The AI trade has stopped being a one-ticker story. The hyperscaler capex cycle — Microsoft, Meta, Alphabet, Amazon and Oracle together guiding to well over $400B of 2026 capex — is now showing up in the order books of utilities, electrical equipment makers, cooling specialists, fiber suppliers and memory vendors. This is the AI Infrastructure Earnings Index: 25 stocks whose results we watch each quarter to gauge the real, physical build-out behind generative AI. Use it as a watchlist, a sector map, and an earnings playbook. This is not investment advice. It is a framework for tracking the companies benefiting from AI data-center expansion and the data points in their earnings reports that actually move the stocks.
Why the AI trade has broadened beyond Nvidia
For most of 2023 and 2024, 'AI stocks' meant Nvidia, a short list of hyperscalers, and a handful of software names. That has changed. Every incremental GPU cluster needs land, transformers, switchgear, backup power, liquid cooling, optical interconnect, high-bandwidth memory and custom silicon. Each of those is a real, capacity-constrained industry with public companies that are now growing earnings at rates more typical of software than industrials.
The AI infrastructure trade is therefore really four overlapping baskets: power and electrical equipment, thermal and mechanical, networking and fiber, and silicon and memory. Together they form what we think of as the AI Infrastructure Earnings Index — a 25-name watchlist for investors who want exposure to data center stocks without betting the entire thesis on one GPU vendor.
The 25-stock AI Infrastructure Index
Power & electrical equipment (AI power stocks): Eaton (ETN), Vertiv (VRT), Schneider Electric (SBGSY), Quanta Services (PWR), Eaton-adjacent peers Hubbell (HUBB) and nVent (NVT), plus utilities with the largest hyperscaler interconnect queues — Constellation Energy (CEG), Vistra (VST), NextEra Energy (NEE) and Talen Energy (TLN).
Cooling, mechanical & construction (AI cooling stocks): Vertiv (VRT) again on liquid cooling, Comfort Systems USA (FIX), EMCOR (EME), Johnson Controls (JCI) and Modine Manufacturing (MOD).
Networking, fiber & optical: Corning (GLW), Arista Networks (ANET), Coherent (COHR), Lumentum (LITE) and Ciena (CIEN).
Silicon, memory & systems: Broadcom (AVGO), Micron (MU), Marvell (MRVL), Advanced Micro Devices (AMD), Super Micro Computer (SMCI) and Oracle (ORCL) as the hyperscale buyer that has effectively repositioned itself as an AI infrastructure operator.
That is the full 25. The names overlap in places — Vertiv shows up in both power and cooling — which is exactly the point: the most interesting AI infrastructure companies sit at the intersection of two scarce inputs.
Power & electrical equipment: Eaton, Vertiv, Schneider
Eaton (ETN) is arguably the cleanest large-cap proxy for US data center electrification. Its Electrical Americas segment has been growing at double-digit organic rates with backlog repeatedly hitting new records, and management has been explicit that data centers are the single largest driver. Watch three things in Eaton's prints: Electrical Americas organic growth, segment margin (a tell on pricing power as the cycle matures), and the 12-month and beyond-12-month backlog split.
Vertiv (VRT) is the purest AI infrastructure name in the index. Roughly three-quarters of revenue is now tied to data centers, spanning power distribution, UPS systems and — increasingly — liquid cooling for GPU racks. Vertiv's book-to-bill, organic orders growth and Americas backlog are the cleanest read on the marginal AI build-out each quarter.
Schneider Electric (SBGSY) is the global analogue: switchgear, power management software (EcoStruxure) and a fast-growing data center systems franchise. For US-listed investors the ADR is thinly traded, but Schneider's European prints are an excellent leading indicator for Eaton and Vertiv.
Cooling & mechanical: Comfort Systems and friends
As rack densities climb past 100 kW, the data center physically cannot be air-cooled at the chip. That has turned a previously sleepy mechanical contracting industry into one of the best-performing corners of the market.
Comfort Systems USA (FIX) is the standout. Its Modular and Industrial segments now do a meaningful share of revenue in technology — read: hyperscale and AI-related data centers — with backlog at multi-year highs. The earnings tell is simple: same-store revenue growth, technology mix as a percent of backlog, and gross margin (a check that the company is not buying growth).
EMCOR (EME) is the larger, more diversified peer with a similar story on the Network & Communications segment. Modine (MOD) is the smaller, higher-beta play through its Data Center Cooling group, which is now its fastest-growing business by a wide margin.
Fiber, optical & networking: Corning, Arista, Coherent
Corning (GLW) has, after years of underperformance, become a credible AI story through its Optical Communications enterprise business — the fiber and connectivity that links GPU clusters within and between data centers. Management has guided to a multi-year doubling of the enterprise optical run-rate driven specifically by generative AI. Watch enterprise carrier and enterprise mix, and any commentary on hyperscaler design wins.
Arista Networks (ANET) sells the high-speed Ethernet switches that interconnect those GPUs. The relevant earnings tells are AI-related deployment commentary from the top four customers, 800G product ramp, and any softening in cloud titan spend.
Coherent (COHR), Lumentum (LITE) and Ciena (CIEN) round out the optics layer: transceivers, lasers and the long-haul gear that connects data center campuses.
Silicon, memory & systems: Broadcom, Micron, Oracle
Broadcom (AVGO) is the second-most important AI silicon company after Nvidia. Its custom AI ASIC business — designing accelerators for Google, Meta and other hyperscalers — plus its AI networking franchise have together pushed AI revenue past a third of the semiconductor segment. The tell each quarter is the AI revenue dollar figure and the forward AI guide; everything else (broadband, software) is secondary for the multiple.
Micron (MU) is the cleanest pure-play on AI memory through high-bandwidth memory (HBM). HBM is sold out for the foreseeable future, ASPs are rising and HBM gross margins are materially above the corporate average. Watch the HBM revenue figure, the qualification cadence on next-generation HBM and the DRAM bit-shipment guide.
Oracle (ORCL) belongs in this index even though it is a buyer, not a supplier. The company's remaining performance obligation (RPO) has surged to multi-year highs on AI training contracts, and Oracle Cloud Infrastructure (OCI) is now growing faster than the hyperscaler average from a smaller base. The earnings tell is RPO, OCI growth and capex — Oracle's capex is itself a leading indicator for everyone else in this index.
How to use this as an earnings playbook
Each quarter, line the index up against the hyperscaler capex prints. If Microsoft, Meta, Alphabet, Amazon and Oracle collectively raise capex, the index should print stronger orders, longer backlogs and — eventually — higher margins. If capex stalls, the earliest cracks tend to show in the shortest-cycle names: Vertiv orders, Arista cloud-titan commentary, Coherent transceiver guide.
We track these companies on the Earnings Compass calendar with implied-move estimates and post-earnings drift analytics, so you can size around the print rather than guess. For deeper dives, see our guides on how to read an earnings report and the implied move calculator.
Bottom line: the AI infrastructure trade has become a broad, multi-industry earnings cycle. Owning a 25-name basket — power, cooling, fiber, silicon, memory and the hyperscaler buyer — gives you a longer-shelf-life way to express the AI thesis than chasing a single ticker.