Custom silicon chips are gaining traction in AI data centers, with Broadcom, Marvell, and Taiwan Semiconductor positioned to capitalize on the shift away from Nvidia's general-purpose GPUs.

Broadcom's custom ASIC sales for customers doubled in its first quarter to $8.4 billion, with Alphabet as a leading customer; the company's management estimates artificial intelligence revenue will reach $100 billion by next year. Marvell reported total sales rising 42% to $8.2 billion in 2026, driven by AI growth and custom design work for Microsoft and Amazon.

Custom semiconductors allow large tech companies to tune processors for their specific AI models more efficiently than general-purpose GPUs. Broadcom expanded its customer designs for Alphabet's Tensor Processing Units (TPUs) for Alphabet's AI data centers through 2031, and Marvell partners with Amazon on its proprietary Trainium chips.

Custom ASIC processors are expected to grow by an estimated 45% this year, compared with GPUs' 2026 growth rate of 15%. TSMC, which manufactures these processors and holds 90% market share in advanced artificial intelligence processors, saw sales surge 41% in the first quarter to $35 billion.