plants. The arrangement highlights the enormous infrastructure requirements of the AI industry, where companies are competing for processors, data centres and networking capacity. Bloomberg reports that Broadcom unveiled an AI platform in June designed to deliver 20 gigawatts of computing power for both Anthropic and OpenAI.
As Broadcom works to expand its AI chip business, Anthropic is also strengthening its semiconductor capabilities by hiring Amir Salek, a founder of Google's custom chip programme, to join its compute team. Salek led Google's Tensor Processing Unit business until 2022 and worked on the first seven generations of the company's chips. At Anthropic, Salek will report to James Bradbury. The company has also agreed an initial order worth about $250 million with UK-based chip start-up Fractile, with the possibility of expanding the deal, and signed computing-capacity agreements with Riot Platforms and Volta Infra Holdings. Anthropic's relationship with Broadcom has grown significantly, with the company planning to use up to one million of Google's Tensor Processing Units in October 2025, a deal worth tens of billions of dollars expected to add more than one gigawatt of computing capacity in 2026. The partnership expanded again in April when Anthropic and Google signed another agreement covering about 3.5 gigawatts of TPU-based computing capacity starting in 2027, part of Anthropic's plan to invest more than $50 billion in computing infrastructure in the US.
The semiconductor push comes as Anthropic currently sources processors from several companies, including Nvidia, Google and Amazon. Developing its own chips could give the AI company greater control over the hardware needed to train and run its models. Custom silicon could also help Anthropic address supply constraints and tailor chip designs to its specific computing requirements. OpenAI is pursuing a similar strategy, having unveiled a chip called Jalapeno, developed with Broadcom, and plans to begin using it later this year. Broadcom itself designs custom chips for companies including Alphabet, Meta, Anthropic and OpenAI, helping large technology firms reduce their reliance on Nvidia, the dominant supplier of AI processors. The semiconductor boom is evident across the industry, with Nvidia announcing this week that it would supply up to $105 billion to fund a new OpenAI data center in Ohio, as detailed in a securities filing, and partnering with six large investment firms to create a $500 billion funding initiative focused on making computing infrastructure an investable asset class.
