Custom AI Silicon Creates New Growth Paths Beyond GPUs
The AI semiconductor market is expanding beyond standard GPUs as Broadcom and Marvell benefit from hyperscaler demand for custom accelerators and networking gear. Nvidia remains the largest accelerated-computing player, while AMD is growing as an alternative across GPUs and CPUs. Recent results show strong growth across these companies, though their scale, valuations, and exposure to future deployments vary widely.
Broadcom's latest quarter showed $29.59B revenue, with $16.7B from AI semiconductors, up 221%, and free cash flow of $13.7B. It guided next-quarter AI chip revenue near $21.7B. Marvell's quarter was smaller at $2.74B, but data-center sales reached $2.17B, up 46% and 79% of total, with management projecting about 60% data-center growth.
Nvidia posted $96.2B quarterly revenue, including $89B data center, up 117%, at 75% gross margin, and guided $108B. AMD's revenue was $11.5B, up 50%, with data center more than doubling to $6.7B and 56% non-GAAP gross margin. Marvell and AMD carried much higher forward earnings multiples than Broadcom and Nvidia.
The shift toward custom AI silicon could affect investors through divergent valuations and earnings, while cloud providers may gain more tailored compute and bargaining power. Workers in chip design, networking, and data-center operations may see demand shift. Broader users could see AI services improve or become cheaper if competition lowers costs, though reliance on a few hyperscalers may concentrate influence.