Why AI Is Waiting Years for Grid Power—and the Dividend Stocks and ETFs Built to Benefit

DV

Written by Dividend Vision

Reviewed by Bert Sweet

Published Aug 16, 2026

✓Fact-checked Aug 16, 2026?

AI's biggest constraint is no longer chips—it's electricity. Here are the dividend-paying utilities, grid-equipment makers and ETFs positioned around the data-center power buildout, screened for payout safety with Dividend Vision's Distribution Safety Score.

Why AI Is Waiting Years for Grid Power—and the Dividend Stocks and ETFs Built to Benefit

Artificial intelligence may be developing at software speed, but the infrastructure powering it still moves at utility speed.

That mismatch is creating what energy investor and technologist Ramez Naam calls the AI industry's most important physical constraint: not chips, models or capital, but access to electricity. In an August 2026 discussion with Peter Diamandis, Salim Ismail and Alexander Wissner-Gross, Naam argued that a developer requesting hundreds of megawatts for a new data center today may wait until 2031 or 2032 for grid power—a queue of five to six years.

Microsoft CEO Satya Nadella has described the same constraint from the buyer's side. Microsoft can obtain AI chips, but it cannot always find enough powered data-center capacity—"warm shells"—in which to install them. Meanwhile, the U.S. Department of Energy estimates that domestic data-center electricity demand could double or triple between 2023 and 2028.

For income-focused investors, this raises a specific question: if electricity is becoming the innermost loop of AI, which dividend-paying companies and ETFs supply the generation, transmission, storage and cooling needed to widen it—and which of those payouts actually look durable?

The answer is not one stock or one technology. It is a value chain.

The thesis in one sentence: AI's power shortage may benefit the companies that generate electricity, move it, condition it, store it and remove the resulting heat—but thematic exposure does not make every security a good investment at every valuation, and it certainly does not make every payout safe.

The AI-power investment map

These are research candidates, not a model portfolio. Many funds own the same companies, and several speculative reactor developers have little or no operating revenue. Investors should examine valuation, liquidity, financial strength and portfolio overlap before turning a compelling theme into an actual position.

From a theme list to an investable shortlist

A long ticker list is useful for mapping an opportunity, but it is not yet a portfolio. The next step is to filter the universe according to what the investor is actually trying to achieve—and for most readers of this site, that means durable income first.

Dividend Vision makes that process explicit. Its proprietary Distribution Safety Score™ grades the apparent durability of a security's payout on a 0–100 scale: 80–100 is Safe, 60–79 is Generally safe, 40–59 indicates Caution, 20–39 indicates Elevated risk, and 0–19 indicates High risk.

The score is a payout-risk screen—not a forecast, credit rating or measure of whether the share price is safe. A company can earn a high score on a dividend too small to interest an income investor. Conversely, a young fund can be capped near 50 because it has not accumulated enough payment history, even when its underlying holdings are established businesses.

The five-filter funnel

The funnel does not identify a universal "best" investment. It produces different shortlists for different jobs.

Income shortlist: require both yield and safety

For a strict income screen, use these starting rules:

• Distribution rate of at least 2.5%
• Distribution Safety Score of at least 80
• At least two years of paying history
• No decreasing distribution trend
• For ETFs, an expense ratio no higher than 0.60%
• Rank the survivors by Safety-Adjusted Yield, then review the factors behind each score

Safety-Adjusted Yield equals the distribution rate multiplied by the Safety Score divided by 100. It discounts a large but fragile payout instead of allowing headline yield to dominate the ranking.

AI-power ETFs that clear the income screen

Dividend Vision data as of August 16, 2026 produces a compact list:

The first three provide the cleaner income exposure. NLR offers more thematic torque but also greater commodity, regulatory and drawdown risk, and it distributes annually rather than quarterly.

Several otherwise relevant ETFs fail this screen. ZAP yields 2.03% but has a score of 50; GRID scores 78 but yields only 0.76%; and POW, VOLT, ELFY, POWR, NUKZ and COOL either lack sufficient yield, fall below the safety threshold or have too little history to qualify. That does not make them poor growth investments—it makes them poor matches for this income mandate.

Individual stocks that clear the income screen

These are screen results, not automatic buys. Utility earnings depend on regulators allowing capital investment into the rate base, and rising data-center demand can require substantial financing before shareholders see the benefit. Valuation, debt, service-territory concentration and allowed returns still matter.

Growth shortlist: prioritize bottleneck exposure, not yield

The growth branch deliberately removes the minimum-yield rule. It asks which businesses have direct exposure to scarce equipment, power capacity or cooling—and whether an ETF owns those businesses efficiently.

