June 14, 2026 · 13 min read
AI data centers are the fastest-growing electricity consumers in history. The companies supplying that power — generators, grid equipment makers, utilities, and infrastructure REITs — are compounding at rates that would make most software investors jealous. This is the comprehensive guide to the best power stocks for 2026.
The AI buildout is not just a semiconductor story — it is one of the largest electricity consumption events in modern history. Three structural forces have converged to create a genuine power crisis that will take a decade to fully resolve.
A traditional server rack in a 2015 data center consumed 5-10 kW of power. The Nvidia GB200 NVL72 — the flagship AI compute rack shipping in 2025-2026 — consumes 120 kW per rack. Hyperscalers are deploying thousands of these racks per facility. A single large AI training cluster can draw 50-100 MW of continuous power, roughly equivalent to the electricity consumption of 50,000 homes. New data center campuses under construction today are planned at 500 MW to 1+ GW of total capacity — matching the output of mid-size power plants.
Microsoft, Google, Amazon, and Meta combined spent over $200B on capital expenditure in 2025, the majority going to data center construction, servers, and power infrastructure. Each company has publicly committed to enormous energy procurement targets: Amazon has contracted over 10 GW of renewable energy capacity, making it the largest corporate renewable energy buyer in history. Google committed to 24/7 carbon-free energy at all data centers by 2030 and has signed 7+ GW in new power purchase agreements. Microsoft signed a 20-year PPA to restart Three Mile Island with Constellation Energy and has committed 3 GW of clean energy for 2025-2026 alone.
New power plants and transmission lines in the US face environmental review, interconnection queue backlogs, and permitting processes that routinely take 5-7 years from application to operation. FERC's interconnection queue currently has over 2,600 GW of pending generation projects — more than double the entire installed US generation capacity. Most will never be built, but the queue itself creates delays for legitimate projects. Transformers and high-voltage switchgear face 2-4 year manufacturer lead times. The physical infrastructure to serve new data center demand simply cannot be built as fast as the hyperscalers can deploy compute.
The AI power theme plays out across four distinct investment segments — each with different risk profiles, valuation methods, and catalysts. A layered portfolio approach across all four provides diversification within the theme.
The most direct way to play AI power demand is through electricity generators that can sign large, long-term power purchase agreements (PPAs) directly with hyperscalers. Nuclear generators are particularly advantaged because hyperscalers will pay a significant premium for carbon-free, dispatchable 24/7 power.
The premier AI power trade. Constellation operates the largest nuclear fleet in the US — 32 GW of largely carbon-free, dispatchable baseload power. Signed a 20-year PPA with Microsoft to restart Three Mile Island (Crane Clean Energy Center), creating a template that every other hyperscaler has since tried to replicate. CEG has a 10 GW+ PPA pipeline for data center customers and management has guided to material revenue uplift from contracted power premiums. The stock tripled from 2022 lows on AI power demand. Nuclear license renewals extending plant life to 80 years are the next major catalyst.
Vistra completed the Energy Harbor acquisition to expand its nuclear fleet to include the Perry, Beaver Valley, and Davis-Besse plants in Ohio and Pennsylvania. The combined company now operates both nuclear (for clean baseload) and natural gas (for peaking), giving hyperscalers a one-stop power solution. Vistra's Texas ERCOT market operations and aggressive capital return program (buybacks + dividends) distinguish it from CEG. Vistra is the second pure-play AI power generator, slightly more asset-diversified than CEG with better near-term free cash flow generation.
NRG is developing differentiated behind-the-meter power solutions for AI data centers and corporate campuses — contracting power directly on-site rather than through the grid, which eliminates transmission costs and interconnection queue delays. NRG's retail electricity business (Reliant, Green Mountain Energy) provides a stable earnings base while the data center solutions business is the growth vector. Smaller nuclear exposure than CEG or VST but significant natural gas generation capacity in Texas and the Northeast.
Before a single kilowatt of AI power reaches a data center, it passes through transformers, switchgear, circuit breakers, and miles of transmission cabling made by a small number of specialized manufacturers. These companies are experiencing demand that structurally exceeds supply — backlog growth is the key metric to watch.
