AI Infrastructure Builders Outpace Utilities as Power Demand Rises

Google News has surfaced a striking AI infrastructure comparison: utilities gained about 8%, while a fund tied to physical data center construction rose roughly 40%.

The figures frame an important market reversal. Investors first treated utilities as the clearest way to capture rising AI power demand. The stronger gains reportedly appeared elsewhere, among companies supplying electrical equipment, cooling systems, construction services, steel, and transportation.

That contrast deserves more scrutiny than the headline alone provides. Performance windows can differ, and an infrastructure fund does not represent a pure data center portfolio. Yet the gap points toward a real shift in the AI spending cycle.

The AI boom is moving beyond chips and electricity generation. It increasingly depends on contractors and manufacturers that can turn capital budgets into working facilities.

Utilities remain essential, but regulated business models limit how quickly new spending becomes earnings. Equipment suppliers can receive orders earlier, sometimes before a utility recovers its investment through customer rates.

The result is a contest between power providers and the companies building the systems around them. It also exposes the risks beneath both trades, including delayed projects, crowded valuations, financing pressure, and uncertain AI demand.

What the Google News Headline Actually Changed

The headline redirects attention from electricity demand toward the companies converting that demand into physical capacity.

The original claim contrasts a utility-sector gain near 8% with a roughly 40% rise for an infrastructure-oriented fund. Those numbers should be treated as a reported market snapshot, not a permanent relationship.

The supplied headline does not establish that both percentages use identical starting dates. A year-to-date return and a trailing-year return would describe different performance periods. Distributions, fund rebalancing, and publication timing can create further differences.

That limitation does not erase the underlying observation. Utility shares have benefited from expectations that AI data centers will increase electricity consumption. Infrastructure suppliers have also captured spending on the equipment required before a data center can operate.

The likely fund behind the comparison is the Global X U.S. Infrastructure Development ETF, which trades under the ticker PAVE. Its official fund mandate covers American infrastructure activity across raw materials, heavy equipment, engineering, and construction.

That wording matters. PAVE is not a dedicated data center fund, and it does not own only businesses serving artificial intelligence. It also holds companies exposed to railroads, public works, manufacturing, roads, and broader construction activity.

Still, several portfolio categories align closely with the AI buildout. Electrical contractors connect generation and substations to large campuses. Equipment manufacturers provide switchgear, breakers, transformers, and power-management systems.

Heating, ventilation, and air-conditioning suppliers address the enormous cooling burden created by dense computing racks. Materials companies provide steel, concrete, and aggregates, while railroads transport bulky equipment across the country.

An earlier infrastructure analysis identified Quanta Services, Eaton, Trane Technologies, Sempra, Union Pacific, and Norfolk Southern among PAVE’s significant positions. The portfolio will change over time, but those examples show its wider exposure.

The fund therefore captures more than one stage of construction. It reaches from raw materials to electrical installation, cooling, transportation, and utility service.

A utility fund occupies a narrower section of that chain. Its companies generate or deliver electricity, maintain networks, and earn regulated returns on approved investments.

That distinction creates the article’s central tension. The businesses selling scarce equipment can benefit when orders arrive, while utilities must navigate planning reviews, interconnection studies, financing, construction, and rate approval.

The headline did not reveal a new technology. It highlighted where investors believe the current bottleneck sits.

For several years, graphics processors received most of the attention. The physical challenge now includes finding enough electricity, equipment, land, labor, and grid access to operate those processors.

That is why the reported performance gap has become meaningful. It suggests the market has rewarded the builders of capacity more aggressively than the regulated companies expected to supply the power.

AI Power Demand Is No Longer a Distant Forecast

Data center electricity consumption has moved from a planning scenario into a measurable source of load growth.

A December 2024 federal energy study estimated that data centers consumed 176 terawatt-hours of U.S. electricity during 2023. That represented about 4.4% of national electricity use.

The same study projected consumption between 325 and 580 terawatt-hours by 2028. Under that range, data centers would account for approximately 6.7% to 12% of U.S. electricity use.

The wide range is not a minor detail. It reflects uncertainty about AI adoption, computing efficiency, facility utilization, and the speed of new construction.

Even the lower estimate implies substantial growth. The higher estimate would force utilities, grid operators, regulators, and large technology companies to make unusually fast infrastructure decisions.

The International Energy Agency added global context in its global power outlook. It projected worldwide data center electricity consumption of around 945 terawatt-hours in 2030, more than twice the current level.

AI-optimized facilities account for much of that increase. These data centers contain accelerators designed for AI training and inference, the process of generating results from trained models.

The IEA expects U.S. data centers to drive nearly half of national electricity-demand growth through 2030. That would reverse a long period in which American demand remained relatively flat.

