OpenAI Unable to Secure $18 Billion for Broadcom Chip Rollout

OpenAI cannot close roughly $18 billion in financing tied to an early tranche of its planned 10‑gigawatt rollout of custom accelerators built with Broadcom, leaving a gap in funding for the program meant to reduce reliance on third‑party GPUs.

The shortfall affects the initial portion of a multi‑year accelerator program that would deploy OpenAI‑designed chips produced in partnership with Broadcom. The Broadcom agreement, announced in October 2025, outlined a broader hardware buildout of about $500 billion covering chips, networking gear, power infrastructure and data center capacity.

OpenAI projects heavy cumulative spending as it scales model training and infrastructure. Company estimates show cumulative spending could reach about $115 billion through 2029 even as revenue grows. Reports of missed internal growth targets have led investors to re‑examine capital plans and vendor commitments.

Industry capital spending for 2026 has risen sharply. Analysts estimate hyperscaler capital expenditures this year at roughly $600 billion to $720 billion, with about three‑quarters aimed at AI infrastructure. Lenders and investors have grown more cautious about the timing of returns on those investments.

Some partners have used their own balance sheets to support related commitments; for example, Oracle issued about $18 billion in bonds last September to back a roughly $300 billion cloud and AI commitment tied to its OpenAI collaboration. Hardware suppliers report strains as well: unpaid customer balances at a major GPU vendor have climbed toward $33 billion as buyers take longer to settle invoices. Some investors have questioned revenue accounting tied to large AI contracts.

OpenAI’s options to address the shortfall include restructuring the tranche to spread payments over a longer period, finding replacement lenders, or scaling back the immediate scope of the rollout. Any of those choices would alter the timing and size of projected capital expenditures in 2026 and beyond and affect how quickly the company can shift away from existing GPU suppliers.