Rising Energy Costs Squeeze AI Infrastructure Stocks

Google News has surfaced a warning for AI infrastructure investors: soaring computing demand does not guarantee equally strong profits for every supplier.

The Simply Wall St headline focuses on rising energy costs, but the pressure extends beyond electricity bills. Grid access, cooling, backup generation, financing, and equipment availability now shape the economics of new data center capacity.

That changes the investment question. Demand for AI servers remains strong, yet the businesses supporting them face sharply different exposure to power costs. Some can pass those costs to customers. Others must absorb them while financing facilities years before revenue arrives.

The divide separates companies selling scarce equipment from operators carrying long-lived energy obligations. Dell, Vertiv, Delta Electronics, and other suppliers appear alongside developers and power specialists in the broader AI stock screen. Their revenue drivers and cost structures are not interchangeable.

This is not a claim that AI infrastructure spending has peaked. It is a warning that electricity has become a constraint, not a background expense. Investors now need to distinguish exposure to AI demand from control over the energy needed to serve it.

What the Google News Headline Actually Signals

The important change is that power availability has become part of the AI infrastructure product.

The Google News headline originated from an investment analysis published by Simply Wall St. It points readers toward companies supporting AI computing while emphasizing an increasingly visible cost problem.

That framing matters because earlier stages of the AI investment cycle centered on accelerator supply. Nvidia chips, high-bandwidth memory, networking equipment, and server manufacturing dominated the discussion.

Electricity often appeared as a secondary operating input. Investors could treat it like rent, labor, or another predictable facility expense.

That assumption no longer fits the largest AI campuses. An AI-focused data center can require a dedicated substation, transmission upgrades, backup systems, and new generation capacity.

It can also spend years waiting for an interconnection agreement. An interconnection agreement defines how a large electricity user connects to the power grid and pays for required upgrades.

The International Energy Agency reported that data center electricity demand grew 17% during 2025. It also said five major technology companies spent more than $400 billion on capital projects that year.

Their combined capital spending was expected to rise another 75% during 2026, according to the agency’s AI power outlook.

Those figures show why the story is not merely about utility rates. Large technology companies are trying to secure computing capacity faster than power systems can expand.

A data center developer can have land, financing, customers, and construction plans without possessing enough electricity to operate. “Time to power” therefore becomes a commercial measure, meaning the period before a facility receives usable grid capacity.

That delay affects returns. Interest accumulates while a project remains unfinished, equipment can sit unused, and customer commitments can become harder to fulfill.

The headline also arrives during a broader reassessment of AI-related stocks. Investors have begun separating companies with current earnings from those whose valuations rely on distant capacity targets.

Google News reflects that shift because stories about power shortages now sit beside earnings reports and chip announcements. Energy risk has moved into mainstream AI market coverage.

However, the headline should not be treated as a complete investment thesis. It identifies a pressure point without showing which companies bear the cost or how contracts allocate it.

That distinction determines whether higher electricity costs reduce margins, raise customer prices, or create more demand for power-management suppliers.

AI Demand Is Growing Faster Than the Grid

The AI buildout is colliding with an electricity system designed for slower, more predictable demand growth.

The United States experienced little electricity demand growth for much of the period after 2005. Utilities could plan around gradual household, commercial, and industrial changes.

That pattern has ended. The U.S. Energy Information Administration found that electricity demand grew about 1.7% annually between 2020 and 2025.

The comparable annual rate between 2005 and 2019 was only 0.1%. Data center consumption is now a major driver of that acceleration, according to the agency’s 2026 forecast.

This growth affects more than total energy production. Data centers cluster in specific regions, so their local impact can exceed their national share.

A new campus may request a load comparable with a power-intensive factory. Several proposed campuses can arrive within one utility territory during the same planning cycle.

Utilities then need generation, substations, transformers, and transmission lines in the same location. Those assets require permits, equipment, construction crews, and regulatory approval.

The concentration problem explains why sufficient national generation does not guarantee a timely connection. Electricity must reach a particular site at the required voltage and reliability level.

AI facilities also want near-continuous service. A short interruption can disrupt computing jobs, damage equipment availability commitments, and reduce expensive server utilization.

Operators address that risk through redundant feeds, batteries, generators, and sometimes dedicated power plants. Each layer adds capital cost before it improves revenue.

The U.S. Department of Energy reported that data centers consumed about 4.4% of national electricity during 2023. Its data center study projected a possible share between 6.7% and 12% by 2028.

The range is wide because future consumption depends on deployment speed, equipment efficiency, server utilization, and the composition of computing workloads.

