Marvell Technology shares shot up nearly 5 percent on Monday after a report that Alphabet’s Google is in serious discussions to team up with the chip designer on two new custom processors aimed at squeezing more performance and lower power bills out of the massive AI models that now dominate Silicon Valley’s spending.
According to The Information, the potential partnership would include a specialized memory processing unit meant to work hand-in-glove with Google’s existing Tensor Processing Units, plus an entirely new TPU variant optimized specifically for inference—the critical job of answering user queries in real time. Two people with direct knowledge of the talks described the discussions to the publication.
The news lands at a moment when the biggest cloud players are no longer content to rely on a single outside supplier for the silicon that powers their AI ambitions. Google already leans on its own TPUs for both training models and handling live traffic, and it has long collaborated with Broadcom on custom designs.
Register for Tekedia Mini-MBA edition 20 (June 8 – Sept 5, 2026).
Register for Tekedia AI in Business Masterclass.
Join Tekedia Capital Syndicate and co-invest in great global startups.
Register for Tekedia AI Lab.
But surging demand for AI infrastructure—and the eye-watering price tags attached to Nvidia’s GPUs—has hyperscalers scrambling to diversify. Meta, for its part, just extended a multigeneration deal with Broadcom and shelled out $2.3 billion last year alone for custom AI processors and related services.
Russ Mould, investment director at AJ Bell, put the calculus plainly, saying: “It should be no surprise that rivals (of Nvidia) will want to grab a piece of the market and the apparent growth on offer by developing their own product. It also makes sense for customers to diversify their sources of supply, if they can, so they can spread technological and supply chain risk.”
Marvell, which specializes in helping big customers design high-performance networking and custom AI silicon, has quietly positioned itself as a flexible alternative in this increasingly fragmented industry. The company already expects its revenue to climb toward $15 billion by fiscal 2028, a forecast that reflects confidence that specialized data-center chips will remain one of the hottest growth areas in tech.
Monday’s pop added more than $6 billion to Marvell’s market value, which stood at $122.15 billion before the trading day began. The stock is now up roughly 64 percent for the year after a bruising 23 percent decline in 2025. It trades at 33.35 times estimated earnings over the next 12 months, a premium to larger rival Broadcom’s 27.84 multiple—an indication that investors are pricing in faster growth as Marvell carves out a bigger slice of the custom-AI pie.
The Google talks also come just weeks after Nvidia itself poured $2 billion into Marvell. That investment was framed around making it easier for customers to mix Marvell’s custom AI chips with Nvidia’s networking gear and central processors, an unusually collaborative move that underscored how even the market leader is adapting to a world where no single company can satisfy every hyperscaler’s needs.
For Google, the potential Marvell partnership would be another step toward building a more cost-effective and resilient AI stack. Inference workloads now account for the bulk of day-to-day compute demand inside its data centers, and any efficiency gains there drop straight to the bottom line.
AI lab Anthropic already taps a mix of chips, including Google’s TPUs, to train and run its Claude models, showing how even cutting-edge labs are spreading their bets.
Neither Google nor Marvell responded immediately to requests for comment, and the talks could still shift or fall apart. But the market’s swift reaction speaks volumes. In an industry where a few percentage points of efficiency or a single reliable supplier can mean billions in savings or risk, even the hint of a new partnership between a hyperscaler and a nimble chip designer moves the needle.
It is thus clear that the AI hardware race is no longer a one-horse contest. Nvidia still commands the lion’s share of mind and wallet, yet the biggest customers are no longer willing to put all their chips, literally, on one vendor.
Marvell’s reported opening with Google, coming on the heels of its Nvidia investment, suggests the company is finding real traction in the spaces where customization, power efficiency, and supply-chain resilience matter most. If the deal materializes, it would hand Marvell a high-profile validation and potentially open the door to more hyperscaler work.