Hubbell makes the distribution transformers, wiring devices, and grid automation equipment that undergird the US electrical grid. The company's Power Systems segment is directly exposed to utility spending on grid upgrades driven by data center load growth. Distribution transformers — the critical last-mile equipment — are on 18-24 month wait times as domestic production capacity cannot keep up with replacement demand plus new data center load additions. Hubbell's pricing power has improved dramatically as utilities accept higher costs to secure limited supply.
Eaton is the dominant maker of the electrical switchgear, power distribution units, uninterruptible power supplies, and large power transformers that sit inside every large data center and utility substation. With a $12B+ order backlog and transformer lead times stretching 2-3 years for large units (400 MVA transformers have 2-4 year wait times), Eaton has exceptional revenue visibility. The Electrical Americas segment — the direct AI data center play — now represents over 60% of total revenue and is growing 20%+ annually. Eaton has pricing power it has not had in decades.
Quanta Services is the largest specialty contractor for electric power infrastructure in North America — building and maintaining transmission lines, substations, and distribution networks. FERC Order 1920's mandate for long-range transmission planning directly expands Quanta's addressable market. The company has a $32B+ backlog, the majority in electric power, and its workforce of 50,000+ skilled linemen represents a moat that cannot be replicated quickly. Grid operators are beginning multi-year transmission buildout programs that Quanta will execute.
| Ticker | Primary Product | Order Backlog | Rev Growth | Fwd P/E | DC Exposure |
| HUBB | Distribution transformers | Record highs | +19% | ~25x | High (transformer supply constraint) |
| ETN | Switchgear, large transformers, UPS | $12B+ | +22% | ~28x | Very high (inside every hyperscaler DC) |
| PWR | Transmission construction | $32B+ | +18% | ~24x | High (builds grid to serve data centers) |
Regulated electric utilities are not typically thought of as growth stocks — but the utilities serving regions with high concentrations of data centers are experiencing load growth they have not seen in decades. For regulated utilities, load growth translates directly to rate base expansion, which drives regulated earnings growth. The best utility plays are those with the highest data center load as a percentage of total load — giving them the most earnings upside from the AI buildout.
AEP serves Ohio, Texas, Virginia, and the Carolinas — all major data center markets, particularly the Northern Virginia (Loudoun County) corridor which is the largest data center hub on earth. AEP has guided to 50%+ load growth from data center customers over the next 5 years, which is extraordinary for a regulated utility. The company is investing $43B+ in infrastructure over 2024-2028 to accommodate this growth, which directly grows its regulated rate base and thus its allowed earnings. Data center load now represents ~18% of total AEP load.
FirstEnergy serves Ohio, West Virginia, Pennsylvania, New Jersey, and Maryland — a region attracting significant data center investment due to available land, fiber infrastructure, and proximity to the Washington DC and New York corridors. FE's data center load has grown from under 5% to approximately 12% of total load since 2022 and is accelerating. The company's transmission business (which is less regulated and carries higher allowed returns) is growing at 15%+ annually as grid upgrades to serve new data center loads are approved.
PPL serves Pennsylvania and Kentucky, two states actively courting data center investment with competitive power rates and land availability. PPL's Pennsylvania operations (PPL Electric) are seeing data center interconnection requests more than double from 2023 levels. The company's recent strategic pivot to focus entirely on regulated utilities (after selling UK operations) simplifies the story and improves earnings quality.
| Ticker | Service Territory | DC % of Load | 5-Yr EPS Growth Guide | Dividend Yield | Key Market |
| AEP | OH, TX, VA, Carolinas | ~18% | 6-8% | ~3.8% | Northern Virginia corridor |
| FE | OH, WV, PA, NJ, MD | ~12% | 6-8% | ~4.0% | Mid-Atlantic / Ohio DC growth |
| PPL | PA, KY | ~9% | 5-7% | ~3.4% | Pennsylvania emerging DC market |
| NEE | FL + national renewables | ~7% | 10%+ | ~2.8% | FPL grid + NextEra Energy Resources PPAs |
Data center REITs are the physical facilities where AI compute lives. The most sophisticated operators have pivoted from simple square-footage leasing to power-constrained capacity pricing — the key metric is megawatts leased and preleased, not square feet. In markets with constrained power (Northern Virginia, Silicon Valley, Singapore, Amsterdam), power availability is the primary gating factor for new supply, which creates strong pricing power for incumbents.