Core growth ETFs:

• Broad electrification: POW, VOLT, GRID, ZAP, POWR and ELFY. Choose one primary vehicle after comparing holdings overlap, expense ratio, fund age and liquidity.
• Nuclear satellite: NLR or NUKZ. NLR has the longer record and stronger current income profile; NUKZ is a newer, more concentrated thematic expression.
• Cooling satellite: COOL. Its rack-power and thermal-management focus is highly relevant, but its August 2026 launch, very small asset base and lack of distribution history place it in the speculative-growth bucket for now.
• AI-and-power blend: AIPO. It mixes the suppliers solving the power constraint with the semiconductor and AI companies creating demand.

GRID currently offers the strongest combination of long operating history and a reasonably established Safety Score among the dedicated grid funds, but its 0.76% distribution rate still makes it a growth holding. POW and VOLT provide more concentrated exposure; both currently score 50 and pay less than 0.5%, making their role capital appreciation rather than income.

Core growth stocks include GEV, ETN, PWR, HUBB, POWL and NVT in grid equipment and construction; CEG and VST in existing generation; VRT in data-center power and cooling; BWXT and CCJ in the nuclear supply chain; and FSLR and NXT in utility-scale solar.

Some of these companies pay very safe but very small dividends. Dividend Vision currently scores ETN at 100, HUBB at 100, PWR at 96, NVT at 96, POWL at 88, VRT at 87 and BWXT at 90—yet their distribution rates range from roughly 0.06% to 1.11%. The correct interpretation is "growth companies with apparently durable incidental dividends," not "income stocks."

Balanced shortlist: income today plus dividend growth

Investors who want both current cash flow and participation in the buildout can relax the yield floor while retaining a high Safety Score. Candidates include:

• NEE and ETR: regulated-utility income with potential load and generation growth.
• ETN, HUBB and CMI: electrical and power-equipment growth with high current payout-safety scores, although yields remain modest.
• NLR: a higher-yielding nuclear basket with more cyclicality than a broad utility ETF.
• SCCO: copper exposure with a 2.00% distribution rate and a score of 75, but distributions and share price remain commodity-sensitive.

This bucket is especially valuation-sensitive. An excellent business can still produce a poor return when purchased at an excessive multiple.

Speculative and watchlist bucket

The final branch preserves ideas that fit the narrative but fail the portfolio-quality or income tests today:

• Pre-commercial nuclear: XE, OKLO and SMR
• Higher-risk power technologies: BE, FLNC, NVTS and VICR
• Solar turnaround exposure: ARRY
• New or small thematic funds: COOL, GALX and selected recently launched products
• Extreme-compute exposure: SPCX and space suppliers held by GALX

These securities may have substantial upside, but the thesis depends more heavily on future commercialization, financing, contract execution or market adoption. They should not sit beside regulated utilities as though they carry the same type of risk.

Recreate the screens on Dividend Vision

Readers can paste the article's tickers into the Dividend Vision screener, then use two saved views:

• AI Power—Income: distribution rate ≥2.5%, Safety Score ≥80, paying history ≥2 years, dividend trend "not decreasing," sorted by Safety-Adjusted Yield.
• AI Power—Growth: no yield minimum; filter for acceptable fund age, assets, volume and expense ratio, then compare total return, drawdown and holdings overlap.

Because yields, distributions and Safety Scores update daily, treat the live screen as the current answer and the tables above as an August 16, 2026 snapshot.

Understanding the bottleneck: nine forces shaping the buildout

1. The "poles and wires" bottleneck

The most important misconception is that every data-center delay can be solved simply by building another power plant. Generation matters, but electricity must also travel through transmission lines, substations, transformers and switchgear before it reaches an AI server.

Interconnection queues illustrate the scale of the problem. Lawrence Berkeley National Laboratory reported approximately 8,200 U.S. projects seeking grid interconnection at the end of 2025, representing 1,312 gigawatts of generation and 749 gigawatts of storage. Queue volume is not the same as future capacity—many proposed projects will never be built—but it demonstrates the burden placed on grid planners and utilities.

For investors, this makes the electrical-equipment supply chain one of the clearest near-term expressions of the thesis:

• GE Vernova (GEV) supplies gas turbines, grid equipment and nuclear technology through GE Hitachi Nuclear Energy.
• Eaton (ETN) supplies switchgear, breakers, power distribution and other equipment used between the grid and the data-center rack.
• Quanta Services (PWR) constructs transmission, distribution and substation infrastructure.
• Hubbell (HUBB), Powell Industries (POWL) and nVent Electric (NVT) provide grid and electrical-distribution components.
• ABB, Hitachi (HTHIY), Schneider Electric (SBGSY), Siemens Energy (SMNEY) and Mitsubishi Heavy Industries (MHVIY) add international exposure. Some of these U.S. symbols are over-the-counter ADRs and may have lower liquidity.

