AI Infrastructure Stocks 2026: Data Centers, Power, Cooling Pick & Shovel
AI infrastructure 2026: power (CEG, VST), cooling (VRT, ETN), data center REITs (DLR, EQIX), networking (ANET) — under-hyped pick & shovel plays for AI capex.
AI Infrastructure Stocks 2026: Data Centers, Power, Cooling Pick & Shovel
TL;DR
The AI capex boom requires more than chips — it needs electrical power, cooling, networking, and physical real estate. The "pick and shovel" thesis: own the infrastructure providers benefiting from hyperscaler capex without taking direct AI silicon risk. Top names include Constellation Energy (CEG) for nuclear power, Vertiv (VRT) for cooling, Digital Realty (DLR) and Equinix (EQIX) for data center REITs, Arista Networks (ANET) for AI networking, and Eaton (ETN) for electrical equipment. These names trade at lower multiples than NVDA (typically 18–35x forward P/E versus NVDA at 47x) with secular tailwinds from data center buildout. Many investors consider AI infrastructure under-owned versus AI silicon — less hype, smaller crowd, similar capex tailwind. The AI thesis carries elevated valuation risk and hyperscaler capex commentary remains the key signal.
Why AI Infrastructure Matters in 2026
The conventional AI investment narrative focuses on the silicon layer — Nvidia, AMD, Broadcom — and the application layer — Microsoft Copilot, Google Gemini, Meta. Both layers attract enormous attention and command premium multiples. The middle layer — the physical infrastructure that turns silicon into productive compute — has historically received less attention but benefits from the same capex tailwind.
Concrete numbers:
- US data center power demand is projected to grow from ~150 TWh in 2024 to ~400 TWh by 2030 — a near tripling in 6 years
- Hyperscaler 2026 data center capex of approximately $330 billion flows roughly 60% to silicon and 40% to physical infrastructure (real estate, power, cooling, networking)
- Liquid cooling adoption is rising from <5% of new builds in 2023 to projected 35–40% by 2026 as Blackwell racks demand it
- Data center construction backlog in major US markets (Northern Virginia, Dallas, Phoenix) has expanded 4x since 2022
This article surveys the highest-conviction AI infrastructure names across power generation, cooling, data center real estate, and networking. These are pick-and-shovel plays — they win regardless of which AI silicon vendor or foundation model dominates, as long as the capex cycle continues.
Investment Thesis
The AI infrastructure thesis rests on five pillars:
Pillar 1: Power is the binding constraint. Hyperscaler capacity expansion is increasingly limited by grid power availability, not silicon supply. Microsoft has signed PPAs (power purchase agreements) for nuclear restart at Three Mile Island. Amazon acquired a $650M data center campus directly adjacent to a Talen Energy nuclear plant. This is structural, not a temporary squeeze.
Pillar 2: Cooling architecture is changing. Air cooling worked for previous-generation servers but Blackwell racks (GB200 NVL72) generate 120kW per rack — 5–10x previous densities. Liquid cooling adoption is mandatory for new AI builds. Vertiv, Munters, and Schneider Electric are direct beneficiaries.
Pillar 3: Data center REITs are leasing at premium pricing. Digital Realty and Equinix report new lease pricing 30–50% above expiring leases due to AI demand. Vacancy in major markets is below 3%. This pricing power compounds over multi-year leases.
Pillar 4: Networking is increasingly mission-critical. AI training clusters require ultra-low latency, high-bandwidth interconnect. Arista Networks has won the majority of hyperscaler AI Ethernet deployments. Optical components (Coherent, Lumentum) ship in every transceiver.
Pillar 5: Lower valuations than the AI silicon layer. Most infrastructure names trade at 18–35x forward P/E versus NVDA at 47x and software AI plays at 50x+. Many investors consider this a more defensive way to participate in the capex cycle.
The bear case: infrastructure capex is more cyclical than software. If hyperscaler capex pauses, infrastructure orders are cancelled or deferred more easily than software contracts. Power and cooling have long lead times that magnify the swing. The AI thesis carries elevated valuation risk and infrastructure plays are not immune.
Top Picks Breakdown
Power Generation
Constellation Energy (CEG)
| Metric | Value |
|---|---|
| Market cap | ~$90 billion |
| Forward P/E | ~28x |
| Revenue (TTM) | ~$24 billion |
| Nuclear capacity | ~22 GW (largest US fleet) |
| AI customer signed | Microsoft (Three Mile Island restart) |
| EPS growth (3y) | ~25% |
The largest US nuclear power generator and the cleanest AI infrastructure play. The Microsoft PPA to restart Three Mile Island (around 2027–2028) signaled that hyperscalers are willing to underwrite nuclear capacity additions. CEG owns existing nuclear plants and is positioned to sell incremental capacity at premium pricing as AI demand grows.
