Broadcom CEO Hock Tan disclosed during the company’s fiscal Q2 2026 earnings call on Wednesday that Broadcom is financing custom AI chips for large language model developers including Anthropic and OpenAI. The financing vehicle, built in partnership with Apollo Global Management and Blackstone, is expected to deploy more than 20 gigawatts of compute capacity by 2028, with Apollo launching the platform’s first $35 billion tranche.
The arrangement reveals a structural shift in how frontier AI companies acquire compute. Rather than purchasing chips outright, model developers are now relying on their chip supplier to fund the silicon they need to scale.
How the Financing Works
Tan described the vehicle as a way to make Broadcom’s custom accelerators accessible to AI companies that would otherwise struggle to afford them at scale. “Creating in partnership with guys, with the best balance sheets around, a vehicle to basically have these chips funded for these LLM players who otherwise might have difficulty getting access to our technology, which provides them with the lowest power and the lowest cost,” Tan said during the call, according to Benzinga.
The need for external financing reflects how rapidly compute costs have escalated. Anthropic placed a $10 billion chip order with Broadcom in December 2025. At that scale, even well-capitalized AI companies face cash flow constraints.
Six Customers, $10.8 Billion in AI Revenue
Broadcom reported $10.8 billion in AI semiconductor revenue for Q2, a 143% increase year-over-year. Tan identified six core custom chip customers: Anthropic, Google, Meta, OpenAI, and two unnamed companies. These relationships are driving the bulk of Broadcom’s AI growth, according to CNBC.
The company forecast $16 billion in AI revenue for Q3, which would represent a tripling year-over-year. Tan reiterated a full-year target of more than $100 billion in AI semiconductor revenue for fiscal 2027 and said demand visibility now extends through 2028, up from 2027 just three months ago.
“The bookings that are coming are not for immediate delivery,” Tan said. “Some they hope to have, but the reality they all accept is they need to align quite a few other things in place before they can deliver.”
Chips Only, Not Systems
Tan also announced a strategic pivot: Broadcom will now offer “chips only” to its AI customers rather than complete integrated AI systems, which the company had previously committed to providing. The shift suggests that AI companies, particularly those building agent infrastructure, prefer to control their own system integration rather than buying turnkey solutions from a semiconductor vendor.
The Capital Stack Behind Agent Scaling
Despite strong AI revenue growth, Broadcom shares fell approximately 15% on Thursday after Tan declined to raise the company’s full-year 2026 AI chip sales forecast. Total Q2 revenue of $22.19 billion narrowly missed analyst expectations of $22.27 billion, and infrastructure software revenue of $7.18 billion came in below the $7.32 billion estimate, per CNBC.
The market reaction underscores a paradox in AI infrastructure: demand for custom compute is accelerating so fast that even 143% year-over-year growth disappoints investors expecting a raised forecast. For the companies actually building agents and frontier models, the bottleneck is no longer algorithmic. It is financial. When your chip supplier has to arrange private equity financing so you can afford the hardware, the capital requirements of scaling autonomous AI systems have entered a new phase entirely.