Alphabet's Google is reportedly in advanced negotiations with Marvell Technology to co-develop a pair of specialized artificial intelligence chips, according to a Sunday report from The Information. The discussions, which have not been officially confirmed by either company, could significantly advance Marvell's position in the fiercely competitive arena of custom silicon design for major cloud infrastructure providers.
Strategic Timing Amid AI Spending Surge
The potential collaboration comes at a pivotal moment for Alphabet, which has projected its capital expenditures could reach as high as $185 billion in 2026. This massive investment is earmarked for scaling up servers, data centers, and networking capabilities to fuel its expansive AI initiatives. The financial momentum is already evident, with Google Cloud revenue soaring 48% year-over-year in the December quarter to $17.7 billion, a surge analysts attribute to robust enterprise demand driven by AI adoption.
On the other side of the table, Marvell Technology has provided optimistic guidance, forecasting nearly $15 billion in revenue for its fiscal 2028. The company's leadership has consistently pointed to the AI wave as a primary catalyst for this upgraded outlook, with spending from large cloud customers described as "still growing massively."
Details of the Chip Designs
The chips under discussion are highly specialized. One is a memory processing unit (MPU) designed to work in tandem with Google's proprietary tensor processing unit (TPU), the engine behind its AI model training and execution. The second is a new TPU variant specifically architected for inference—the critical phase where a trained AI model generates responses to user prompts. According to the report, the two companies are targeting to finalize the design for the memory chip sometime next year, which would then pave the way for test production runs.
It is crucial to note that these discussions do not guarantee a finalized agreement. Reuters has stated it could not independently verify The Information's story, and both Google and Marvell declined to comment on the report.
Complex Web of Chip Alliances
Google's semiconductor strategy involves multiple partners. Earlier this month, Broadcom announced a long-term agreement to collaborate on and supply future iterations of Google's custom AI chips and related components, a partnership extending through 2031. Both Broadcom and Marvell are known for assisting cloud service providers in developing tailor-made data center chips, offering an alternative to Nvidia's dominant general-purpose AI accelerators.
Marvell itself has been strengthening its industry ties. On March 31, Nvidia disclosed a $2 billion investment in Marvell, with CEO Jensen Huang proclaiming, "The inference inflection has arrived." Analysts noted this partnership enhances Nvidia's access to more specialized silicon and Marvell's optical interconnect technology. Furthermore, Meta Platforms recently extended its own custom-chip engagement with Broadcom through 2029, signaling a broad-based rush among tech giants to secure specialized silicon supply chains for burgeoning inference workloads.
Market Context and Financial Impact
For Marvell, custom chip designs currently constitute an estimated 10% to 15% of its total revenue. The company's management has highlighted a record pace of design wins, which peaked in fiscal 2026. Marvell's stock closed at $139.69 on Friday, representing a gain of $6.37, before The Information's report surfaced on Sunday. With Nasdaq trading confined to weekdays, the market's reaction to this news will first be seen at Monday's opening bell.
The broader trend is clear: the largest buyers of AI computing power are aggressively pursuing custom silicon solutions to optimize performance, power efficiency, and cost for their specific workloads, moving beyond off-the-shelf offerings. This reported dialogue between Google and Marvell is a fresh indicator of that strategic shift, placing Marvell squarely at the intersection of several key industry relationships and the intensifying race for AI hardware supremacy.
