Google, Amazon, and OpenAI are designing custom AI chips instead of relying solely on Nvidia, and chip designers Broadcom and Marvell are capturing billions in revenue from this shift.

What happened: Broadcom reported AI chip and networking revenue of $10.8 billion（約1.7兆円） in its fiscal second quarter (ended May 3, 2026), up 143% from the prior quarter's $8.4 billion（約1.3兆円）. Marvell's data center segment, which includes custom chip design work, grew 27% to $1.83 billion（約2900億円） in its fiscal first quarter (ended May 2, 2026). Both companies are co-designing custom processors for major cloud operators including Google, Amazon, and OpenAI.

Why it matters: Large cloud providers are moving away from depending entirely on Nvidia for AI hardware, choosing instead to build custom chips tuned to their own software needs. Broadcom and Marvell benefit directly as the designers powering this shift, capturing a fast-growing portion of AI spending without having to manufacture the chips themselves. However, both stocks already reflect this optimism: Broadcom trades at a price-to-earnings ratio in the low 60s and Marvell near 100, leaving less room for disappointment.

What to watch: Broadcom has booked more than $30 billion（約4.8兆円） in AI orders in the quarter and guided for AI revenue to more than triple in the current quarter, with a possibility its AI business could top $100 billion（約16兆円） in annual revenue in fiscal 2027. Marvell expects its custom silicon business to top $10 billion（約1.6兆円） in annual revenue by fiscal 2029. Both companies face concentration risk, as their custom-chip revenue leans largely on a small number of cloud customers.