TL;DR
- Morgan Stanley projects $3 trillion in data center capital expenditure through 2030, with less than 20% deployed so far. The vast majority of this spending still needs to flow through power infrastructure, grid construction, and electrical equipment companies — creating a multi-year demand runway that most investors are underestimating.
- The “picks and shovels” thesis is working. Eaton has roughly tripled from 2023 lows. Vertiv has risen over 400%. Quanta Services has doubled. These companies benefit from every dollar of AI capex regardless of which model or hyperscaler wins the AI race.
- Power is the binding constraint on AI scaling. US data center power consumption is projected to grow from 17 GW to 35–45 GW by 2030. Transmission bottlenecks, grid interconnection queues exceeding 2,600 GW nationally, and 3–5 year lead times for new transmission lines create structural scarcity that benefits incumbents.
- Nuclear is the dark horse. Constellation Energy's deal with Microsoft to restart Three Mile Island Unit 1 signals that hyperscalers will pay premium prices for 24/7 clean baseload power. Nuclear offers what renewables cannot: firm, dispatchable power without carbon emissions.
- The risks are real: overbuilding parallels to the fiber optic bust, behind-the-meter generation bypassing utilities, regulatory pushback on cost allocation, and valuations already pricing in years of growth. Use platforms like DataToBrief to monitor the capex commitments, power purchase agreements, and earnings guidance that separate signal from hype.
The $3 Trillion Buildout: Why 80% of AI Infrastructure Spend Is Still Ahead of Us
Morgan Stanley's projection of $3 trillion in cumulative data center capital expenditure through 2030 is the single most important number for infrastructure investors to understand. It is not a theoretical forecast — it is derived from committed hyperscaler capex guidance, signed power purchase agreements, land acquisitions, and equipment orders with multi-year lead times. And the critical insight is this: less than 20% of that spending has been deployed.
What has been built so far is impressive but insufficient. Microsoft guided to $80 billion in AI-related capex for fiscal year 2025. Amazon committed approximately $75 billion. Google and Meta each signaled $50–65 billion ranges. These are annual numbers, and they are accelerating, not decelerating. When Satya Nadella says Microsoft will spend more on AI infrastructure in a single year than most countries spend on their entire military budgets, he is not exaggerating — and the utility and grid companies that feed these data centers are the direct beneficiaries.
The math is straightforward. Every $1 billion of data center construction requires roughly $200–400 million in power infrastructure, electrical equipment, cooling systems, and grid upgrades. Over $3 trillion in total data center capex, that implies $600 billion to $1.2 trillion flowing specifically through the power and grid infrastructure value chain. That is the addressable market for the companies we are going to analyze.
What “Less Than 20% Deployed” Actually Means
To put the deployment gap in context: the US currently has approximately 5,400 data centers, with about 35 GW of total power capacity. To reach the 80–100+ GW that Goldman Sachs and McKinsey project for 2030, the industry needs to roughly triple installed capacity in five years. The construction pipelines are there — CBRE tracks over 5 GW of data center capacity under construction in the US alone as of Q4 2025 — but the power infrastructure to support it lags badly. Grid interconnection queues across US ISOs and RTOs total over 2,600 GW of proposed projects, with average wait times of 3–5 years. This is the bottleneck that creates pricing power for the companies that can actually get power infrastructure built.
The Five Companies That Own the AI Power Value Chain
Five companies sit at the center of AI power infrastructure investment. Each occupies a distinct niche in the value chain, and together they capture the vast majority of incremental spending driven by data center expansion. Here is our assessment of each.
Eaton Corporation (ETN): The Electrical Backbone
Eaton manufactures the electrical distribution equipment that sits between the grid and every piece of computing hardware in a data center: switchgear, transformers, uninterruptible power supplies, power distribution units, and busways. The company has been the single best-performing large-cap industrial stock in the AI infrastructure theme, roughly tripling from its October 2023 lows to trade above $340 in early 2026.