Even the lower estimate implies a meaningful system change. The upper estimate would require utilities to support an industrial-scale demand expansion within a short period.

Global forecasts point in the same direction. Gartner projected worldwide data center electricity consumption of 565 terawatt-hours during 2026, compared with 447 terawatt-hours in 2025.

It also estimated that AI-optimized servers would account for 31% of data center power use during 2026. Their consumption was expected to surpass conventional servers by 2027.

These forecasts carry uncertainty, but operators must make investment decisions before actual demand becomes known. They cannot build a substation after customers need it.

That creates two opposing risks. Building too slowly can leave profitable demand unserved, while building too aggressively can strand capital if AI demand disappoints.

Investors should watch how companies manage that uncertainty. Announced megawatts are less informative than energized capacity supported by signed customers and credible power arrangements.

The grid constraint can benefit selected suppliers. Transformer manufacturers, switchgear companies, cooling providers, and electrical contractors can receive orders regardless of which AI model wins.

Yet the same constraint can hurt developers promising rapid campus expansion. Every delayed connection pushes expected revenue further into the future.

The Real Divide Is Cost Control Versus Cost Exposure

AI infrastructure winners will be defined by contract structure and power control, not by proximity to the AI label.

The Simply Wall St list covers businesses across servers, storage, networking, electrical equipment, thermal management, and data center development.

Those companies participate in the same spending cycle, but they do not carry the same energy risk.

A server manufacturer generally sells equipment to a customer. Electricity affects product design and customer demand, but the manufacturer does not usually pay the server’s lifetime utility bill.

A cooling supplier occupies a similar position. Higher rack density increases demand for thermal management, although component and manufacturing costs still affect margins.

Vertiv, Delta Electronics, nVent, and Advanced Energy Industries sit closer to this supplier model. Their products help facilities distribute power, convert electricity, and remove heat.

Rising energy costs can strengthen the case for efficient equipment. Customers become more willing to pay for systems that reduce wasted electricity or support denser computing.

That does not make suppliers immune. Competition can compress prices, customers can delay projects, and manufacturing capacity can overshoot demand.

However, their exposure differs from a data center operator that owns a facility and purchases electricity continuously. The operator’s economics depend on lease terms, utilization, financing, and energy procurement.

A long-term customer contract can protect the operator if it passes utility costs through. A fixed-price contract can create risk when electricity prices rise faster than expected.

Contract duration matters as well. A multiyear commitment improves revenue visibility but can lock a company into assumptions made before construction costs changed.

Some developers pursue behind-the-meter power, meaning generation located on or near the customer’s site. This approach can reduce dependence on a congested public connection.

It can also introduce fuel risk, environmental permitting, equipment maintenance, and local opposition. Owning generation changes a digital infrastructure company into a partial energy operator.

Public utilities occupy another position. Data center demand can expand their rate base, which represents infrastructure approved for recovery through regulated customer charges.

However, regulators decide which costs enter that rate base and which customers pay them. A utility cannot assume households will absorb every upgrade serving a private campus.

Federal regulators have tried to accelerate connections for large loads. Recent grid connection rules addressed faster access for projects that can consume as much power as smaller cities.

Faster procedures do not create transformers or generation overnight. They primarily clarify planning, cost allocation, and the treatment of exceptionally large customers.

Investors therefore need to ask a practical question: who pays when the grid must expand?

If hyperscalers pay through dedicated agreements, infrastructure developers and utilities receive stronger protection. If costs enter general utility rates, political and regulatory resistance becomes more likely.

The answer varies by jurisdiction. It also changes as regulators respond to household affordability concerns.

Power procurement quality deserves the same attention as customer announcements. A signed AI contract has limited value when the operator lacks a firm path to energization.

Conversely, secured power without a creditworthy customer can leave an expensive site underused. The strongest projects align land, electricity, financing, and demand.

This is why broad AI infrastructure baskets can hide risk. Two companies with similar revenue growth can possess completely different obligations after a facility opens.

Google News coverage tends to group them under one theme. Financial analysis must separate their cash-flow mechanics.

What Rising Energy Costs Can Do to AI Infrastructure Stocks

Higher power costs affect valuation through margins, project timing, financing needs, and customer behavior.

The most direct effect appears in operating margins. A company that absorbs electricity costs earns less from each unit of computing sold when utility expenses rise.

The company can raise prices, but customers may respond by reducing usage or moving workloads. AI training is important, yet some jobs can wait for cheaper locations or hours.

Inference workloads create a different pattern. Inference is the process of running a trained model to generate answers, images, or decisions.

