Share with your CTO
Nvidia posted $81.6 billion in quarterly revenue, up 85% year over year, with data center alone contributing $75.2 billion, while Broadcom’s AI chip revenue hit $10.8 billion last quarter, up 143%, with management guiding toward $16 billion next quarter. The comparative performance of these two AI chip strategies now lands on a surprising valuation flip: Nvidia, the faster grower, trades at roughly 20 times forward earnings versus Broadcom’s 25 times, inverting the usual growth-premium logic.
What this means for your business
Your infrastructure vendor mix is quietly being decided by forces you don’t control. Nvidia holds the general-purpose GPU layer where training and inference for frontier models runs, while Broadcom owns the custom-silicon layer where hyperscalers design their own accelerators to reduce that dependency. If your AI roadmap runs on cloud, both companies are inside your bill of materials whether you’ve thought about it or not. The question isn’t which stock to own; it’s which architectural direction your primary cloud partners are betting on, because that bet flows straight into which hardware your workloads land on.
The Broadcom custom-silicon story deserves sharper attention from infrastructure planners than it typically gets. Custom AI accelerators, chips designed specifically for a single customer’s workload rather than sold as general-purpose hardware, are how Google’s TPUs and Meta’s MTIA chips get built. Broadcom designs and manufactures these for a handful of the largest cloud operators. As those customers scale custom silicon, they reduce GPU purchases from Nvidia, which is why Nvidia’s stock cooled in 2026 even while revenues climbed: the market is pricing in eventual share erosion at the margin, even if the absolute numbers remain extraordinary.
The falsification condition for Nvidia’s continued dominance is straightforward: if the three or four hyperscalers commissioning Broadcom custom chips successfully generalize those designs beyond their own internal workloads and begin offering them as competitive inference products to enterprise customers, Nvidia’s pricing power compresses faster than current forward estimates suggest. Watch whether AWS, Google, or Meta start aggressively marketing custom-silicon inference tiers to enterprise buyers as a price alternative to Nvidia-backed instances. That’s the leading indicator that the architectural competition has moved from internal cost optimization into your vendor renewal conversation.
Based on reporting from $1,000 in Nvidia vs. $1,000 in Broadcom: Which AI Chip Bet Paid Off More?, originally published 2026-07-12 04:30:00.