The bull case is undeniable: Eaton's data center revenue has been growing at 30–40% year-over-year, its backlog extends over two years, and management has increased its long-term organic growth guidance from 6–8% to 8–10%. The company's Q4 2025 earnings call was unambiguous — CEO Craig Arnold stated that data center demand was “the strongest secular trend I've seen in my career.”
The bear case: Eaton trades at roughly 33x forward earnings, well above its 10-year average of 18–20x. This premium assumes sustained data center demand growth with no meaningful deceleration through the end of the decade. Any softening in hyperscaler capex guidance or evidence of order cancellations could trigger a sharp multiple compression.
Vertiv Holdings (VRT): Power and Cooling Convergence
Vertiv provides critical power management and thermal management solutions for data centers — power conditioning, distribution, and precision cooling systems. The stock has been the most dramatic re-rating story in the AI infrastructure space, rising from under $15 in early 2023 to over $130 by early 2026 — roughly a 9x move.
Vertiv's positioning is uniquely favorable because AI workloads require both more power and more cooling per rack than traditional computing. GPU-dense racks running NVIDIA H100 or Blackwell chips draw 40–70 kW each and generate enormous heat that conventional air cooling cannot efficiently dissipate. Vertiv's liquid cooling and direct-to-chip cooling solutions address this bottleneck, and the company has seen liquid cooling orders grow at triple-digit rates.
Our concern with Vertiv at current levels is valuation: at approximately 40x forward earnings, the stock is priced for perfection. Management has executed superbly, but the market capitalization now exceeds $50 billion for a company generating roughly $8 billion in revenue. Any miss on orders or margin guidance will be punished severely.
Quanta Services (PWR): Building the Grid
Quanta Services is the largest specialty contractor in North America for electric power transmission and distribution infrastructure. If Eaton provides the components and Vertiv provides the power management, Quanta provides the labor force and project management to actually build, upgrade, and maintain the grid infrastructure that connects data centers to the power grid.
This is arguably the most defensible position in the entire AI power value chain. Skilled electrical construction labor is scarce, permitting expertise is difficult to replicate, and Quanta's scale provides competitive advantages in bonding capacity and equipment fleet management. The company's backlog exceeded $33 billion as of Q4 2025, representing approximately 2x annual revenue, providing exceptional revenue visibility.
We think Quanta is the most attractively valued of the five companies discussed here, trading at roughly 25x forward earnings despite best-in-class revenue visibility and a structural labor scarcity that supports pricing power. The risk is execution — large infrastructure projects are inherently lumpy and subject to weather, permitting, and supply chain disruptions.
NextEra Energy (NEE): Renewables Meet Data Centers
NextEra is the world's largest generator of renewable energy from wind and solar, and the parent of Florida Power & Light, the largest regulated electric utility in the United States. The AI infrastructure angle is twofold: hyperscalers have aggressive sustainability commitments that require renewable energy procurement, and NextEra's development pipeline of wind, solar, and battery storage projects directly serves this demand.
NextEra Energy Partners and the parent company's development arm have signed multiple power purchase agreements with major data center operators. However, our analysis shows a structural limitation: renewables alone cannot provide the 24/7 firm power that AI data centers require. Solar generates power for 4–6 peak hours daily, wind is intermittent, and battery storage at the scale needed to firm up renewable power for a 100+ MW data center remains prohibitively expensive. This creates an opening for nuclear and natural gas — which brings us to Constellation.
Constellation Energy (CEG): The Nuclear Renaissance
Constellation Energy owns the largest nuclear fleet in the United States — 21 reactors generating approximately 21 GW of carbon-free baseload power. The company's September 2024 announcement that it would restart Three Mile Island Unit 1 under a 20-year power purchase agreement with Microsoft was a watershed moment for the nuclear-AI thesis.