Vistra Energy (VST)
| Metric | Value |
|---|---|
| Market cap | ~$50 billion |
| Forward P/E | ~22x |
| Nuclear capacity | ~6 GW (post-Energy Harbor acquisition) |
| Combined cycle natural gas | ~30 GW |
Diversified power generator with growing nuclear and natural gas exposure. Less pure than CEG but cheaper. Texas (ERCOT) market exposure provides AI demand growth from the Texas data center build-out.
NextEra Energy (NEE)
| Metric | Value |
|---|---|
| Market cap | ~$160 billion |
| Forward P/E | ~22x |
| Renewables capacity | Largest US wind/solar |
| Florida regulated utility | FP&L (regulated returns) |
Diversified utility with strongest renewable energy buildout. AI demand for clean power positions NEE to monetize wind, solar, and battery storage at premium PPAs. More defensive than pure merchant power names.
Cooling and Electrical
Vertiv Holdings (VRT)
| Metric | Value |
|---|---|
| Market cap | ~$35 billion |
| Forward P/E | ~32x |
| Revenue (TTM) | ~$8 billion |
| AI infrastructure revenue share | ~50% (rapidly growing) |
| EPS growth (3y) | ~35% |
| Liquid cooling | Market leader |
The pure-play AI cooling and power infrastructure name. Vertiv designs and manufactures liquid cooling systems (CDUs, rear door heat exchangers), uninterruptible power supplies, and rack systems for hyperscaler AI deployments. Order growth has accelerated through 2024–2025 as Blackwell racks demand liquid cooling.
Eaton (ETN)
| Metric | Value |
|---|---|
| Market cap | ~$135 billion |
| Forward P/E | ~28x |
| Revenue (TTM) | ~$26 billion |
| Data center revenue share | ~25% (growing) |
| Electrical Americas operating margin | ~24% |
Diversified industrial with significant data center electrical infrastructure exposure (switchgear, busways, transformers, UPS). Less pure than VRT but more diversified across industrial verticals.
Schneider Electric (SU.PA)
| Metric | Value |
|---|---|
| Market cap | ~€140 billion |
| Forward P/E | ~26x |
| Revenue (TTM) | ~€38 billion |
| Listing | Paris (EUR-native for EU investors) |
| Data center exposure | ~20% of revenue |
European peer to Eaton with strong data center electrical and cooling exposure. EUR-listed, no FX cost for EU investors. Acquired Motivair (liquid cooling) in 2024 to deepen AI infrastructure positioning.
Data Center REITs
Digital Realty (DLR)
| Metric | Value |
|---|---|
| Market cap | ~$50 billion |
| FFO multiple | ~22x |
| Dividend yield | ~3.4% |
| Total power capacity | ~3 GW |
| AI bookings (2024) | Record year |
Largest pure-play data center REIT globally. New lease pricing 30–50% above expiring leases driven by AI demand. Power-constrained markets (Northern Virginia, Frankfurt, Singapore) command premium pricing. Tax-efficient REIT structure with growing dividend.
Equinix (EQIX)
| Metric | Value |
|---|---|
| Market cap | ~$80 billion |
| FFO multiple | ~28x |
| Dividend yield | ~2.1% |
| Markets served | 70+ globally |
| Interconnection revenue share | ~20% |
The premium data center REIT focused on interconnection (network exchange, cloud on-ramps). Higher multiple than DLR reflects sticky interconnection revenue model. AI demand drives both raw capacity leasing and interconnection volume.
Networking
Arista Networks (ANET)
| Metric | Value |
|---|---|
| Market cap | ~$130 billion |
| Forward P/E | ~38x |
| Revenue (TTM) | ~$7.5 billion |
| Hyperscaler customer concentration | Top 5 ~50% |
| AI Ethernet share | Majority of new builds |
| Operating margin | ~40% |
The dominant AI Ethernet vendor for hyperscaler networks. Won majority of Microsoft, Meta, Oracle AI fabric deployments. Competes with Cisco (broader portfolio) and Nvidia Spectrum-X (vertically integrated alternative). Customer concentration is the key risk.
Cisco Systems (CSCO)
| Metric | Value |
|---|---|
| Market cap | ~$240 billion |
| Forward P/E | ~16x |
| Dividend yield | ~2.7% |
| AI infrastructure revenue (Splunk + Silicon One) | Growing segment |
The legacy networking incumbent, repositioning around AI through Splunk acquisition and Silicon One custom silicon. Lower growth, much lower multiple than ANET. Defensive AI infrastructure exposure.
