Tech Giants Build Custom AI Chips to Reduce Nvidia Reliance

Major technology firms are increasingly designing custom application-specific integrated circuits (ASICs) to improve power efficiency and reduce their dependence on Nvidia Corporation. Amazon has developed Graviton, Trainium, and Nitro chips, while Microsoft has deployed its second-generation Maia 200 accelerator to support OpenAI models and Microsoft 365 Copilot. Alphabet is expanding its tensor processing units (TPUs) through eighth-generation systems and a joint venture with Blackstone to offer them as a rentable cloud service.

This shift toward tailored hardware is benefiting designers like Broadcom Inc. and Marvell Technology. Broadcom generated $10.8 billion in AI semiconductor revenue, largely driven by its work on Alphabet's TPUs. Marvell reported record first-quarter fiscal 2027 revenue of $2.42 billion and received a $2 billion investment from Nvidia to partner on future technologies. TrendForce Corp. projects that AI ASIC shipments will grow by 44.6% in 2026, significantly outpacing the 16.1% growth expected for GPUs.

Despite the rise of custom silicon, Nvidia remains the dominant force in the market. The company reported an 85% year-over-year revenue increase to $81.6 billion in its fiscal first quarter of 2027. Hyperscalers and AI startups continue to purchase record amounts of Nvidia GPUs, contributing to a projected $725 billion in total capital expenditures across the sector for 2026.