Nuclear provides exactly what AI data centers need: 24/7 dispatchable power at 90%+ capacity factors, zero carbon emissions, and the ability to scale to hundreds of megawatts at a single site. Constellation's existing fleet is already the lowest-cost generator in PJM (the regional transmission organization covering the Mid-Atlantic and Midwest) when you account for the value of clean energy credits.
The stock has reflected this thesis aggressively. Constellation traded below $100 in mid-2024 and exceeded $340 by early 2026 — a remarkable re-rating for what was traditionally a stable utility. At current levels, the market is pricing in significant contract upside from additional data center PPAs and favorable power market conditions. We believe the thesis has merit but the easy money has been made; from here, returns depend on the pace and pricing of new contracts.
Transmission Bottlenecks: The Most Underappreciated Constraint on AI Growth
You can build a data center in 18–24 months. You cannot build a high-voltage transmission line in less than 5–7 years. This mismatch is the single most underappreciated constraint on AI infrastructure scaling, and it creates investment opportunities that the market has not fully priced.
The numbers are stark. According to Lawrence Berkeley National Laboratory, the US grid interconnection queue contained over 2,600 GW of proposed projects at the end of 2024 — roughly double the entire existing US generation fleet. Average wait times for interconnection have stretched from 2–3 years a decade ago to 4–5 years today. Only about 20% of projects that enter the queue ultimately reach commercial operation.
For data centers, this means that securing grid interconnection is becoming more valuable than the land the data center sits on. A site with 200 MW of secured grid capacity in Northern Virginia, Central Texas, or the Phoenix metro area is worth a massive premium. And the companies that can actually build transmission — Quanta Services, MasTec, Pike Electric — have pricing power that reflects the scarcity of their capabilities.
FERC's May 2024 Order 1920, which requires long-range transmission planning and cost allocation, is a positive catalyst for the transmission buildout. But implementation will take years, and the order faces legal challenges from states that object to bearing costs for transmission that primarily benefits data centers in other jurisdictions. This is a classic case where the regulatory framework has not caught up with the pace of demand.
Our view: Transmission is where the structural scarcity is greatest and the market is least efficient in pricing it. For investors focused on AI infrastructure investing, grid and transmission companies offer a better risk/reward at current valuations than the more heavily-followed data center REITs and chip stocks.
AI Power Infrastructure Stocks: Valuation Comparison
The following table compares the five core AI power infrastructure stocks across key financial and operational metrics as of February 2026.
| Company | Ticker | Fwd P/E | Rev Growth (YoY) | Backlog / Revenue | AI Revenue Exposure |
|---|---|---|---|---|---|
| Eaton Corp | ETN | ~33x | +12% | ~2.0x | ~25–30% |
| Vertiv Holdings | VRT | ~40x | +18% | ~1.5x | ~50–60% |
| Quanta Services | PWR | ~25x | +14% | ~2.0x | ~15–20% |
| NextEra Energy | NEE | ~22x | +8% | N/A (regulated) | ~10–15% |
| Constellation Energy | CEG | ~28x | +15% | PPA contracted | ~20–30% |
Grid Modernization: The $300 Billion Opportunity Nobody Is Talking About
Beyond new transmission lines, the existing US grid requires massive modernization to handle the load characteristics of AI data centers. Data centers draw constant, high-density power loads that stress grid infrastructure designed for the variable demand patterns of residential and commercial users. Substations need upgrading, transformers need replacing, and distribution networks need reinforcing.
The American Society of Civil Engineers gives US energy infrastructure a C– grade. Much of the grid was built in the 1960s and 1970s, with an average transformer age exceeding 40 years — well past the 25–30 year design life. Adding tens of gigawatts of data center load to aging infrastructure requires not just new capacity but replacement and hardening of existing assets.
The Department of Energy estimates that achieving the Biden-era grid modernization goals requires $300–400 billion in transmission investment alone over the next decade. The Inflation Reduction Act provides tax incentives and loan programs that partially fund this buildout, but the majority of capital must come from utility rate base investment and private infrastructure funds.
