Marvell Stock at a Record $147: The Google Call That Changed the Chip Game

Only semiconductor analysts and business IT purchasers could recognize Marvell Technology at first glance two years ago. Storage controllers and networking chips are produced by a reputable but unglamorous Santa Clara-based company. Then came the AI capital expenditure cycle, and at some point in mid-2024, Marvell started to subtly change into something the tape couldn’t ignore. That change resulted in a particular and unsettling milestone for the bears on Monday, April 20, 2026: shares ended the day at $147.84, up 5.83%, following reports from The Information and Reuters that Google and the company are in advanced talks to co-develop two new AI chips.

To be honest, the stock has been nearly vertical. Up about 84% so far this year from its close of $80.46 on December 31, 2025. roughly 170% higher than the previous 12 months. And currently trading above the $131 consensus analyst target price, which typically occurs when the news flow outpaces the sell-side models. Tuesday’s pre-market saw MRVL ticking up at $151.61 once more. The market capitalization has increased to approximately $129 billion, with the Google headline alone adding over $6 billion in a single session.

What was stated in the headline? Google is investigating two particular chips with Marvell, according to reports in The Information and Reuters. The first is a memory processing unit, which is essentially an on-package or near-package component that sits next to Google’s current Tensor Processing Units and speeds up data transfer during AI inference and training. The true bottleneck in contemporary AI systems is memory bandwidth rather than compute, which is why Broadcom has spent years creating unique accelerators with memory as the primary constraint and why Nvidia has developed the Grace Hopper architecture. The second chip is a next-generation TPU that has been specifically tuned for inference, which is the portion of AI that actually responds to user inquiries and has quietly grown to account for a larger portion of hyperscaler spending.

Marvell will join Broadcom and MediaTek as Google’s custom-silicon partners if both chips are put into production. That is a significant structural change. Since 2016, Broadcom has essentially controlled the Google TPU partnership. Google’s desire for parallel suppliers for its most strategic roadmap is indicated by the addition of a second named U.S. partner. This move is similar to Meta’s diversification strategy and challenges Broadcom’s near-monopoly over the hyperscaler custom-chip market. As a result of the news, Broadcom’s stock fell roughly 1.7%, ending at $399.63.

The Google story coincides with a potentially even more significant Nvidia deal. In order to integrate Marvell’s unique XPUs and networking silicon into NVLink Fusion, Nvidia’s interconnect ecosystem, Nvidia announced a $2 billion strategic investment in the company on March 31. The dollar amount is essentially irrelevant. What it really means is that instead of competing with Marvell’s custom chips, Nvidia has chosen to accept them as compatible guests inside its own stack. For any hyperscaler considering a Marvell-designed accelerator rather than an off-the-shelf Nvidia GPU, that kind of co-existence significantly lowers the switching cost. It’s difficult to ignore how corporate-Darwinian Nvidia’s decision to turn a rival into an ecosystem partner is.

In the meantime, the operating numbers are catching up to the chart. Revenue for the fiscal year 2026 reached a record $8.195 billion, and adjusted earnings increased by 81% from the previous year. In fiscal 2026, data center revenue increased by 46%, surpassing a $6 billion run rate. Estimates from Counterpoint Research By 2027, Marvell might have about 25% of the market for custom AI accelerators, compared to Broadcom’s estimated 60%. Gross margins are still higher than 51%. The management has made it clear that by fiscal 2028, revenue should be close to $15 billion. As of right now, MRVL is trading at about 44 times forward earnings, which is more than Broadcom’s 28, which the market appears ready to accept given Marvell’s quicker growth rate.

This is a stock pricing in almost perfect execution, so proceed with caution as usual. According to a Reddit thread that surfaced late on Monday, JPMorgan has reportedly denied the details of the Google TPU story, but the relationship as a whole has been less contested. The headlines won’t cooperate for a few weeks. There will be quarters in which the stock falls 15% before lunch due to a single hyperscaler customer delaying a program. This is the nature of a name whose beta is greater than two and whose valuation is predicated on the uninterrupted compounding of AI capital expenditures. As this develops, it appears that Marvell has at last earned a place at the table, but the market is still determining how much that seat is worth. According to what I’ve read, the story of this quiet chip company that recently became well-known is far from over.