We're coming to the end of another round of closely watched AI earnings, and investors are already scanning the sector to see where AI spending goes next. That will keep the focus on NVIDIA Corp. (NASDAQ: NVDA). But the AI buildout is no longer a one-stock story.
Every hyperscaler capital expenditure (CapEx) update points in the same direction: spending keeps rising. That money doesn't just buy GPUs. It funds custom chips, data center networking, and the power and cooling systems that keep it all running.
Three companies sit at different points along that chain: Broadcom (NASDAQ: AVGO), Arista Networks (NYSE: ANET), and Vertiv (NYSE: VRT). Each represents a distinct bottleneck in the AI infrastructure buildout, not a substitute for NVIDIA itself. Here's how each one fits into the broader picture.
Why "Second-Layer" Beats "Better Than NVIDIA"
These three stocks aren't alternatives to NVIDIA. They don't compete for the same dollar. NVIDIA sells the GPUs doing the AI work. These companies sell what has to exist around those GPUs for a data center to function at scale.
A building full of the fastest processors in the world is useless if it can't move data between racks, pull enough power, or stay cool under load. That's the second layer of the AI trade.
Broadcom covers custom silicon and chip-level networking. Arista covers the data center network fabric. Vertiv covers power and cooling. If any one of those layers breaks down, the buildout slows, regardless of how good the chips are.
Broadcom: Custom Silicon and the Networking Glue
Broadcom sits in two places in the AI stack. It designs custom AI accelerators for the largest hyperscalers, built to each customer's own specifications rather than sold off the shelf. It also makes the networking silicon that connects those chips, allowing thousands of processors to function as one system.
Management has pointed to AI-related revenue scaling sharply higher over the next few years across both businesses combined. The exact figure matters less than the direction: AI is becoming a larger share of total revenue.
The stock swings between two narratives. One treats Broadcom as the next high-growth AI name. The other prices it as a diversified semiconductor conglomerate with AI exposure attached.
That gap between perception and fundamentals shows up in how sharply shares move on any single earnings report.
The real risk is concentration. A small number of hyperscaler customers drive AI revenue, and the company carries meaningful debt from its acquisition history.
Arista Networks: The Plumbing Between the Racks
Arista makes the high-speed Ethernet switching connecting servers inside a data center. Increasingly, it also connects data centers to each other. Training a large AI model isn't one chip working alone. It's thousands of chips communicating constantly, and Arista builds the network fabric that lets them do it without bottlenecking.
Recent quarters showed revenue growth in the high-20% range, with management guiding to continued acceleration and calling out AI networking as a segment growing faster than the business overall.
Arista often gets compared directly to Broadcom, but the two aren't fully interchangeable. Broadcom sells components used across many vendors' equipment. Arista sells complete networking systems directly to hyperscalers. The market sometimes prices Arista as a cheaper way into the same trade, and other times as the purer AI-networking bet worth a premium.
That premium is an honest risk. Arista trades above Broadcom on a forward basis, meaning strong execution is already priced in. Any softness in hyperscaler networking budgets could quickly compress that multiple.
Vertiv: Power and Cooling as the Physical Limit
Vertiv doesn't sell chips or networking gear. It sells the power distribution and thermal management systems that keep data centers running. Every new generation of AI chips draws more power and generates more heat than the last, and many older data centers weren't built for that density.
This is a physical constraint, not a demand constraint. It doesn't depend on which chipmaker or model architecture wins the broader AI race. Whoever wins still needs power and cooling.
Vertiv's most recent quarter showed revenue growth in the mid-20% range, with organic growth in the high teens. Shares have climbed sharply over the past year, and the company's inclusion in the S&P 500 broadened its institutional ownership base.
Perception has lagged fundamentals here more than with the other two names. Vertiv can look like an industrial company having a good year, when it's actually a direct bet on the physical limit of the entire AI buildout. That gap has been closing as more investors link AI growth to demand for power infrastructure.
The risks: shares have already run hard, project timing can make revenue lumpy, and a high beta means sharp moves in both directions around earnings.
The AI infrastructure trade is broader than the chip headlines suggest. Broadcom, Arista, and Vertiv each carry their own risk profile. Understanding which bottleneck each one represents is the first step before deciding where they fit in a portfolio.