Nvidia (NVDA 0.78%) holds roughly 85% to 92% of the artificial intelligence accelerator market today. That kind of dominance invites serious competition -- and in this case, much of it is coming from the very cloud giants currently paying Nvidia billions every quarter. The four major hyperscalers -- Alphabet (NASDAQ: GOOGL) (NASDAQ: GOOG), Microsoft, Amazon, and Meta Platforms -- are all aggressively developing custom AI silicon to reduce their dependence on any single supplier.
One company sitting at the center of that shift is Marvell Technology (MRVL +1.16%).
Image source: Getty Images.
Why Marvell is different
Marvell Technology is not a brand most casual investors will recognize. It doesn't make consumer devices. It doesn't market to developers. But it does design some of the custom chips and connectivity fabric that allow modern AI data centers to function.
Its custom application-specific integrated circuit (ASIC) business grew from near zero to $1.5 billion in annual revenue in a single fiscal year. The company has 18 design wins with hyperscalers, including Microsoft and Amazon, meaning it already sits inside the hardware stacks of some of the most aggressive AI infrastructure build-outs in the world.
This month, reports surfaced that Alphabet's Google is in talks with Marvell to co-develop two custom AI chips: a memory processing unit and a next-generation Tensor Processing Unit (TPU) for inference workloads. Marvell stock jumped more than 13% in a single session on that news. Critically, this isn't a one-time contract announcement. It is a signal that hyperscalers are turning to Marvell as a strategic co-design partner, not just a vendor.
NASDAQ: MRVL
Key Data Points
The architecture of the opportunity
Why would Google, with one of the world's most sophisticated chip teams, need Marvell? Because building cutting-edge inference silicon while staying compatible with the broader Nvidia CUDA infrastructure requires a level of chip interconnect expertise that Marvell has spent decades building. Its PCIe, Ethernet, and photonic networking offerings tie into virtually every major AI infrastructure architecture.
In Marvell's fiscal 2026, which ended Jan. 31, its revenue grew by 42% to $8.2 billion. But management now expects to reach $15 billion in total sales in fiscal 2028, with top-line growth of more than 30% in the current year, and close to 40% in the next. The stock currently trades at around $152 per share -- down a bit from the all-time high it touched last week -- and analysts have been boosting their price targets recently in light of the Google news.
The company also recently acquired Celestial AI for $3.25 billion to bring its photonic interconnect technology in-house. This technology enables data to move between chips at the speed of light, which is the next binding constraint in multichip AI systems. That acquisition signals management is already thinking past the current generation of AI infrastructure.
Beyond raw speed, Celestial AI's photonic fabric is designed to break the so-called "memory wall" by delivering massive bandwidth and low-latency connections between processing chips and memory hardware. This lets AI systems more effectively scale across entire racks and clusters. In layman's terms, this means easy speed, easy speed, and more easy speed.
NASDAQ: NVDA
Key Data Points
Is the "next Nvidia" just clickbait?
The "next Nvidia" framing isn't about Marvell unseating the GPU giant. Nvidia will likely remain the dominant processor provider in AI training for the foreseeable future, and the moat it enjoys due to the ubiquity of its CUDA software platform is genuinely deep. What the comparison is intended to reflect is the trajectory that I think Marvell is on. Like Nvidia, it started as a specialized niche supplier. Now, it too is earning design win after design win with the major hyperscalers, and eventually, it appears likely to become so deeply embedded in the infrastructure stack with its custom ASICs and photonic interconnect technology that it's effectively irreplaceable.
That's exactly what Nvidia did with gaming and then with data center GPUs.
Marvell stock is up over 80% over the last six months, but the combination of 18 hyperscaler design wins, a Google co-development partnership, photonic interconnect hardware, and 30%-plus projected annual growth makes Marvell Technology the most compelling AI play that most retail investors haven't discovered yet. Those with a long-term view should consider adding it to their portfolios.