A later demand update found that global data center electricity use increased 17% during 2025. Overall worldwide electricity demand grew 3% during the same period.

The update also identified tightening supplies of transformers, gas turbines, advanced chips, and other information-technology components. Planning systems and grid-connection queues created additional delays.

These constraints explain why physical infrastructure suppliers have attracted attention. A utility cannot serve a large data center merely by producing additional annual energy.

The power must arrive at the correct location, voltage, quality, and reliability level. That requires substations, transmission capacity, distribution equipment, backup systems, and often dedicated generation.

A data center campus also needs redundant connections because an interruption can disrupt expensive computing workloads. Redundancy increases equipment requirements beyond the facility’s average consumption.

Cooling adds another layer. AI accelerators concentrate enormous computing output inside relatively small spaces, producing heat that must be removed continuously.

Traditional air cooling remains common, but higher rack densities are increasing demand for liquid-cooling systems. These systems move heat through fluid circulating near processors or server components.

This transition benefits specialized manufacturers before the local utility receives any new electricity revenue. The cooling system must be selected, ordered, installed, and tested during construction.

The same sequence applies to switchgear and transformers. Long production lead times can encourage developers to reserve equipment before every project detail becomes final.

That behavior pulls spending toward manufacturers and contractors. It also creates a risk that reported order books include projects that later change, shrink, or disappear.

Nevertheless, the direction is clear. AI power demand now shapes construction schedules, utility forecasts, and equipment supply chains.

The open question is not whether data centers use significant electricity. It is whether every announced facility will secure power and produce an acceptable financial return.

Why Builders Can Outrun Regulated Utilities

Infrastructure suppliers can recognize the AI spending impulse sooner because utilities face a slower and more regulated path to returns.

A regulated utility generally invests in generation, transmission, or distribution assets after forecasting demand and receiving required approvals. Regulators then decide which costs can enter the rate base.

The rate base is the value of approved assets on which a utility can earn a regulated return. It gives the business predictable economics, but it also creates delay.

A utility cannot assume that every proposed project will be fully recoverable from customers. Regulators examine whether investments are necessary, reasonably priced, and fairly allocated.

Data centers complicate that process because their loads are unusually large. A single campus can require power comparable with a manufacturing complex or a small city.

Developers also evaluate several locations at once. Utilities risk building infrastructure for demand that moves to another state, changes schedule, or never reaches operation.

Large-load tariffs attempt to manage that problem. These agreements can require minimum payments, longer commitments, or direct contributions toward infrastructure.

The goal is to prevent ordinary customers from absorbing construction costs created by a small number of technology companies. However, each tariff can trigger regulatory and political debate.

Infrastructure suppliers operate under different rules. A manufacturer can record orders from utilities, contractors, or data center developers without waiting for decades of customer rate recovery.

An engineering company can earn revenue while designing and constructing a substation. A utility earns its approved return after the project enters service and regulators accept the cost.

This timing difference helps explain the reported 8%-versus-40% split. The suppliers stand closer to current capital spending, while utilities convert that spending into earnings over longer periods.

Scarcity strengthens the effect. When transformer or switchgear production capacity is limited, suppliers gain pricing and scheduling leverage.

Utilities rarely receive the same immediate advantage from scarcity. They must still provide reliable service, manage customer affordability, and satisfy regulatory requirements.

The market also values the two groups differently. Traditional utilities often attract investors seeking stable cash flow and dividends. Their valuations can therefore respond to interest rates as well as electricity growth.

Higher bond yields can make utility dividends less attractive relative to government securities. They also increase financing costs for capital-intensive expansion.

Industrial suppliers are not immune to rates, but their earnings can accelerate when order volumes and factory utilization rise. Investors may assign higher growth expectations to that operating leverage.

PAVE spreads this exposure across several industries. Electrical equipment is only one component, while railroads, machinery, and construction materials bring different economic sensitivities.

That diversification can help when U.S. infrastructure spending rises broadly. It can also weaken the direct relationship between the fund and AI data centers.

A railroad holding may gain from construction shipments, but it also depends on industrial production, fuel costs, labor agreements, and freight volumes. An aggregates producer can serve highways and housing alongside data centers.

The fund’s gain therefore cannot be attributed entirely to AI. Domestic manufacturing, public infrastructure programs, and broader economic expectations also influence its holdings.

That is the crucial qualification missing from a dramatic Google News comparison. PAVE represents the American construction and equipment cycle, with AI acting as one important demand source.

Utilities provide more direct exposure to electricity consumption. The infrastructure fund provides wider exposure to the spending needed before that electricity can be delivered and used.

Neither route perfectly isolates AI. One captures regulated power delivery, while the other captures an overlapping collection of industrial bottlenecks.

The market has recently preferred the second route. That preference can persist only if orders become revenue and announced projects become operating facilities.

