When everyone knows that Nvidia is incredibly strong, GPUs are in short supply, and AI data centers are expanding at a frantic pace, the smartest money on Wall Street has already started asking the next question: Where will AI capital flow after Nvidia?
Over the next few days, a seemingly low-profile company——Marvell, ticker MRVL on the U.S. stock market,Marvell Technology, is about to release a blockbuster earnings report that may provide the key answer.
On May 27, Marvell will report its FY2027 first-quarter earnings. On the surface, it looks like an ordinary chip company earnings release, but in reality it is areport card for the second phase of the AI rally. It carries three of the hottest labels at once: custom AI chips, optical networking, and data center interconnects. Cloud giants want to build their own AI compute highways, and Marvell may be the one helping them build the roads, lay the cables, and make the critical components.
1. Core question: Will AI capital spread from GPUs outward?
The focus of this issue is not only Marvell as a company, but a bigger question: now that Nvidia is familiar to everyone, will the next wave of AI money spread from GPUs into custom chips, optical communications, and data center networking—the deeper infrastructure layers?
-
Phase one of the AI rally: the logic was simple——whoever has GPUs has the power to set the narrative. The profit formula was just buy Nvidia.
-
The logic of phase two of the AI rally: once a story becomes universally known, capital starts looking for the second layer of opportunity.
Gold rush analogy: the first people to get rich were the miners digging for gold (Nvidia). Once everyone rushed into the mines, the people selling shovels, tents, jeans, and water (infrastructure suppliers) also started making money.
An AI data center cannot consist of GPUs alone. It also needs complete servers, high-speed networks, optical communications, switching chips, and custom AI chips to make tens of thousands of GPUs work together. That is where Marvell’s value becomes visible.
2. Marvell’s role: an infrastructure provider for the AI superfactory
Think of an AI data center as a superfactory:
-
Nvidia sells the most critical engine (GPU).
-
Marvell builds the conveyors, highways, wires, control systems, and custom machines. It participates in custom AI chips and makes high-speed interconnect, optical communications, and data center networking chips.
The bottleneck in AI is not only compute, but data movement. When thousands of GPUs are placed together, the core issue becomes: how do they communicate with each other? How is data transmitted? How are bandwidth, latency, and power consumption solved? It is like a city: even if the cars are great, if the highways are completely jammed, nobody can move. The next phase of AI is not only about who computes faster, but who makes data travel faster. That is exactly why the market is repricing Marvell.
3. The driver: the “collaboration dividend” from cloud giants’ in-house chips
Microsoft, Google, Amazon, Meta, and other cloud giants are all doing one thing: developing their own chips.
The reason is not that Nvidia is bad, but precisely that it is too good, too expensive, and too scarce. For giants spending tens of billions of dollars a year building AI infrastructure, relying solely on external GPUs long term is too risky: costs are high, supply is unstable, power consumption is heavy, and chips need to be customized for their own models and scenarios. So these giants will not abandon Nvidia, but they will definitely keep a second plan: continue buying general-purpose GPUs while accelerating in-house and custom chip development.
That is Marvell’s opportunity. It is not directly replacing Nvidia, but capturing another piece of the pie——the “collaboration dividend” behind cloud giants’ in-house chips and custom AI chips.
-
Nvidia makes money from general-purpose GPUs.
-
Marvell makes money from the part that big customers must spend when they want to build their own AI compute systems.
The stronger Nvidia gets, the more expensive GPUs become, the better the custom chip value proposition looks; the larger AI data centers become, the more important high-speed interconnect and optical networks become. Marvell is not standing opposite Nvidia, but rather in the second layer of AI infrastructure expansion.
4. Earnings watchlist: five signals that must be tracked closely
For AI stocks, looking only at revenue and EPS is far from enough. For Marvell’s earnings this time, the five key signals are:
-
Whether data center revenue remains strong: high growth would show that AI demand has truly flowed through to basic layers such as networking, interconnects, and custom chips.
-
Whether custom AI chip projects have entered a volume ramp: the market cares most about whether big-customer projects have real orders, production timelines, and stronger forward guidance. If management releases positive signals, capital will continue trading it as a leading AI ASIC name.
-
Whether optical communications and electro-optical products continue to benefit: the larger the model, the more critical data transmission becomes; optical networking is a key piece of the AI data center expansion puzzle.
-
Customer concentration risk: landing orders from a mega-customer can create huge earnings upside, but if that customer delays orders or slows the pace, short-term volatility will also be very severe.
-
Valuation and guidance: the stock has already had a run, so expectations are not low. An earnings report being “good” may not be enough to make it rise; it must be “good enough,” and guidance must be “strong enough” to support market imagination. Otherwise, even if results meet expectations, it could still be a sell-the-news event.
5. Competitive landscape: Marvell vs. Broadcom
Many people first think of Broadcom when they hear custom AI chips. A comparison:
-
Broadcom: the older brother already seated at the main table. It has a very strong position in custom chips and networking chips, with larger scale, stronger customers, and greater certainty.
-
Marvell: a shadow player now being rediscovered by capital. Its advantage is that its upside may be larger, and the room for market repricing is more imaginative.
Key point: this is not to say Marvell is definitely better than Broadcom, but if AI capital starts spreading outward from Nvidia, Marvell is very likely to become one of the directions the market focuses on.
6. Macro significance: verifying whether the AI rally has entered phase two
Marvell’s earnings are not only about its own performance, but also about helping the market verify one judgment: Has the AI rally really entered phase two?
-
Phase one: the market only looks at GPUs.
-
Phase two: the market starts looking at AI servers, custom chips, optical communications, networking, power, data centers, and more. AI changes from a single-chip story into a massive infrastructure buildout story.
-
Dell answers whether AI server orders continue to explode.
-
Marvell answers whether custom chips and data center interconnects continue to ramp.
-
Broadcom answers whether demand for cloud ASICs and networking chips is accelerating.
-
If these companies all deliver strong guidance, it means the AI rally is not Nvidia dancing alone, but the whole industry chain expanding. Capital will then dig further down the chain, from GPUs to servers, custom chips, optical communications, memory, power, liquid cooling, and data center REITs; the entire chain will be reexamined.
7. Risk warning: this is not a story to buy blindly
Do not hear this story as “Marvell will definitely rise”; the risks are also obvious:
-
Expectations are already full: the market has already been trading it as an AI core stock, so earnings must be “much better than expected.”
-
Customer concentration: changes in the order pace from major customers will significantly affect short-term performance.
-
Fierce competition: Broadcom, Nvidia, AMD, and even cloud giants’ own chip teams are all fighting for the pie. This is not a blue ocean, but a battlefield of giants.
-
AI capex cycle: once cloud giants rein in data center spending in some future quarter, infrastructure stocks like Marvell will be very sensitive.
8. Core conclusion
Nvidia is not the end of the AI rally, but the starting point for the diffusion of the AI industry chain. Whether Marvell can catch the second wave of capital depends on how strong this earnings report is.
The real highlight of phase two of the AI rally is: after Nvidia, who can catch the second wave of money? Will it be custom chip and optical communication companies like Marvell, ASIC giants like Broadcom, or complete AI server manufacturers like Dell? This may determine the flow of U.S. AI capital in the coming period.
What really matters in the May 27 earnings report is not how much Marvell earned in one quarter, but whether it can prove one thing: AI capital is spreading from Nvidia into deeper infrastructure layers such as custom chips, optical communications, and data center interconnects.
