Marvell shares gain on report of deal talks with Google to develop two AI chips

Marvell shares jumped 6% premarket after a Reuters-reported deal discussion that Google may work with the chip designer on two new AI chips, including a memory processing unit and a new TPU. The report suggests Google could diversify away from Broadcom as demand for custom AI chips remains strong, supporting Marvell's AI-related growth narrative. Nvidia's recent $2 billion investment in Marvell further underscores strategic interest in the company's custom chip efforts.

Analysis

This is less about one customer and more about a widening strategic wedge in the AI foundry stack. If Google is truly diversifying its custom silicon design partner set, it reduces Broadcom’s bargaining power on future TPU generations and raises the probability that design wins become more modular, with memory, networking, and compute increasingly sourced from different vendors. That is structurally favorable for the best-in-class merchant designer with exposure to multiple hyperscalers, but the near-term market move likely overstates how quickly a “talks” headline converts into revenue given chip design cycles, qualification, and tape-out risk.

The second-order winner is not just Marvell, but also the ecosystem around AI inference efficiency: anything that lowers power and memory bottlenecks should extend the economic life of deployed racks and increase the number of tokens generated per watt. That creates a subtle negative for Nvidia’s pricing power at the margin over the next 12-24 months, because customers keep trying to peel off the highest-cost workloads into custom silicon while still using Nvidia networking and CPUs as glue. The Nvidia investment in Marvell looks less like altruism and more like hedge construction—if custom designs proliferate, Nvidia wants to remain embedded in the interconnect layer regardless of who wins the accelerator socket.

For Broadcom, the risk is not immediate revenue loss but a valuation multiple ceiling: the market may start assigning a lower terminal share of wallet assumption if Google follows Meta in broadening its design partner base. The key catalyst window is the next 2-4 quarters, when hyperscaler capex guidance and commentary on inference efficiency will determine whether this is a one-off procurement story or the start of a broader vendor diversification trend. If the report is noise, MRVL likely gives back a meaningful portion of the gap; if Google confirms a multi-supplier strategy, the rerating can persist because it implies a longer revenue runway for custom silicon vendors beyond a single TPU program.

Key Decisions for Investors

• Go long MRVL on a 1-4 week horizon into any intraday pullback below the premarket gap, targeting a continued rerating as the market prices in broader hyperscaler design-share gains; stop if the stock closes back below the gap midpoint for two sessions.
• Pair trade: long MRVL / short AVGO for 1-3 months to express vendor diversification risk at Google and other hyperscalers; the trade works best if upcoming capex commentary emphasizes multi-sourcing and custom silicon expansion.
• Use NVDA weakness to add via call spreads 6-12 months out; the core thesis is that even with custom chips gaining share, Nvidia’s networking and system-level attach remains embedded in the AI buildout.
• If already long AVGO, trim 10-20% into strength and re-enter only after the next Google/Broadcom capex readout; the near-term upside is likely capped by multiple compression if design-share diversification becomes the new consensus.