Marvell Technology’s stock has been on a tear this April, and the latest reason is a report that Google is in early talks with the chipmaker to develop two new artificial intelligence chips. The news, first reported by The Information, sent Marvell shares climbing and added fuel to what has already been a remarkable year for the company.
To understand why this matters, you need to know a little about the quiet but fierce competition happening behind every AI product you use.
Marvell Technology is not a household name, but the company plays a critical role in the AI supply chain. While Nvidia gets most of the headlines for selling the powerful GPU chips that train AI models, Marvell operates in a different lane. The company designs custom silicon, which means it builds chips tailored to a specific customer’s needs rather than selling one-size-fits-all processors.
Think of it this way: Nvidia sells the equivalent of a high-performance sports car that works well for almost anyone. Marvell, on the other hand, builds a car from scratch to fit exactly what one particular driver needs. That kind of custom work is increasingly valuable as tech giants look for ways to run AI more cheaply and efficiently.
Marvell’s custom silicon business now generates roughly $1.5 billion per year, and the company has landed design wins with 18 different cloud providers. Its client list reads like a who’s who of big tech: Amazon (where Marvell helps build Trainium processors), Microsoft (for the Maia AI accelerator), and Meta (for a new data processing unit). Marvell also already works with Google on the Axion ARM-based CPU.
Google has its own line of AI chips called Tensor Processing Units, or TPUs. These are the processors that power Google’s AI services, from the Gemini chatbot to Search features to cloud computing products offered to outside customers. For years, Broadcom has been Google’s primary partner for manufacturing these chips, and the two companies recently locked in a TPU supply agreement that runs through 2031.
So why would Google also start talking to Marvell? The answer comes down to two things: diversification and a shift in how AI computing works.
According to reports, Google is discussing two specific chips with Marvell. The first is a memory processing unit designed to work alongside Google’s existing TPUs, essentially making them faster and more efficient at handling data. The second is a brand-new TPU built specifically for inference.
If “inference” is a new word for you, here’s what it means. Training an AI model is the phase where the system learns from enormous amounts of data. Inference is what happens after: it’s when the trained model actually does useful work, like answering your questions, generating images, or recommending videos. As AI tools become more widely used, inference is quickly becoming the bigger cost for companies like Google. Building chips optimized specifically for inference could save Google billions of dollars over time.
Marvell’s stock has gained roughly 55% since the start of 2026, hitting a new 52-week high near $139. A significant chunk of that climb happened in April alone, driven by two major developments back to back.
First, at the end of March, Nvidia announced a $2 billion investment in Marvell through convertible preferred stock. That deal gives Nvidia about a 2.5% ownership stake in the company and ties the two firms together through a technology partnership called NVLink Fusion. Under this arrangement, Marvell will design custom AI accelerators and networking products that plug directly into Nvidia’s computing ecosystem. The two companies will also collaborate on research into silicon photonics and next-generation wireless networks.
The Nvidia deal alone was enough to push Marvell shares up nearly 20% in a single session. Then the Google reports landed just a couple of weeks later, giving the stock another push higher.
Barclays analyst Tom O’Malley upgraded Marvell to overweight (Wall Street’s way of saying “buy”) and raised his price target from $105 to $150, citing what he called massive demand growth for optical networking and AI infrastructure through 2027.
There’s a larger trend at work here that explains why Marvell has become such a hot stock. The biggest technology companies are increasingly designing their own custom chips instead of relying solely on off-the-shelf products from Nvidia and AMD.
Amazon has its Graviton and Trainium chips. Google has TPUs. Microsoft is building Maia. Meta is developing its own accelerators. The reason is straightforward: when you’re spending tens of billions of dollars a year on AI computing, even small efficiency gains from custom-designed chips can translate into enormous savings.
But these companies don’t actually manufacture their own silicon. They need partners like Marvell and Broadcom to turn their designs into real, working chips. The custom chip market (known in the industry as custom ASICs) is projected to grow 45% in 2026 alone, and analysts expect it to reach $118 billion by 2033.
Marvell is positioning itself as the go-to partner for this work. If the Google deal goes through, Marvell would have relationships with essentially every major cloud computing company on the planet. That’s a strong position to be in as AI spending continues to accelerate.
Before anyone gets too carried away, there are real caveats worth understanding. The Google talks are just that: talks. No contract has been signed. Chip development takes years, so even if a deal is reached soon, the resulting products likely wouldn’t hit production until 2028 or later.
There’s also the question of valuation. After a 55% rally, Marvell’s stock is priced for a lot of good news. If the Google deal falls through, or if any of its other partnerships stumble, the stock has room to fall. Chip development is expensive and complex, and not every project makes it to market.
Finally, Marvell isn’t the only company competing for this business. Broadcom remains Google’s established TPU partner and recently secured that long-term contract extension. Other chip designers are also eager for a piece of the custom silicon pie.
For anyone following the AI space, the Marvell-Google talks highlight a shift that will shape the industry for years. The era of “just buy more Nvidia GPUs” is giving way to something more nuanced: a world where the largest tech companies build highly specialized chips for specific AI workloads, partnering with firms like Marvell to get the job done.
This is good news for AI users. More competition in the chip market means more innovation, lower costs, and ultimately better AI products. When Google can run its AI services on cheaper, more efficient custom chips, those savings eventually trickle down to the cloud computing customers and everyday users who rely on those services.
The custom chip race is heating up, and Marvell just moved closer to the front of the pack.