The closest diversified funds are POW (active, global picks-and-shovels), VOLT (concentrated grid, utilities and nuclear), GRID (established global smart-grid fund), ZAP (U.S.-focused electrification), POWR (broad U.S. power infrastructure) and ELFY (diversified, less concentrated electrification). Of these, POW, VOLT, GRID and ZAP most directly express the grid-equipment thesis. Their holdings overlap significantly, so owning all four may create the illusion of diversification while repeatedly purchasing the same electrical-equipment companies.

2. Unlocking capacity with flexible—or interruptible—loads

The fastest source of new AI power may be electricity the grid already has but does not use continuously.

Power systems must be sized to survive periods of peak demand, such as extremely hot summer afternoons. During many other hours, portions of that capacity sit unused. Naam describes this gap as roughly 200 gigawatts across the United States.

A Duke University study led by Tyler Norris estimated that nearly 100 gigawatts of new large loads could potentially be integrated with minimal reliability impact if those customers accepted brief curtailments—averaging roughly 0.5% of hours annually, not routine shutdowns.

AI facilities could respond by moving deferrable training workloads to off-peak hours, shifting workloads geographically between data centers, reducing consumption briefly during grid emergencies, or using batteries and on-site generation during peaks.

Naam compares batteries to a cache for electrons: charge when grid capacity is abundant, then discharge when transmission capacity is constrained. The analogy is imperfect—batteries cannot create energy—but it captures their role in time-shifting supply.

Potential public-market beneficiaries include Fluence Energy (FLNC), Tesla (TSLA) through Megapack, and electrical-control suppliers such as ETN. Battery-oriented funds include BATT and LIT, although both remain heavily tied to lithium, battery materials and electric vehicles rather than pure stationary-storage exposure.

3. Sodium-ion batteries: promising, but not yet a guaranteed 10× breakthrough

Battery costs have already changed the economics of renewable power and grid storage. The International Energy Agency estimates that lithium-ion battery prices fell about 90% between 2010 and 2023, and battery energy-storage-system prices continued declining in 2025.

Naam presents sodium-ion chemistry as a potential next leap because sodium is abundant and stationary storage does not require the highest possible energy density. His upside case envisions storage costs eventually falling by another order of magnitude.

Treat that as an aggressive technology forecast—not a present-day cost assumption. The IEA has estimated that scaled sodium-ion batteries could cost approximately 20% less than incumbent technologies, while warning that supply chains, material quality and low lithium prices could slow adoption.

There is currently no clean U.S.-listed sodium-ion ETF, and many leading sodium-ion manufacturers trade primarily in Asian markets. That makes the nearer-term investable opportunity less about choosing the winning chemistry and more about owning storage integrators, power electronics and grid equipment capable of working with multiple chemistries.

4. Behind-the-meter power: stop waiting for the utility

When a grid connection will not arrive quickly enough, data-center developers can build generation directly on-site, or "behind the meter." The economics may be worse than ordinary utility power, but electricity is only one component of a data center filled with tens of billions of dollars of rapidly depreciating AI hardware. Speed can matter more than the lowest possible power price.

Naam highlights a severe backlog for large natural-gas turbines. The public companies with the clearest exposure include GEV (large and aeroderivative gas turbines), Caterpillar (CAT, which owns industrial-turbine maker Solar Turbines), Bloom Energy (BE, rapidly deployable on-site fuel cells), Cummins (CMI, generators and distributed power), and Siemens Energy and Mitsubishi Heavy Industries through SMNEY and MHVIY.

One clarification: GE Hitachi is not a separate turbine stock—it is a nuclear joint venture. GE Vernova is the relevant U.S.-listed GE security, while the former Mitsubishi Hitachi thermal-power business is now part of Mitsubishi Heavy Industries.

Defiance AI & Power Infrastructure ETF (AIPO) combines AI hardware, data centers, generation, utilities and power infrastructure—a single-fund approach to both the source of AI demand and the equipment needed to serve it.

5. Utilities and existing power plants: the fastest generation is already operating

Regional utilities may experience substantial load growth as data centers enter their service territories—the names in the income shortlist above are the clearest candidates.

This is not free money. Regulated utilities may earn returns on approved capital investment, but regulators can require data-center customers to pay for upgrades and protect residential ratepayers. Large projects also create financing, execution and political risks.