Optical Components
Coherent (COHR)
| Metric | Value |
|---|---|
| Market cap | ~$20 billion |
| Forward P/E | ~28x |
| AI optical transceiver share | Material |
| Customer concentration | Hyperscalers and OEMs |
Manufactures optical components (lasers, transceivers) for AI data center networking. Smaller, more cyclical than ANET but pure-play optical exposure.
Valuation Analysis
| Stock | Forward P/E | EPS growth (3y) | PEG | AI revenue share |
|---|---|---|---|---|
| CEG | 28x | 25% | 1.1 | 100% (power for AI) |
| VRT | 32x | 35% | 0.9 | ~50% |
| ETN | 28x | 14% | 2.0 | ~25% |
| SU.PA | 26x | 12% | 2.2 | ~20% |
| DLR | 22x (FFO) | 8% | 2.7 | ~40% lease growth |
| EQIX | 28x (FFO) | 9% | 3.1 | ~30% lease growth |
| ANET | 38x | 22% | 1.7 | ~40% |
| CSCO | 16x | 6% | 2.7 | ~10% |
| COHR | 28x | 25% | 1.1 | ~35% |
Compared to silicon layer: Most infrastructure names trade at 22–32x forward earnings versus NVDA at 47x and AMD at 32x. On growth-adjusted PEG, VRT and CEG screen most attractive — both offer growth comparable to NVDA at lower multiples.
Compared to broader market: Infrastructure names trade at premiums to S&P 500 (22x) but discounts to dedicated AI silicon. Many investors consider this the appropriate "second derivative" exposure for the AI thesis.
Historical context: Pre-AI capex cycle (2019–2022), most of these names traded at 15–22x forward earnings. Multiples have re-rated 30–50% as the AI thesis took hold. Whether multiples sustain depends on capex continuation. If hyperscaler capex pauses, infrastructure names will derate faster than NVDA because their revenue concentration to AI capex is similar but their valuation cushion is thinner.
EU Investor Access
Most names are accessible via UCITS structures or direct US listings:
Direct US stock access:
- CEG, VST, NEE, VRT, ETN, DLR, EQIX, ANET, CSCO, COHR — all listed in US, available via XTB, Trading 212, IBKR, Trade Republic, Saxo
- W-8BEN form reduces dividend withholding from 30% to 15%
EUR-listed direct access:
- Schneider Electric (SU.PA): Paris listing, no FX cost, French dividend withholding 12.8% with treaty
- Siemens (SIE.DE): German listing, partial AI infrastructure exposure via electrical and automation
UCITS ETF exposure:
- iShares Global Infrastructure UCITS (INFR / IE00B1FZS467): Broad infrastructure including utilities and pipelines
- SPDR Morningstar Multi-Asset Global Infrastructure UCITS: Similar approach
- iShares Listed Private Equity UCITS — adjacent (data center developers)
- Invesco Real Estate UCITS — partial exposure to data center REITs
There is no pure-play "AI infrastructure" UCITS ETF as of 2026. EU investors targeting this thesis must build it via direct stocks or use partial exposure through broad infrastructure and tech ETFs.
Tax efficiency: REITs (DLR, EQIX) require attention — US REIT dividends have specific tax treatment that differs from regular dividends. Polish IKE/IKZE accounts shelter capital gains but withholding rules at source still apply.
Real-World Example Portfolio
A €100,000 portfolio with AI infrastructure tilt for a moderate EU investor seeking AI exposure with less concentration risk than direct silicon names:
| Position | Allocation | Amount | Rationale |
|---|---|---|---|
| iShares Core MSCI World (IWDA) | 40% | €40,000 | Global core |
| Constellation Energy (CEG) | 8% | €8,000 | Pure power-for-AI play |
| Vertiv (VRT) | 6% | €6,000 | Pure cooling pure-play |
| Arista Networks (ANET) | 5% | €5,000 | AI networking leader |
| Digital Realty (DLR) | 5% | €5,000 | Data center REIT, dividend yield |
| Schneider Electric (SU.PA) | 4% | €4,000 | EUR-listed electrical |
| Equinix (EQIX) | 3% | €3,000 | Premium data center REIT |
| Eaton (ETN) | 3% | €3,000 | Diversified electrical |
| iShares MSCI Semiconductors (SEMI) | 8% | €8,000 | Silicon layer exposure |
| Vanguard FTSE All-World (VWCE) | 13% | €13,000 | Diversification |
| Cash / short bonds | 5% | €5,000 | Dry powder |
Combined AI infrastructure exposure: 34% across power, cooling, networking, and REITs. Adding SEMI brings effective AI thesis exposure to ~42% of portfolio while diversifying away from NVDA-specific concentration. This portfolio benefits from AI capex without single-stock binary risk on Nvidia.
