Google Diversifies AI Chip Strategy, Tapping Marvell for Custom Silicon

Marvell Technology shares climbed nearly 6% on Monday following reports that Google intends to utilize the firm's design expertise for two new artificial intelligence processors. The potential collaboration is expected to encompass both a Tensor Processing Unit (TPU) and a memory processing unit. Historically, Google has relied on Broadcom for the development of its in-house TPUs. While the Google-Broadcom partnership remains intact—having been extended through 2031 earlier this month—the shift toward Marvell suggests a strategic effort by Google to diversify its silicon pipeline. Broadcom shares dipped nearly 2% following the report. This development follows a strong period for Marvell, which saw a $2 billion investment from Nvidia in March. The company's stock has surged nearly 50% in April, bolstered by robust fourth-quarter earnings and escalating demand for AI-driven application-specific integrated circuits (ASICs). The move reflects a wider industry scramble among hyperscalers like Amazon, Meta, and Microsoft to develop lower-cost alternatives to Nvidia's dominant GPU offerings. Google, a pioneer in this space since 2015, recently released its 7th generation 'Ironwood' TPU and may unveil further hardware at the upcoming Google Cloud Next conference. The diversification of chip design partners helps mitigate supply chain risks and addresses critical bottlenecks, particularly in memory processing. As Meta and other firms continue to scale their custom silicon, the competition between design firms like Marvell and Broadcom is expected to intensify.