OpenAI’s multi‑vendor commitments
OpenAI is reportedly committing tens of billions to AMD, plus $10B+ to Cerebras and Broadcom and massive spend on AWS Trainium, according to industry posts argued. Those allocations point to OpenAI spreading inference and accelerator risk across multiple suppliers rather than relying on one vendor argued.
OpenAI and AMD announced a multi‑year pact to deploy 6 gigawatts of AMD Instinct GPUs, with an initial 1 GW slated for deployment in the second half of 2026 (openai.com), and AMD issued OpenAI a warrant for up to 160 million shares (≈10% of AMD) that vests against deployment milestones (ir.amd.com). OpenAI and Broadcom unveiled a collaboration to co‑develop and deploy 10 gigawatts of OpenAI‑designed custom accelerators, a program OpenAI says will deliver next‑generation systems and networking by 2029 (openai.com); Broadcom previously disclosed a >$10 billion order from a new hyperscaler customer during its earnings call, a figure market reports linked to the OpenAI relationship (tech.yahoo.com). Bloomberg and CNBC reported a separate multiyear agreement with Cerebras for roughly 750 megawatts of wafer‑scale compute capacity through 2028, part of a deal Bloomberg described as exceeding $10 billion in committed spend for OpenAI’s training and inference needs (bloomberg.com). Amazon and OpenAI formalized an expanded AWS partnership that includes a planned consumption of about 2 gigawatts of AWS Trainium capacity and an upsized multi‑year cloud commitment (reports place the expanded AWS total near $138 billion, with Amazon also announcing a $50 billion strategic investment) (aboutamazon.com). Taken together the headline commitments on public record — AMD 6 GW, Broadcom 10 GW, AWS Trainium ~2 GW and Cerebras ~0.75 GW — sum to roughly 18.75 gigawatts of targeted accelerator capacity across multiple vendors, with timelines stretching from 2026 through at least 2028–2029 per the respective releases and reports. (openai.com) OpenAI’s agreements include non‑standard commercial levers — equity warrants with AMD and OpenAI‑designed hardware programs with Broadcom — and the company’s public statements and industry analysts have framed the package as an explicit move to diversify supplier risk and secure scale ahead of forecasted demand spikes. (ir.amd.com)