Broad utility ETFs such as XLU, VPU and FUTY reduce company-specific exposure. ZAP and POWR combine utilities with more of the supporting infrastructure.

Independent power producers and nuclear fleet owners may offer more direct exposure to rising wholesale-power demand. Constellation Energy (CEG) is working to restart Three Mile Island Unit 1—renamed the Crane Clean Energy Center—under a 20-year agreement with Microsoft. Vistra (VST) has agreements supporting nuclear output for Meta. Talen Energy (TLN) offers additional but more concentrated data-center and nuclear exposure.

6. Nuclear power: extend, restart, then manufacture

The fastest nuclear strategy is not necessarily a futuristic reactor. It is keeping existing plants operating, increasing their output and restarting viable plants that closed for economic reasons.

The longer-term thesis involves moving nuclear construction toward repeatable manufacturing:

• Large reactors: Westinghouse's AP1000 remains an established design, but Westinghouse itself is private. Cameco (CCJ) owns 49% of Westinghouse; Brookfield and its partners own the remaining 51%, providing indirect exposure through BEP and BEPC.
• Small modular reactors: X-energy (XE), Oklo (OKLO) and NuScale Power (SMR) provide publicly traded development-stage exposure.
• Fuel and components: CCJ, BWX Technologies (BWXT) and Centrus Energy (LEU) supply uranium, components, fuel and enrichment services.
• Reactor technology: GEV participates through GE Hitachi Nuclear Energy.

The principal ETF choices differ meaningfully. NLR spans uranium mining, nuclear generation, construction, engineering and equipment; NUKZ emphasizes advanced reactors, utilities, nuclear services and fuel; URA and URNM are more directly tied to uranium miners and the fuel cycle.

Development-stage nuclear stocks can trade primarily on contracts, regulatory milestones and expectations years before meaningful commercial revenue. They should not be evaluated like established utilities or equipment manufacturers.

Fusion's regulatory advantage—without the "hospital machine" shortcut. Fusion may eventually face a lighter regulatory framework than commercial fission because it does not sustain the same kind of chain reaction or meltdown scenario. The Nuclear Regulatory Commission is developing a framework based largely on its rules for byproduct material rather than automatically treating fusion facilities like conventional fission power reactors. That is a meaningful potential advantage, but commercial fusion still faces unresolved engineering, licensing, financing and fuel-cycle questions. Commonwealth Fusion Systems and Helion remain private, so public-market exposure is currently indirect and highly diluted through partners such as Microsoft.

7. Solar plus storage: the fastest modular build

Solar and batteries can be manufactured in factories and deployed in repeatable increments. Naam argues that a large solar-plus-storage project can be delivered much faster than a new transmission line, large turbine or nuclear plant.

Potential beneficiaries include First Solar (FSLR), a U.S. utility-scale solar manufacturer; Nextracker (NXT) and Array Technologies (ARRY), which supply solar tracking systems; NEE, FLNC and TSLA through renewable development and storage; and the Invesco Solar ETF (TAN) for broader solar-industry exposure.

Solar is not equally effective everywhere. Daily battery storage can shift power from afternoon to night; it does not cheaply solve seasonal shortages in high-latitude winters. That leaves room for a mixed system of solar, storage, gas, geothermal and nuclear rather than one universal technology.

8. Cooling: every watt eventually becomes heat

Supplying electricity is only half of the physical problem. Nearly every watt consumed by an AI rack ultimately becomes heat that must be removed.

Vertiv (VRT) is one of the clearest public beneficiaries through data-center power distribution, thermal management and liquid cooling. ETN and NVT also participate in power delivery and enclosures, while Vicor (VICR) and Navitas Semiconductor (NVTS) provide higher-risk exposure to power conversion and semiconductors.

VegaShares AI Thermal, Cooling & Power Management ETF (COOL) is the most specialized ETF in this group, targeting server liquid cooling, power supplies, battery backup, rack power distribution and power semiconductors. COOL launched in August 2026 with very limited assets and trading history, so investors should pay particular attention to bid-ask spreads, trading volume and fund-closure risk.

VistaShares Artificial Intelligence Supercycle ETF (AIS) is broader—semiconductors, memory, data-center infrastructure and AI applications. AIS benefits from AI capital spending, but it represents the demand side of the power shortage more than a pure grid solution.

9. Extreme data centers: oceans and orbit

Naam's most speculative scenarios move computing away from conventional land-based infrastructure.

