Alphabet's Google is reportedly in talks with Marvell Technology to develop two new types of chips aimed at running artificial intelligence models more efficiently. According to sources familiar with the matter, one of the chips is an in-memory processing unit designed to work alongside Google's Tensor Processing Units, while the other is a new TPU specifically built for operating AI models. Marvell Technology would take on a design services role, similar to how MediaTek participated in Google's latest Ironwood TPU project. No formal contract has been signed yet.
NVIDIA's GPUs currently dominate the AI chip market, but Google has been positioning its TPUs as an alternative. As Google seeks to demonstrate that its significant AI investments are translating into commercial returns, TPU sales have become a key growth driver for its cloud business. Sources indicate that Google and Marvell aim to finalize the design of the in-memory processing unit as early as next year, followed by a trial production phase.
The negotiations come just days after Broadcom secured a TPU supply agreement extending through 2031. The move also signals Google's increasing focus on inference, a major cost component in AI applications. The custom ASIC market is projected to grow 45% by 2026, reaching an estimated $118 billion by 2033. Analysts suggest Google's strategy is not to replace Broadcom but to add a third design partner to its supply chain, emphasizing diversification rather than substitution.
Google's current custom chip supply chain includes Broadcom for high-performance TPU variants, MediaTek for cost-optimized "e" series TPUs (20%-30% cheaper), and TSMC for manufacturing. Separately, Google and NVIDIA's successive engagements with Marvell highlight the chip designer's growing role in the AI data center custom chip market. Marvell recently signed a $2 billion strategic cooperation deal with NVIDIA.
In the fiscal year ending February 2026, Marvell's data center business achieved a record $6.1 billion in revenue, with total revenue reaching $8.2 billion, a 42% year-over-year increase. The company's custom chip business generates approximately $1.5 billion in annualized revenue and has secured design orders from 18 cloud providers, including chips for Amazon's Trainium processors, Microsoft's Maia AI accelerators, and Meta's new data processing units. Marvell is also handling existing design work for Google's Axion ARM CPU.
Year-to-date, Marvell's stock has surged over 56%, with a 30% gain in April alone driven by news of its collaboration with NVIDIA and negotiations with Google. Barclays analyst Tom O'Malley upgraded the stock to "Overweight" and raised his price target from $105 to $150.
