Marvell Technology’s stock didn’t just inch up in March—it rallied. Shares of the chipmaker rose 21.3% during the month, according to data from S&P Global Market Intelligence.
Part of the momentum came the old-fashioned way: numbers that landed better than expected. In its fiscal fourth quarter, Marvell reported a 22.1% jump in revenue to $2.2 billion. Adjusted (non-GAAP) earnings per share rose 33.3% to $0.80, and management guided for a strong 9% sequential jump in revenue in the first quarter, with a guide for $0.79 in adjusted EPS. Misryoum newsroom reported that the reported results and guidance both handily beat analysts’ expectations—so yeah, the market had something to grab onto.
Then the late-month news hit, and it was the kind investors usually circle in red. Marvell also announced a landmark investment from and product collaboration with Nvidia (NVDA +0.87%). The plan, at least in broad strokes, suggests Nvidia isn’t just buying into the AI pipeline—it’s looking to integrate more deeply with Marvell’s side of the infrastructure.
Inside the quarter, Marvell management said it expected data center revenue to grow 40% in fiscal 2027 (its current fiscal year), higher than the average analyst estimate of 25%. There’s been some consternation over the past year about whether Marvell has lost market share with its largest customer, Amazon, for Amazon’s custom Trainium chips. But forward guidance pushed back on that worry. Misryoum editorial desk noted that the all-important XPU and XPU-attach business—where Marvell provides crucial IP for custom AI chips—is alive and well. Also, it hasn’t been a one-customer story. Marvell has diversified, adding Microsoft to its roster. Microsoft unveiled its updated XPU, the Maia2 chip, back in January.
Agentic inference might be part of why this all keeps snowballing. In agentic setups, AI systems can be constantly communicating with large language models in data centers—and with each other. That means more networking traffic, and Marvell is positioned there. It’s a little counterintuitive, honestly: the “agent” concept sounds like software, but it still drags hardware demand along with it, and networking is where the bill lands.
Misryoum analysis indicates the networking piece likely explains the Nvidia investment angle too. Nvidia will invest $2 billion in Marvell, and the deal includes a product partnership. Typically, these days, AI infrastructure is either Nvidia-based, or companies deploy their own custom XPU chips and Ethernet technology. However, it now appears Nvidia is looking to integrate with Marvell, in order to enable heterogeneous infrastructures that can mix XPUs with Nvidia’s other technologies, such as NV-Link, its Vera CPUs, or even hybrid Nvidia GPU-plus-XPU architectures. The press release also mentioned a collaboration on silicon photonics. This stood out while I was reading—there’s something about the phrase “silicon photonics” that makes you think of clean rooms and humming equipment, even if you’re just staring at a screen.
This matters because silicon photonics may replace copper-based networking in the next generation of AI data centers. Nvidia’s current NV-link fusion is based on copper, so it appears Nvidia is seeking Marvell’s networking expertise to develop new optical networking products built on Nvidia technology in the future. Over the past year, Nvidia has been making similar types of deals with other semiconductor companies that might have been considered competitors, but have instead chosen to collaborate with the chip giant. These announcements are typically met with a jump in the target company’s shares, as investors appear to appreciate the vote of confidence and anticipate more future growth, offsetting concerns over dilution.
Misryoum editorial team stated Marvell should remain an AI winner. Marvell had a difficult 2025, but its resilient results have shown it as a beneficiary of the AI infrastructure build-out. It’s possible things could get even better in the age of agentic AI inference. Marvell is no longer as cheap as it was at times last year, but at 27 times this year’s earnings estimates, the stock is still buyable for those enthusiastic about the networking growth prospects of generative AI. And honestly, the market seems to believe that enthusiasm is going to keep showing up in earnings—maybe not forever, but for now.