Data center power and cooling
AI training racks consume five to ten times the power of traditional servers and require dense liquid cooling, dedicated transmission upgrades, and massive site development. This theme groups the companies actually building, powering, and cooling the AI data center footprint.
How does data center cooling work?
Data center cooling removes the heat that servers generate so the chips stay within safe operating limits. Traditional facilities used air cooling: computer room air handlers push cold air across racks and exhaust the hot air out. That worked when a rack drew a few kilowatts. AI training racks are far denser (an NVL72-class rack can pull 100kW or more), and air alone can no longer carry that much heat away, so operators move to liquid cooling and direct-to-chip cold plates that run coolant straight across the processors, plus rear-door heat exchangers and immersion in some designs.
This shift is the core mechanism behind the data center cooling side of the theme. Vertiv (VRT) sells exactly this stack: liquid cooling units, coolant distribution, high-density power distribution, and integrated thermal management built for AI rack densities, which is why its product mix tracks the move from air to liquid so closely. Corning (GLW) benefits on an adjacent axis, because denser, hotter campuses still need high-count optical fiber to interconnect all those racks. The simple rule of thumb: the more power a rack draws, the more the data center cooling problem shifts from air to liquid, and the more equipment the named beneficiaries supply per megawatt built.
How does data center power flow from the AI chip to the grid?
Data center power is a chain that runs from the utility grid all the way down to the processor. A hyperscaler campus connects to high-voltage transmission, steps voltage down through substations and switchgear, distributes it through busways and power distribution units inside the building, and finally feeds the racks through uninterruptible power supplies that smooth out any interruption. Every link in that chain has to scale up together when an AI campus is sized in the hundreds of megawatts or gigawatts rather than the tens.
That is where the contractors and equipment makers in the data center power and cooling theme come in. Quanta (PWR) builds the transmission lines and substations that utilities add to serve new AI load. Comfort Systems (FIX) and EMCOR (EME) do the mechanical, electrical, and plumbing work inside the buildings, wiring the power and cooling systems together. Sterling (STRL) prepares the site itself: earthwork, foundations, and utility tie-ins before any equipment arrives. WESCO (WCC) distributes the wire, cable, switchgear, and busways those crews install. Each one earns revenue per facility built, so a faster data center power buildout flows directly into their backlogs.
Why do AI data centers need so much more power than traditional ones?
AI workloads concentrate enormous compute into a small footprint. A conventional CPU server rack might draw a handful of kilowatts, while AI accelerator racks pack many high-power GPUs together and draw many times more, often an order of magnitude higher per rack. Multiply that across thousands of racks in a single campus and the result is data center power demand measured in hundreds of megawatts or full gigawatts, comparable to a small city. Because the chips run hot and constantly during training, the load is also dense and sustained rather than intermittent.
This demand does not stay inside the building. To serve a new gigawatt-scale campus, a utility often has to build fresh transmission, add substation capacity, or contract new generation, which lifts utility capital spending and the contracting work that supports it. That is the structural link from AI compute to the data center power and cooling theme: the chips create the load, the load forces grid and on-site upgrades, and the contractors (PWR, FIX, EME, STRL) and distributors (WCC) get paid to build them. Vertiv (VRT) sits at the rack end of the same chain, supplying the power and cooling gear that turns delivered electricity into a safe, dense computing environment.
How does data center power and cooling work as a single buildout?
On a real AI campus, data center power and cooling are designed and constructed together, not as separate projects. A developer secures land and grid access, a contractor like Sterling (STRL) prepares the site, utilities and specialists like Quanta (PWR) bring in transmission and substations, and MEP contractors like Comfort Systems (FIX) and EMCOR (EME) install the electrical distribution and the cooling loops in one coordinated build. WESCO (WCC) supplies the electrical materials throughout, and Vertiv (VRT) provides the integrated power and cooling equipment that lands in the data hall.
Framing it this way shows why the data center power and cooling theme spans several stock types rather than one. No single company covers the full stack from grid to chip, so each named beneficiary captures a different slice of the same campus budget. Industrial gas suppliers Linde (LIN) and Air Products (APD) sit one step upstream, supplying the specialty gases used in the fabs that make AI accelerators, which is why they appear as adjacent exposure rather than direct construction beneficiaries.
What gets a stock into the Data center power and cooling theme?
Hardware, contracting, or materials that scale directly with AI data center buildouts: power and cooling equipment, electrical infrastructure construction, MEP contracting, site development, optical fiber, industrial gases at hyperscale fabs and DCs.