The 40% Gain Does Not Prove a Pure AI Trade

The strongest counterargument is that infrastructure performance combines AI enthusiasm with several unrelated economic forces.

Headline comparisons can make diversified funds appear more targeted than they are. PAVE’s name emphasizes infrastructure development, not artificial intelligence or data centers.

Its portfolio includes companies tied to transportation, construction machinery, engineering, raw materials, and industrial automation. Many benefit from trends that would exist without generative AI.

Government infrastructure programs support road, bridge, water, and transportation projects. Manufacturing investment increases demand for factory construction, freight, and electrical systems.

Housing and commercial construction can influence materials suppliers. Commodity prices affect steelmakers, while rail traffic responds to the wider economy.

A 40% fund gain can therefore reflect several forces arriving together. Calling the entire return an AI result would overstate what the available evidence supports.

The comparison with utilities creates another measurement problem. A broad utility fund holds regulated electric companies, independent power producers, and businesses with different regional exposures.

Some serve areas with enormous data center pipelines. Others operate where AI demand has little immediate effect.

Utility performance also depends on weather, fuel expenses, storm costs, wildfire liability, regulatory outcomes, and interest rates. AI represents only one part of the earnings picture.

The infrastructure side carries its own concentration risks. Several highly valued equipment suppliers can pull a diversified fund upward when investors expect persistent shortages.

If manufacturing capacity catches up, those shortages can ease. Customers might then receive equipment faster while suppliers lose some pricing leverage.

Project delays create a different problem. Orders can be postponed when developers fail to secure interconnection agreements, permits, financing, or suitable land.

The IEA estimates that grid constraints threaten a meaningful share of planned data center projects with delay. A delay does not always cancel demand, but it pushes revenue further into the future.

Efficiency creates another uncertainty. New processors can perform more AI work per unit of electricity, reducing the energy required for a specific task.

However, lower computing costs can stimulate more usage. This rebound effect occurs when efficiency makes a service cheaper, causing total consumption to rise despite lower unit requirements.

Nobody can yet measure the final balance with confidence. AI agents, video generation, scientific computing, and enterprise inference could expand demand faster than hardware efficiency improves.

Financing also matters. Data centers require large commitments before operators know the lifetime value of future AI workloads.

Technology companies with strong cash flow can support extensive construction. Smaller developers often depend more heavily on loans, leases, or long-term customer contracts.

Higher financing costs raise the return required from every project. A facility that looks attractive under optimistic utilization assumptions can become marginal when construction runs late.

Supply-chain companies face this demand quality risk indirectly. A full order book looks encouraging, but investors still need to examine cancellations, customer concentration, and delivery schedules.

Utilities face a related risk through load forecasts. Building a new substation for speculative demand can burden customers if the expected data center never arrives.

Regulators are responding with stricter contract requirements. Those protections can improve utility economics, but they can also make some locations less appealing to developers.

Local opposition adds further uncertainty. Communities increasingly question water consumption, land use, backup generators, transmission corridors, and the effect on electricity bills.

These concerns can extend approval timelines. They can also shift development toward regions with available generation, faster permitting, or more favorable rate structures.

The headline’s 40% figure should therefore be read as market evidence, not operational proof. It shows where investors have placed greater confidence, not where every future profit will appear.

The reported gain may be justified if infrastructure shortages persist and AI investment remains high. It may also contain expectations that leave little room for normal construction setbacks.

The Real Opponents Are Capital Speed and Regulatory Speed

The decisive contest is not builders against utilities, but fast technology spending against slow physical and regulatory systems.

Major technology companies can approve computing budgets quickly. Their data center partners can select sites and begin procurement before all grid work is complete.

Electric infrastructure moves on a different schedule. Transmission lines require routing, permits, equipment, construction crews, environmental reviews, and coordination across jurisdictions.

Generation projects add fuel supply, interconnection, market, and emissions considerations. Nuclear projects face especially long development periods, while gas turbines and transformers confront manufacturing constraints.

Renewables can often be developed faster, but they require storage, transmission, flexible demand, or dispatchable generation to provide continuous service.

Data centers need high reliability throughout the day. That requirement makes annual renewable-energy matching different from physical hour-by-hour power availability.

A company can purchase enough renewable energy over a year while still relying on gas, nuclear, storage, or grid imports during particular hours.

On-site generation offers another path. Fuel cells, gas turbines, batteries, and microgrids can reduce dependence on delayed grid connections.

A microgrid is a local electrical network that can operate with the wider grid or independently. It can improve resilience, but it adds equipment, fuel, maintenance, and regulatory questions.

Some developers may also accept flexible service agreements. Under these arrangements, a data center reduces consumption during grid emergencies or shifts computing workloads to another region.