For investors interested in how AI is reshaping macroeconomic analysis and forecasting, the power infrastructure buildout is a significant input to GDP growth, industrial production, and employment data over the next several years. It is one of the few clearly identifiable sources of sustained capital investment growth in the US economy.
Behind-the-Meter Generation: The Threat to the Utility Thesis
Every bull case deserves a steel-manned bear case, and for utilities the biggest threat is behind-the-meter generation. Hyperscalers are actively exploring options to generate their own power on-site, bypassing the utility grid entirely. If this trend accelerates, it undermines the core thesis that AI demand flows through regulated utilities.
Microsoft's investment in nuclear microreactors, Amazon's acquisition of a nuclear-powered data center campus from Talen Energy, and Google's agreement with Kairos Power for small modular reactors all point in the same direction: hyperscalers want to control their own power supply rather than depend on utilities and grid constraints.
However, we think the behind-the-meter threat is overblown in the near term. Small modular reactors are at least 5–7 years from commercial deployment at meaningful scale. On-site natural gas generation faces emissions and permitting challenges. And even hyperscalers that generate some power on-site will still need grid interconnection for redundancy and peak demand. The realistic scenario is that behind-the-meter generation supplements rather than replaces grid power for most data centers through 2030.
How to Build an AI Infrastructure Portfolio: Our Framework
For investors looking to build diversified exposure to AI power infrastructure, we suggest a framework based on three tiers of risk/reward.
Core holdings (60%): Eaton, Quanta Services, and a regulated utility with significant data center exposure (Dominion Energy or Southern Company). These provide the most defensible revenue streams and the best risk-adjusted returns over a 3–5 year horizon.
Growth holdings (30%): Vertiv and Constellation Energy. Higher upside potential but elevated valuations that require sustained execution on backlog conversion and contract wins. Position sizing should reflect the higher beta.
Opportunistic holdings (10%): Mid-cap names like nVent Electric (NVT), Regal Rexnord (RRX), or MasTec (MTZ) that offer specialized exposure at more attractive valuations. These are higher-conviction ideas that require more detailed fundamental analysis using tools like AI-powered research platforms to monitor order backlogs, margin trends, and customer concentration.
Frequently Asked Questions
Why are utility stocks benefiting from AI infrastructure demand?
Utility stocks are benefiting from AI infrastructure demand because data centers require enormous amounts of electricity, and this demand is reversing two decades of flat US power consumption growth. Morgan Stanley estimates that $3 trillion in data center capital expenditure is planned through 2030, but less than 20% has been deployed so far. This buildout requires grid interconnections, transmission upgrades, and new generation capacity that directly benefit regulated utilities earning allowed returns on deployed capital. Utilities in data center-heavy regions like Virginia (Dominion Energy), Texas (Oncor/Sempra), and Georgia (Southern Company) are seeing load growth forecasts revised upward from 1-2% to 5-8% annually, which supports higher rate base investment and correspondingly higher earnings growth. The market is re-rating these utilities from their traditional 12-15x P/E multiples to 18-22x, reflecting the secular growth dimension that AI demand adds to what was historically a low-growth, income-oriented sector.
What is the 'picks and shovels' thesis for AI power infrastructure?
The 'picks and shovels' thesis for AI power infrastructure argues that investors should focus on the companies providing essential physical infrastructure for AI rather than betting on which AI software or model wins. Just as hardware and equipment suppliers profited reliably during the Gold Rush regardless of which individual miners struck gold, companies like Eaton (electrical components), Vertiv (power management and cooling), Quanta Services (grid construction), and NextEra Energy (power generation) benefit from every dollar of data center capital expenditure regardless of whether that data center runs OpenAI, Google, or Meta workloads. The thesis has proven remarkably durable: Eaton stock has roughly tripled from its 2023 lows, Vertiv has risen over 400%, and Quanta Services has doubled, all driven by AI-related demand visibility. The key advantage of this approach is that it reduces technology risk — you don't need to pick the winning AI model, just correctly forecast that AI compute demand will grow, which is a much higher-confidence bet.