Data center power and cooling stocks
Every public name that fits the Data center power and cooling thesis, with the rationale for inclusion. Click any ticker for the full stock guide. The basket above starts equal-weighted; you set your own target weights inside Walnut.
UPS, PDUs, liquid cooling, and integrated thermal management for AI training racks. Pure-play AI infrastructure beneficiary.
Largest specialty electrical infrastructure contractor in North America. Transmission and substation work for AI-driven utility load growth.
Mechanical, electrical, and plumbing contracting with specialty data center expertise; backlog from hyperscaler customers has grown materially.
Large MEP contractor winning hyperscaler data center buildouts alongside Comfort Systems.
E-Infrastructure Solutions specializes in data center site development (earthwork, foundations, utilities) for hyperscalers.
Distributes the wire, cable, switchgear, busways, and datacom infrastructure that data center construction consumes at scale.
Industrial gases, including the specialty electronics gases used in fabs that produce AI accelerators.
Industrial gases and large-scale hydrogen projects; semiconductor and data center customers consume meaningful share of revenue.
Largest optical fiber manufacturer; AI data center interconnect drives Optical Communications segment growth.
How to invest in Data center power and cooling
There are a few practical ways to get exposure to data center power and cooling. The most concentrated is a small pair of leaders, for example Vertiv (VRT) for rack-level power and cooling equipment alongside Quanta (PWR) for the transmission and contracting side. That is simple to hold, but the two names tend to move together and leave out the rest of the build. If you prefer something passive, the closest ETF proxies in 2026 are XLU (utilities) and XLI (industrials), and both are diluted: they hold their data center beneficiaries at small weights inside hundreds of unrelated names, so neither is a clean expression of the theme. No pure-play data center power and cooling ETF exists yet, because index providers have not defined a tight enough rule set around it.
The alternative is to build a data center power and cooling basket in Walnut. You describe the theme to Walnut's AI assistant (for instance, "data center power and cooling beneficiaries"), and it proposes constituents and starting weights spanning equipment (VRT), grid and contracting (PWR, FIX, EME, STRL), distribution (WCC), and adjacent industrial gases (LIN, APD) and fiber (GLW). You review the rationale, adjust the weights, and fund the basket through your existing connected broker. Walnut never trades for you: you approve every order before it is placed, and the data center power and cooling basket then tracks as a single performance line you can compare against XLU or XLI.
ETFs used as passive proxies for Data center power and cooling
If you want the theme as a single ticker rather than as a basket, these are the ETFs people most commonly use. Each has trade-offs (concentration, expense ratio, sector overlap) covered in the individual ETF guides.
FAQ
What is the data center power and cooling theme?
+
Data center power and cooling covers the companies that build, electrify, and thermally manage the physical buildings that AI runs in. That's electrical contractors (Quanta, Comfort Systems, EMCOR), site developers (Sterling), UPS and liquid cooling equipment makers (Vertiv), electrical distributors (WESCO), optical fiber for interconnect (Corning), and the industrial gases used in adjacent fab and DC operations (Linde, Air Products). Picks-and-shovels for the picks-and-shovels.
Why is AI driving data center power demand?
+
AI training racks consume 5 to 10 times the power of traditional CPU server racks. A single NVIDIA H100 server pulls ~10kW; an NVL72 GB200 rack is 100kW+. Hyperscalers planning 1-5 GW campuses now have to either build new transmission, contract new generation, or co-locate next to existing power. The economic effect is a multi-year upcycle in utility capex, transmission contracting, and on-site cooling and power equipment.
Which stocks are in the data center power theme?
+
Walnut groups 9 names as of early 2026: VRT (Vertiv, power and cooling equipment), PWR (Quanta, electrical contracting), FIX (Comfort Systems, MEP), EME (EMCOR, MEP), STRL (Sterling, site development), WCC (WESCO, distribution), LIN (Linde, industrial gases), APD (Air Products, gases), GLW (Corning, fiber).
What's the biggest data center power stock?
+
Vertiv (VRT) is the most concentrated pure-play in the theme: UPS, busways, PDUs, and liquid cooling specifically optimized for AI rack densities. Quanta Services (PWR) is bigger by market cap but its exposure is broader (grid + renewables + AI), and it's the typical contractor when utilities expand transmission to serve new AI campuses.
What ETFs cover data center power?