Ocean compute. The concept combines wave-generated electricity with cold-water cooling: factory-built floating or submerged units could theoretically avoid some land, cooling and grid constraints. Naam discusses a private portfolio company pursuing this model and targeting extremely low power costs. Those targets remain unverified at commercial scale, and there is no clean publicly traded ocean-data-center stock. Treat the concept as venture-stage technology, not an earnings forecast for an existing public company.

Orbital compute. Orbit offers long periods of solar exposure, but it replaces terrestrial constraints with launch cadence, radiation, communications, maintenance and regulation. Naam estimates that deploying 10 gigawatts of orbital compute annually could require five or six Starship launches every day. His conclusion is not that orbital compute is imminent, but that it may remain prohibitive for 15 to 20 years without extraordinary progress.

SpaceX (SPCX) is now publicly traded and supplies the most direct exposure to Starship launch economics. VistaShares Space Supercycle ETF (GALX) owns launch, satellite, communications and space-infrastructure companies. GALX is relevant to the orbital-compute thought experiment, but it is not a substitute for POW, GRID or VOLT—it belongs in the speculative frontier, not the core grid basket.

Where HALX fits—and where it does not

Tuttle Capital Heavy Assets Low Obsolescence ETF (HALX) selects tangible, capital-intensive businesses whose assets may be difficult for AI to replace, including utilities, energy infrastructure, railroads and commodity producers.

HALX may serve as a broad physical-economy complement to technology exposure, but it is not a dedicated AI-power or electrification ETF. Holdings such as retailers, railroads, gold companies and conventional energy producers substantially dilute the grid thesis.

Comparing the principal AI-power ETFs

The efficiency horizon: what could break the thesis?

The largest risk to the "AI consumes limitless electricity" narrative may be AI itself.

Naam argues that intelligence gains are sublinear with compute: each incremental improvement may require disproportionately more hardware and energy. But algorithms, model architecture, inference optimization and specialized chips can move the curve in the opposite direction by producing more useful output per watt.

That creates two competing forces:

• Jevons-style expansion: efficiency makes AI cheaper, increasing adoption and total electricity consumption.
• Saturation: businesses eventually stop paying for marginal intelligence that does not create enough additional value.

Investors should therefore avoid assuming that every announced gigawatt will be built or that current equipment shortages will persist indefinitely. Interconnection requests can be speculative, technologies can change, and high expected growth can already be embedded in stock prices. For income investors specifically, a regulated utility's dividend does not depend on the AI boom persisting—which is precisely why the income shortlist leans on regulated names rather than thematic torque.

Bottom line

The near-term AI-power opportunity is grounded in ordinary but essential infrastructure: transformers and switchgear, transmission lines and substations, gas turbines and distributed generation, batteries and flexible loads, existing nuclear plants and their supply chains, utility-scale solar, and power conversion and liquid cooling.

Ocean and orbital data centers may eventually matter, but the present bottleneck is terrestrial. The race is taking place in factories, utility commission hearings, interconnection studies and electrical-equipment order books.

For income investors, the cleanest expressions today are broad utility ETFs (FUTY, VPU, XLU), high-Safety-Score regulated utilities (D, FE, PNW, DUK, SO, AEP, NI, NEE) and, for those accepting more cyclicality, NLR. For diversified growth exposure, POW, VOLT, GRID, ZAP and POWR are the closest broad ETF matches, while COOL isolates rack-level power and cooling. AIS, GALX and HALX can complement the discussion, but they are not direct substitutes for electrification funds.

Before investing, use Dividend Vision to compare each fund's holdings, expense ratio, distribution history, total return, drawdowns and overlap. Owning five electrification ETFs that all hold Eaton, Quanta Services and GE Vernova is still one concentrated bet—just wearing five ticker symbols.

Sources and further reading

AI's Energy Wall: From Gridlock to Abundance

• Ramez Naam discussion: "200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Data Centers"
• Lawrence Berkeley National Laboratory: Queued Up—interconnection queue data
• U.S. Department of Energy: Report on U.S. Data Center Energy Use
• Duke University Nicholas Institute: Rethinking Load Growth
• International Energy Agency: Batteries and Secure Energy Transitions
• International Energy Agency: Sodium-ion battery potential and limitations
• U.S. Nuclear Regulatory Commission: Part 30 byproduct-material framework
• GE Vernova: Gas power technology for data centers
• Constellation Energy: Crane Clean Energy Center restart
• Vistra and Meta nuclear-power agreements
• SpaceX: June 2026 initial public offering
• Dividend Vision: Distribution Safety Score methodology
• Dividend Vision: Safety-Adjusted Yield methodology

For informational and educational purposes only. Dividend Vision does not provide individualized investment, tax or legal advice. All investments involve risk, including possible loss of principal. ETF holdings and strategies can change.