How much power do AI data centers actually consume?
AI data centers consume significantly more power than traditional cloud computing facilities. A single NVIDIA H100 GPU server rack can draw 40-70 kW, compared to 5-15 kW for a traditional enterprise server rack. A large-scale AI training cluster with thousands of GPUs can consume 50-100+ MW — equivalent to powering a small city of 40,000-80,000 homes. Goldman Sachs estimates that US data center power consumption will grow from approximately 17 GW in 2023 to 35-45 GW by 2030, representing roughly 8-10% of total US electricity generation capacity. Globally, the International Energy Agency projects data center electricity consumption could reach 1,000 TWh by 2030, roughly doubling from current levels. The growth is driven by both the expansion in the number of AI-capable data centers and the increasing power density per facility as next-generation chips (NVIDIA Blackwell, AMD MI400) push rack densities even higher.
What are the biggest risks to the AI utility and grid stock investment thesis?
The biggest risks include: (1) Overbuilding — if AI demand growth slows due to efficiency improvements (like DeepSeek's more efficient training methods), model commoditization, or recession, utilities and grid companies could face stranded capital investments and lower-than-expected load growth; (2) Regulatory risk — utility commissions may resist passing data center-related grid upgrade costs to residential ratepayers, creating cost allocation disputes that delay investment recovery; (3) Permitting and timeline risk — new transmission lines and generation facilities face multi-year permitting processes, environmental reviews, and community opposition that can delay or kill projects; (4) Valuation risk — many AI-adjacent utility and infrastructure stocks have already re-rated significantly, with Vertiv, Eaton, and Quanta trading at historically elevated multiples that leave little room for disappointment; (5) Behind-the-meter generation risk — hyperscalers are increasingly exploring on-site nuclear, natural gas, and solar generation that bypasses the utility grid entirely, which would reduce the demand flowing to regulated utilities and transmission companies.
Which AI infrastructure stocks are most undervalued in 2026?
Identifying undervalued AI infrastructure stocks requires comparing current valuations against the sustained multi-year demand buildout that Morgan Stanley's $3 trillion estimate implies. Our analysis suggests that transmission and grid construction companies — particularly Quanta Services (PWR) and MasTec (MTZ) — remain more attractively valued than their electrical equipment and cooling peers because the market has not fully priced in the multi-year backlog of grid modernization and transmission buildout required to support data center growth. In the utility space, Constellation Energy (CEG) offers unique exposure through its nuclear fleet, which provides the 24/7 baseload power that data centers require and that renewables alone cannot deliver — its power purchase agreement with Microsoft for the Three Mile Island Unit 1 restart demonstrates the premium that data center operators place on firm, clean power. Among mid-caps, nVent Electric (NVT) and Regal Rexnord (RRX) provide specialized electrical and mechanical infrastructure exposure at lower multiples than Vertiv or Eaton. That said, 'undervalued' is relative — the entire AI infrastructure complex has re-rated, and investors should stress-test assumptions against scenarios where demand growth comes in 30-40% below consensus.
Research AI Infrastructure Stocks with DataToBrief
Tracking the AI power infrastructure buildout requires monitoring capex guidance, backlog data, power purchase agreements, and regulatory filings across dozens of companies. DataToBrief automatically extracts these data points from 10-K, 10-Q, and 8-K filings with source citations, enabling you to monitor the entire AI infrastructure value chain from a single platform.
Request Early AccessThis article is for informational purposes only and does not constitute investment advice. Stock prices, valuations, and financial metrics cited reflect estimates at the time of writing and are subject to change. Past performance is not indicative of future results. Always conduct your own due diligence and consult a qualified financial advisor before making investment decisions.