Risk Factors
Hyperscaler capex pause. The single biggest risk. If Microsoft, Amazon, Google, or Meta materially cut 2026 or 2027 AI capex guidance, infrastructure orders soften within 1–2 quarters. Watch quarterly capex commentary closely.
Customer concentration. ANET's top 5 customers represent ~50% of revenue. VRT has similar concentration in hyperscalers. CEG's Microsoft PPA is a single-customer concentration. Loss of any major contract would meaningfully impact earnings.
Power transmission constraints. Even with generation capacity available, US transmission grid constraints could delay data center power delivery by 2–4 years in some markets. This affects realized growth more than thesis validity but extends payback timelines.
Cooling technology risk. Liquid cooling adoption is mandatory for high-density AI but technology choice (direct-to-chip, immersion, rear-door heat exchanger) is not yet settled. Vertiv covers multiple architectures; competitors may emerge with superior solutions.
Regulatory risk. US Inflation Reduction Act subsidies for clean power (CEG, NEE, VST) face ongoing political risk. EU electricity market reforms could affect Schneider Electric's regulated returns. Policy reversals are not zero-probability.
Real estate concentration. Data center REITs depend on a few major markets (Northern Virginia, Frankfurt, Singapore, Tokyo). Local regulatory action (water consumption limits, zoning, grid moratoriums) can impact specific REIT submarkets disproportionately.
Long lead times cut both ways. Power and cooling have 2–4 year lead times. This protects pricing during shortages but means oversupply can persist if demand softens, magnifying the cycle downside.
Multiple compression. Infrastructure multiples have re-rated 30–50% on the AI thesis. A reversal of investor enthusiasm — without earnings missing — could compress multiples by 20–30%.
The AI thesis carries elevated valuation risk across the entire infrastructure complex.
Time Horizon Considerations
Short-term (0–12 months): Quarterly capex commentary from hyperscalers drives sector sentiment. Earnings are less binary than NVDA but the sector trades together. Limited tactical edge for non-professional traders.
Medium-term (1–3 years): This is the AI infrastructure capex window. Power buildouts, cooling deployments, REIT re-leasing, and networking upgrades all play out. Most thesis returns are realized in this period. Position-sizing decisions in 2025 determine 2027–2028 outcomes.
Long-term (3–10 years): Data center power demand is projected to keep growing through 2030 even after the immediate AI capex peak. Power, cooling, and REITs have multi-decade asset lives. Networking technology refresh cycles continue. Many investors consider infrastructure a more defensible 10-year hold than pure AI silicon.
The most defensible AI infrastructure holding period is 3–7 years with active monitoring of hyperscaler capex commentary as the primary thesis indicator.
FAQ
Q: Why invest in AI infrastructure instead of Nvidia directly? A: Lower entry multiples (22–32x vs 47x), less single-customer concentration risk, diversification across power, cooling, networking, and real estate. Many investors consider infrastructure a more defensible way to participate in the AI capex thesis. Trade-off: lower upside if NVDA continues to dominate value capture.
Q: What is the most pure-play AI infrastructure stock? A: Vertiv (VRT) for cooling and Constellation Energy (CEG) for power are the cleanest pure-plays. Both have ~50%+ revenue exposure to AI data center demand and trade at attractive growth-adjusted multiples.
Q: Are data center REITs (DLR, EQIX) good AI plays? A: Yes, with caveats. New lease pricing is rising 30–50% above expiring leases driven by AI demand. But existing leases roll over slowly (5–10 year terms), so lease re-pricing benefits compound over years rather than quarters. Lower beta than direct AI silicon names.
Q: Can EU investors buy these names easily? A: Yes. All US-listed names available via XTB, Trading 212, IBKR, Trade Republic. Schneider Electric available natively in EUR via Paris listing. No pure-play UCITS AI infrastructure ETF exists yet.
Q: How do I track AI infrastructure positions across multiple sectors? A: For consolidated multi-broker, multi-sector tracking with sector tagging (power, cooling, REIT, networking), Freenance supports custom asset categorization and unrealized P&L tracking — useful when building diversified AI infrastructure exposure across many names.
Related Articles
Want full control over your finances?
Try Freenance for free