Flexible computing could change the economics of grid connection. Training jobs that do not require immediate completion can move toward periods with more available electricity.

Real-time services are harder to interrupt. Search, enterprise applications, and user-facing AI products need predictable response times, limiting how much load can shift.

This technical distinction matters for both sides of the investment comparison. Utilities benefit when flexible customers allow existing assets to serve more demand.

Equipment suppliers benefit when developers instead build redundant private systems. The chosen reliability strategy directs spending toward different parts of the chain.

Google News coverage can flatten these distinctions into a single AI power narrative. The physical market is more fragmented.

One region may need new transmission. Another may have available generation but insufficient substations. A third may face water limits or community opposition.

That fragmentation favors companies able to serve many projects and geographies. It also prevents any single fund from tracking the theme perfectly.

PAVE owns companies positioned across the construction chain, but it does not isolate grid hardware. A smart-grid fund may offer more concentrated equipment exposure, alongside greater company and valuation risk.

A utility fund owns the electricity-delivery layer, but regulated holdings can lag during periods of high interest rates. Independent power producers can behave more like commodity or growth stocks.

Semiconductor funds capture accelerator demand but miss much of the surrounding facility. Data center real-estate funds capture buildings and leases but face power availability and financing constraints.

The AI infrastructure stack therefore includes several investable layers, each with different timing. Chips can be ordered before power is available, while grid assets may operate for decades after a server generation becomes obsolete.

The 8%-versus-40% reversal shows which timing the market currently prefers. Investors appear to value near-term order growth more than the slower expansion of regulated assets.

That judgment can change. Once factories add capacity and equipment shortages ease, the scarcity premium may move toward energized sites with secured power.

Utilities serving proven data center clusters could then capture more value. Their advantages include land access, grid knowledge, existing rights of way, and established regulatory relationships.

The builders have led while construction remains the urgent problem. The operators become more important when the question shifts from building capacity to earning durable returns from it.

Three Signals Matter More Than the Next Google News Headline

The next phase will depend on completed connections, order quality, and protection for ordinary electricity customers.

The first signal is actual data center energization. Announced megawatts matter less than facilities receiving power and beginning commercial operation.

Energization proves that equipment arrived, construction finished, regulators approved service, and the customer accepted the connection. It converts a speculative pipeline into measurable electricity demand.

Utility earnings reports should distinguish signed agreements from active load. Investors should also watch whether new customers meet minimum payment obligations.

A widening gap between announced projects and connected load would weaken the infrastructure thesis. A steady stream of completed campuses would support both builders and utilities.

The second signal is the quality of supplier backlogs. Electrical equipment and construction companies should explain order growth, cancellation rates, lead times, and customer concentration.

Long lead times support revenue visibility only when customers remain committed. Falling lead times can indicate improved supply, weaker demand, or both.

Margins also provide evidence. Strong volumes with stable margins suggest healthy execution, while rising sales and falling margins can reveal cost pressure or project problems.

Backlog growth across utilities, hyperscalers, and colocation providers would strengthen the case. A backlog driven by a few speculative developers would deserve more caution.

The third signal is the spread of protective large-load tariffs. Regulators must decide who pays for generation and grid upgrades created by data centers.

Contracts with minimum terms and direct infrastructure contributions can protect households from abandoned projects. They can also give utilities greater confidence to invest.

Weak protections increase the risk of cost shifting. Residential and small-business customers could face higher bills for assets built around demand that never materializes.

Tariffs that are too strict create another risk. Developers can move projects toward states offering faster connections or more favorable terms.

The most durable framework will balance speed with accountability. Data center customers should carry the costs they create, while utilities should provide clear and realistic connection schedules.

These three signals cover the entire mechanism behind the headline. Energization measures completed demand, supplier disclosures measure construction quality, and tariffs measure who carries the financial risk.

They also determine whether the current market reversal continues. Infrastructure suppliers should retain their advantage if projects move forward and equipment remains scarce.

Utilities can close the gap if connected load accelerates and regulators approve fair cost recovery. Both groups can disappoint if AI capacity commitments exceed profitable usage.

That final uncertainty reaches beyond electricity. AI services must eventually generate enough economic value to support chips, buildings, cooling, networks, and long-lived power assets.

Readers following Google News should therefore look past the next percentage comparison. Fund returns summarize expectations, but operating evidence determines whether those expectations survive.

Watch completed power connections, supplier backlog quality, and large-load tariff decisions during the next several months. Together, they will reveal whether AI infrastructure is becoming a durable industrial cycle.

They will also show whether the reported 40% leader truly owns the bottleneck, or merely received the market’s highest expectations first.

The useful question is no longer which fund appeared in a Google News headline. It is which companies can convert AI demand into completed, paid, and reliably powered facilities without transferring excessive risk to customers.