Part 5/11:
OpenAI’s choice to partner with Broadcom stems from the latter’s rich history and expertise in custom silicon and system design. Unlike relying solely on GPU giants like Nvidia or AMD, Broadcom offers a long-standing track record of working with hyperscale cloud providers including Google, Meta, and Amazon.
Historically, Broadcom has been instrumental in co-developing Google's TPU, enabling Google to process over a quadrillion tokens per month—more than four times OpenAI's current processing rate. Their involvement with Meta’s MTIA chip and Amazon's Tranium and Inferentium hardware stacks underscores Broadcom’s extensive experience in building specialized, large-scale AI infrastructure.