+
No clean pure-play ETF exists for the theme. XLU (Utilities sector) and XLI (Industrials sector) get partial exposure. Some thematic infrastructure ETFs like GRID and PAVE include Quanta and Vertiv but at low weights. A Walnut basket of 5-7 names is tighter on the thesis than any ETF can be without rule-set changes.
How do I invest in data center power?
+
Two practical paths. (1) Buy VRT and PWR as a two-stock concentrated bet, simple but tightly correlated. (2) Build a Walnut basket spanning equipment (VRT), contracting (PWR, FIX, EME, STRL), distribution (WCC), and industrial gases (LIN), which diversifies execution risk while preserving the theme. Most users do option 2 because no individual name spans the entire stack.
What's the difference between data center power and AI infrastructure?
+
AI infrastructure is the silicon and software stack: chips, networking, foundries, cloud platforms. Data center power is the physical building and grid stack: power equipment, cooling, contracting, electrical distribution, fiber. They're sequential layers. A new GW-scale AI campus consumes both: NVDA H100s plus VRT cooling plus PWR substations plus GLW fiber. Walnut treats them as separate themes so users can size each independently.
Is data center power a good investment in 2026?
+
Walnut isn't an investment adviser. Factually, multi-quarter backlog at VRT, PWR, FIX, EME has visibly grown through 2024-2025 on hyperscaler commitments, and US utility capex plans have stepped up materially. Valuations on the most-loved names (VRT, PWR) have expanded with the thesis; concentrations of expectation are real. Risk: any hyperscaler capex pause shows up here within 1-2 quarters.
What are the risks of a data center power basket?
+
Three. (1) Hyperscaler capex concentration: ~70% of recent backlog growth has come from 5 customers; budget revisions move the complex. (2) Margin compression from labor and copper inflation: contracting margins are sensitive to these inputs. (3) Sentiment: the entire theme rerated in 2024-2025; multiple compression on AI capex disappointment is the asymmetric risk now versus 18 months ago.
Why is Vertiv such a popular AI play?
+
Vertiv is one of the few public companies whose product mix is explicitly tilted toward what AI data centers buy. Liquid cooling, high-density PDUs, integrated thermal management, and high-availability UPS systems are essentially the AI rack stack. Backlog grew several multiples between 2022 and 2025; the company is among the clearest revenue-pulls from hyperscaler AI capex commitments.
Is data center power cyclical?
+
Underneath the AI wave, yes. Equipment cycles can swing 30%+ peak-to-trough; contracting margins are sensitive to labor inflation; utility capex is regulated and gated by rate cases. The current up-cycle has been driven by structural AI demand, but pretending data center capex grows in a straight line through 2030 has been wrong in every prior infrastructure boom.
Can I build a data center power basket in Walnut?
+
Yes. Tell Walnut's AI assistant something like 'data center power and cooling beneficiaries' and it proposes a 5-7 stock basket usually anchored by VRT, PWR, FIX, with optional adds from STRL, EME, WCC, LIN. You set the weights, review the rationale, and fund through your broker. The basket tracks as one line so you can compare it to XLU or XLI cleanly.
What ETF should I buy if I just want one ticker?
+
There isn't a clean single ticker for this theme. The closest passive proxies are GRID (Nasdaq Clean Edge Smart Grid) and PAVE (Global X US Infrastructure Development), but both dilute the data center exposure with broader infrastructure names. The structural reason there's no AI data center ETF: the index providers haven't yet defined a tight enough rule set.
Build the Data center power and cooling basket in Walnut
Walnut's AI assistant takes the thesis above, proposes 5 to 6 constituents with target weights, and lets you fund the basket through your existing broker. You approve every order; we never trade on your behalf.
Other themes
- AI infrastructure. Picks and shovels of the AI buildout: GPUs, networking, foundries, and the software platforms training the largest models.
- Semiconductors. The full chip stack: designers, foundries, equipment makers, materials suppliers, and packaging specialists.
- Defense and modernization. Software, sensors, and specialty materials at the center of US and allied defense buildouts.
- Critical materials. Rare earths, specialty metals, and strategic materials at the center of supply chain reshoring.
- Dividend growth. Companies that compound a growing dividend through cycles. The boring core of many long-term portfolios.
Walnut is informational, not investment advice. Theme membership is descriptive, not prescriptive; nothing on this page should be read as a recommendation. Always verify current financials and your own circumstances before investing.