Broadcom Seeks $80 Billion Debt to Finance AI Chip Supply for Anthropic

People think the artificial intelligence boom is simply a story of tech companies buying Nvidia graphics processing units with venture capital. The reality is that the scale of compute required for next-generation frontier models has outgrown traditional equity financing. It has become a heavy infrastructure play, financed by massive debt instruments usually reserved for building national power grids, toll roads, or telecom networks.

Broadcom is currently negotiating between $70 billion and $80 billion—and potentially up to $100 billion—in debt financing to fund custom AI chip deployments for Anthropic and other frontier labs. The massive private credit deal signals a fundamental shift in how the industry finances the staggering physical costs of the AI race.[1][2]

The financing is expected to be issued through a special purpose vehicle, or SPV. Under the proposed structure, Broadcom will use its investment-grade credit to partially guarantee a senior secured debt tranche of roughly $60 billion to $70 billion.[4][5]

A subordinated, or junior, tranche of approximately $30 billion to $35 billion completes the package. This tiered structure is designed to attract institutional investors, such as pension funds and insurance companies, who are seeking reliable, utility-like yields from the AI boom without taking on venture-style equity risk.[4][5]

This is not Broadcom's first foray into massive private credit arrangements. In June 2026, the semiconductor giant partnered with Apollo Global Management and Blackstone to launch the "AI XPV Platform."[3]

That initial June transaction provided $35 billion to help Anthropic deploy one gigawatt of compute capacity—enough electricity to power roughly 750,000 homes—using Broadcom's custom chips and networking equipment.[3]

The new $80 billion package dramatically expands that initiative. It aims to support Broadcom's broader goal of enabling more than 20 gigawatts of compute capacity for leading AI labs, including Anthropic and OpenAI, by 2028.[1][3]

The shift toward Broadcom highlights a technical pivot within the AI industry. While Nvidia's general-purpose GPUs currently dominate the market, they are expensive and highly power-intensive. Frontier labs are increasingly shifting toward Application-Specific Integrated Circuits, or ASICs.[5]

ASICs are custom silicon chips designed specifically for a lab's unique AI models and workloads. Because they are hardwired for a single purpose, they offer significantly better speed, energy efficiency, and operating margins than general-purpose hardware.[5]

Broadcom is a leading designer of these custom ASICs. By organizing a dedicated "chip-purchasing capital pool," Broadcom essentially finances its own customers. This allows the company to lock in massive, long-term orders for its custom silicon and networking hardware, securing sticky revenue.[5]

The financial engineering here is just as important as the silicon. Holding the hardware in an SPV keeps the massive capital expenditure off Anthropic's balance sheet. The AI lab effectively leases the compute capacity from the vehicle, while Broadcom secures guaranteed sales.[5]

Broadcom is not alone in deploying this strategy. Nvidia recently announced it would supply up to $105 billion to fund a new OpenAI data center in Ohio, and launched a $500 billion funding initiative with asset managers to establish computing infrastructure as a distinct asset class.[2]

Meanwhile, Marvell Technology, another major custom chip designer, recently expanded its relationship with Google, highlighting the intense race among semiconductor firms to secure exclusive hyperscaler partnerships.[4][6]

This transition marks the rapid commoditization of AI compute. Computing power is becoming a standardized utility, financed, packaged, and traded much like commercial real estate or energy infrastructure.

However, debt on this scale carries significant weight. By acting as a financier, Broadcom is taking on substantial customer concentration risk and guarantee exposure.[4]

If the AI market cools, or if a frontier lab fails to generate the revenue required to service its compute leases, Broadcom's backstop guarantees on the senior debt could trigger massive corporate liabilities.[4]

Ultimately, the Broadcom-Anthropic financing model proves that the primary bottleneck in artificial intelligence development is no longer just algorithmic innovation. It is the mobilization of institutional capital at an unprecedented scale, requiring the world's largest asset managers to physically build the next era of computing.