Top 10 Picks-and-Shovels Stocks for the AI Buildout
The public companies cashing the hyperscaler capex checks, ranked by sector criticality and conviction.
The public companies cashing the hyperscaler capex checks, ranked by sector criticality and conviction.
Published May 16, 2026. The one-year returns below are as of that date.
Disclosure: I hold low-single-digit-percent positions in several names below (VRT, IREN, CRWV, BE) per the house rules. This is the list of names I follow, not a model portfolio, and I update it quarterly. 1Y returns last refreshed 2026-05-16 against /api/watchlist; for live values see the live watchlist.
When there's a gold rush, sell shovels. The AI buildout's hyperscaler customers (Amazon, Google, Meta, Microsoft, Oracle) will combine to spend more than $675 billion on capex in FY2026 alone (the full ranking is here). Most of that money flows out the door to a relatively narrow group of public infrastructure suppliers: power equipment makers, cooling specialists, electrical contractors, nuclear utilities, and one weird fuel-cell company.
The ten names below are the highest-conviction picks-and-shovels public stocks we track. Ranked first by sector criticality (which layer of the buildout is most supply-constrained, per the supplier map), then by company-specific exposure. One-year total returns are listed for sanity check, not as the ranking criterion. The dispersion is wide on purpose: some of these names already had their re-rating (ETN, CEG) and are now compounding off a high base (or, in CEG's case, partially unwinding it); others (POWL, MOD, NVT) are smaller-cap pure plays that ran harder. The list is what I'd own, not what's been the best trade.
The top 10
1. Vertiv ($VRT): cooling and power management
1Y return: +250% | Layer: Cooling, UPS, thermal management
The single most-discussed name in the AI infrastructure trade for a reason. Full liquid cooling stack, UPS systems, and the most direct exposure to the rack-density-tripled-in-four-years problem. Margins expanding every quarter; backlog visibility through 2027. If I had to own one supply-chain name, this would be it.
2. Eaton ($ETN): electrical infrastructure
1Y return: +22% | Layer: UPS, switchgear, backup power
Total backlog over $19B, with Electrical Americas alone at \~$10B and data-center orders up roughly 200% in Q3 2025 (record-quarter release). The clearest pure-play on data center electrical equipment, and the layer that's bottlenecked by 128-week transformer lead times. The 1Y return looks downright modest because most of ETN's re-rating happened in 2024; the next leg is earnings catching up to multiple, not multiple expansion.
3. Constellation Energy ($CEG): nuclear baseload
1Y return: -8% | Layer: Nuclear utility
Nuclear fleet, signed hyperscale PPAs at premiums to grid rate. The cleanest 24/7 baseload solution to the AI power problem. Hyperscalers have been signing decade-plus PPAs at $80+/MWh against a wholesale grid rate of \~$45. The 2024 re-rating already happened, and has partially unwound in the last six months. The trade from here is PPA economics flowing through earnings, not another multiple re-rate. The fundamentals haven't broken; the multiple did.
4. Bloom Energy ($BE): fuel cells
1Y return: +1,272% (\~13x, the outlier) | Layer: On-site generation
Up roughly 13x in the past year on AI data center orders. Oracle's master services agreement covers up to 2.8 GW of Bloom solid-oxide systems for US AI and cloud sites. The trade with the highest catalyst risk on the list (if grid lead times collapse, this rolls over fast) but also the cleanest expression of "hyperscalers cannot wait for utilities." Worth owning small; do not chase.
5. GE Vernova ($GEV): grid-scale electrical
1Y return: +145% | Layer: Transmission, turbines, grid
The utility-side counterpart to Eaton. Sells the high-voltage equipment that connects new substations to the grid, plus gas turbines for capacity additions. Beneficiary of every regional grid reinforcement project the AI buildout has triggered.
6. Quanta Services ($PWR): electrical construction
1Y return: +124% | Layer: EPC, transmission, site work
The most levered to power build-out specifically, of the major US construction firms we track. Builds the transmission lines and substations that feed every new data center site.
7. Talen Energy ($TLN): nuclear utility
1Y return: +37% | Layer: Nuclear utility
Susquehanna nuclear, the much-discussed Amazon PPA. Sharper exposure than CEG to a single deal, which is good or bad depending on whether the deal closes on the original terms. The FERC review has been a recurring catalyst, and the modest 1Y return reflects that overhang more than the operating story.
8. nVent Electric ($NVT): electrical pure-play
1Y return: +150% | Layer: Electrical, liquid cooling
A smaller-cap pure play in the electrical and liquid-cooling layers. Less analyst coverage than ETN, which means more idiosyncratic moves when the data shows up.
9. Modine Manufacturing ($MOD): thermal management
1Y return: +160% | Layer: Cooling, HVAC
Thermal management specialist; the smaller-cap pure play in cooling next to Vertiv. It earned a spot this cycle as the rack-density story has gotten more obvious.
10. Powell Industries ($POWL): switchgear
1Y return: +381% | Layer: Custom switchgear, electrical
Smallest cap of the cohort, most levered to new-build capex. The order book reads as a leading indicator for the broader power equipment cycle. The strongest 1Y return on the list, which is what happens when a small-cap pure play meets a multi-year demand surge in its core product.
What's not on this list (and why)
A few absences worth flagging:
- No hyperscalers. AMZN, GOOGL, META, MSFT, ORCL are the customers, not the suppliers. The thesis is that suppliers have priced-in less and have more room to run. The 1Y dispersion above is consistent with that: even the names that have unwound (CEG, ETN) sit at much higher absolute returns than the hyperscaler customers did over the same window.
- No chip fabs. TSM, Samsung, SK Hynix and Micron are left off on purpose. They have their own dynamics (cyclical, geographically concentrated, capital-intense in a different way) that aren't a clean read on data center build velocity.
- No GPU vendors. NVDA is not in the picks-and-shovels list because it sits above the supply chain, not in it. A different trade entirely.
How to actually act on this
The fastest way to track all ten as a unit: the live picks-and-shovels watchlist: same names, same order, with current prices, 1D, 1Y, FY capex, and a per-ticker trade button into Public.com (the brokerage we use for these names; they support the full list including the smaller-cap pure plays like POWL and MOD).
To see where these names actually get the work, the campus directory is the other half of this piece: each project page lists the companies tied to that site, the capacity figure, and who published it. How we build and check it.
The mental model worth holding: hyperscaler capex is a $675B line item this year. Each name on the list above gets some narrow slice of it. The investable question is which names get the durable slice, the kind that compounds for five years rather than spiking on one cycle. That's a separate piece, coming next quarter.