Marvell jumps on report of Google AI chip deal

Marvell Technology, a US chip design firm, rose nearly 6% after Google may use it for AI chips.

Broadcom, which has handled the design of Google’s in-house TPUs so far, fell almost 2% after the report.

Marvell and Broadcom both help customers turn chip designs into silicon before fabrication.

Google recently extended its Broadcom partnership through 2031.

The report comes as large tech companies build more custom AI chips to reduce reliance on Nvidia, while memory shortages remain a constraint across the sector.

🔗 Source: CNBC

🧠 Food for thought

Implications, context, and why it matters.

Google’s move could broaden a large custom chip plan

Google is looking beyond Broadcom and could add another design partner for strategic AI chips, which would make its supply chain sturdier if talks become a contract 1.
Marvell already has depth in custom silicon, with a business that brings in about US$1.5 billion in revenue 2.
A Google deal would give Marvell another cloud customer. The cited sources do not establish the claim that this would amount to a clean sweep of the top four US cloud providers 3.
The talks focus on AI bottlenecks. One chip would use a memory processing unit to speed data movement. Another would run AI models after training, a step called AI inference 4.

The AI chip fight is moving toward lower costs

Across the sector, companies are leaning toward custom chips called application-specific integrated circuits (ASICs). For some AI work, they can use 30-40% less power than general-purpose graphics processing units (GPUs) 5.
The ASIC market is projected to grow 45% in 2026, well ahead of the 16% forecast for GPUs 5.
More efficient hardware could trim the hidden costs of running AI, including overhead, aging infrastructure, and compute demands. That could cut spending on work such as ad targeting or data analysis if gains lower total cost of ownership 6.
Companies may shift toward custom silicon, yet Nvidia still holds an important role through partnerships. The cited source does not support the draft’s claim of a US$2 billion Nvidia-Marvell deal that would let Marvell’s custom chips plug into Nvidia’s network ecosystem 3.