Broadcom AI chip revenue jumps 221% to $16.7 billion

Broadcom's numbers come from a position few chipmakers occupy: designing custom AI silicon for the largest buyers rather than selling standard parts. It is the design partner for Google's Tensor Processing Units (TPUs, custom chips built for AI work) and for OpenAI's Jalapeno inference chip, and it expects to ramp shipments of Ironwood TPU version 7 to Anthropic in the current quarter. Tan said on the Sept. 2 call that Broadcom plans to deliver "tens of billions of dollars of TPUs annually over the next several years" for just one customer, Google.

The supply side cuts both ways. Higher memory prices are pressuring gross margin, and component shortages — memory especially — could make the fiscal 2028 goal harder to hit. Tan said Broadcom has already secured the supply it needs, and added that "our demand actually exceeds this outlook, and we will work to improve supply." The customer list is also narrow: six buyers account for the bulk of AI chip orders, so its revenue depends on their willingness to keep funding data center construction. Regulation that slows new data center building, or rising public opposition over utility bills, is a stated risk.

What the outcome hinges on is whether AI infrastructure spending outlasts current expectations. If it does, the gap between Broadcom's roughly 12 times fiscal 2028 earnings and the roughly 25 times current-year estimates of leading chip stocks could narrow — a possibility the article frames as investors doubling their money. If the build-out slows, that same narrow customer base works against the stock.