Equinix is the world's largest data center REIT by revenue, operating 260+ colocation facilities across 71 metros globally. The interconnection model — where thousands of networks, clouds, and enterprises meet in the same physical location — creates switching costs that make Equinix's premium pricing sustainable. AI workloads are driving record demand for Equinix's xScale hyperscale product and its standard IBX colocation. In constrained markets like the DC metro and Silicon Valley, Equinix's existing facilities command premium power pricing that new entrants cannot easily replicate.
Digital Realty focuses on large hyperscale wholesale facilities leased to Microsoft, Google, Meta, and Oracle. Less interconnection-intensive than Equinix — DLR sells raw capacity (power, cooling, floor space) to hyperscalers building their own environments. DLR's JV structure (PlatformDIGITAL) allows it to develop capacity alongside customer commitments, reducing speculative risk. The lower multiple vs Equinix reflects the more commodity-like wholesale model — but AI buildout is pushing both companies' leasing pipelines to record levels.
Iron Mountain has successfully pivoted from physical records storage (its legacy business) to data centers, operating 100+ MW of data center capacity with a multi-GW development pipeline. The records storage business provides steady cash flows that fund data center development — a self-funding growth model. IRM's data center segment is growing 30%+ annually and is expected to become the majority of revenue by 2027. The combination of defensive cash flows and high-growth data center development makes IRM a uniquely positioned REIT.
A surprising sub-theme within AI power infrastructure is former Bitcoin mining companies. These firms built large, high-powered facilities with dedicated power contracts — exactly the infrastructure needed for AI GPU compute. As Bitcoin mining profitability compressed post-halving, the most sophisticated operators pivoted their power infrastructure to AI and high-performance compute (HPC) workloads at dramatically higher revenue per megawatt.
IREN (Iris Energy) built large renewable-powered Bitcoin mining operations in North America. Following the Bitcoin halving, management pivoted to AI/HPC GPU cloud hosting, leveraging the same 100% renewable power contracts and high-density electrical infrastructure. IREN's facilities in British Columbia and Texas are now hosting Nvidia H100/H200 clusters for AI workloads at $2-3/GPU-hour — a revenue profile dramatically better than Bitcoin mining at current prices. The pivot is still in early stages but represents the strategic direction for the business.
MARA is the largest publicly traded Bitcoin miner by hashrate, operating at 50+ EH/s. While primarily still a Bitcoin miner, MARA is exploring AI and HPC compute at its large-scale facilities with dedicated power. The company operates power-heavy infrastructure in Texas and North Dakota that could support GPU clusters. MARA is primarily a Bitcoin investment, but its power infrastructure creates optionality to partially pivot if Bitcoin mining margins compress further.
For investors who want broad exposure to the AI power theme without individual stock selection, three ETFs cover different slices of the market. Each has distinct characteristics in terms of holdings, cost, and thematic purity.
| ETF | Full Name | YTD Return | Expense Ratio | Top Holdings | Best For |
| GRID | First Trust NASDAQ Clean Edge Grid & Infrastructure | +34% | 0.57% | ETN, HUBB, PWR, VST, CEG | Pure grid infrastructure play |
| AMPS | Pacer American Energy Independence ETF | +28% | 0.75% | CEG, VST, NRG, utilities | US energy independence focus |
| ICLN | iShares Global Clean Energy ETF | +18% | 0.41% | NEE, ENPH, SEDG, utilities | Broader clean energy; less AI-specific |
GRID is the most direct ETF play on AI power infrastructure — its top holdings are the transformer and grid equipment manufacturers (ETN, HUBB) alongside the AI-exposed generators (CEG, VST). AMPS provides similar exposure with more US energy independence framing. ICLN is broader and includes solar and wind equipment that is less directly tied to AI data center power demand.
The AI data center power theme is one of the clearest secular investment opportunities of this decade — but the best portfolio approach is layered rather than concentrated in a single segment.
The single biggest catalyst to watch in H2 2026 is the FERC interconnection queue reform implementation — if large data center projects are able to advance more quickly through the queue, it will accelerate load growth for utilities and generators simultaneously. Eaton's Q3 2026 earnings backlog update will also be closely watched as a real-time indicator of grid equipment demand.
Use our comparison tool to analyze these stocks side by side on valuation, growth, and analyst targets.