Consumer and business applications often need rapid responses. That requirement limits the operator’s ability to pause computing during expensive periods.

Some workloads remain flexible. Model training, data processing, and selected batch jobs can shift across time or geography if software supports that movement.

Flexible computing can reduce power costs and help balance the grid. It requires compatible infrastructure, customer consent, and enough network capacity between locations.

Efficiency improvements offer another defense. Better accelerators can complete more work per unit of electricity, while liquid cooling can support dense racks more effectively.

Yet efficiency does not automatically reduce total consumption. Lower computing costs can stimulate more use, offsetting savings achieved by each server.

This rebound effect complicates forecasts. Chips can become more efficient while data center electricity demand continues rising.

Project delays create a second valuation channel. A facility expected to begin earning revenue this year may slip into the next reporting period.

That delay reduces near-term cash flow and can require additional financing. It also increases the risk that customers renegotiate or select another site.

Financing becomes especially important for smaller developers. Large hyperscalers can fund projects from substantial operating cash flow, although shareholders still scrutinize returns.

Smaller operators often depend on debt, equity issuance, joint ventures, or customer prepayments. Rising construction expenses can increase dilution or leverage before revenue stabilizes.

Energy costs also influence location decisions. Regions with available generation, supportive regulators, and faster permits gain an advantage over congested markets.

Cheap electricity alone is insufficient. Operators also need fiber connectivity, water or alternative cooling systems, construction labor, and acceptable political conditions.

The IEA found that the United States represented 45% of global data center electricity consumption during 2024. China held 25%, while Europe accounted for 15%.

Those shares show the scale of geographic concentration. They also reveal why national energy policy increasingly shapes AI competition.

Still, investors should resist one-sided conclusions. Rising electricity demand can create revenue for utilities and equipment suppliers while increasing costs for data center operators.

It can also support power producers that sign long-term agreements with technology companies. The relevant outcome depends on regulation, fuel mix, and contract design.

The skeptical case centers on demand durability. Current capacity plans assume that AI services will create enough revenue to justify sustained infrastructure expansion.

If models become cheaper to operate, demand might grow enough to absorb that capacity. If customers resist paying for AI services, utilization could fall below projections.

A second uncertainty concerns duplicate requests. Utilities sometimes receive connection applications from developers exploring several sites for one eventual project.

Those applications can make future demand appear larger than committed construction. Investors should prefer projects with deposits, permits, equipment orders, and customer obligations.

A third uncertainty involves political tolerance. Communities increasingly question water use, emissions, land requirements, and electricity-rate effects.

Public opposition can delay facilities even when technical demand remains strong. It can also change which party pays for grid upgrades.

None of these risks invalidates the AI infrastructure thesis. They determine which companies convert demand into durable returns.

How to Evaluate AI Infrastructure Stocks Under Power Pressure

Investors should replace headline megawatts with evidence of energized, contracted, and economically protected capacity.

The first useful metric is energized capacity. This measures the power already available to operating equipment, not the size of a future development pipeline.

A company may control land capable of supporting a large campus. That land does not generate AI revenue until electricity, buildings, cooling, and servers are ready.

The second metric is contracted capacity. Investors should examine customer credit quality, contract duration, cancellation rights, and the treatment of energy costs.

A long contract with a strong customer can support financing. It becomes less attractive if the operator accepted a fixed rate without protection against utility increases.

The third metric is the construction schedule. Management should identify permits, interconnection milestones, equipment delivery dates, and expected commissioning periods.

Repeated delays deserve attention, even when management continues highlighting total pipeline size. A delayed megawatt is not equivalent to an operating megawatt.

The fourth metric is capital intensity. Investors should compare spending requirements with operating cash flow, available financing, and future debt maturities.

High growth can destroy shareholder value when each new project requires frequent equity issuance. Revenue growth alone does not reveal that tradeoff.

The fifth metric is customer concentration. A developer serving one hyperscaler can enjoy visible demand but also face substantial negotiation risk.

Large customers understand their leverage. They can demand performance guarantees, favorable renewal terms, and strict completion schedules.

Suppliers face concentration too. An equipment maker can grow quickly because one customer accelerates orders, then suffer when that customer pauses spending.

The sixth metric is backlog quality. A backlog records expected orders or contracted work, but accounting definitions differ across companies.

Investors should determine whether customers can cancel, whether deposits exist, and when backlog converts into revenue.

The seventh metric is efficiency. Power usage effectiveness compares a facility’s total electricity consumption with the power delivered to computing equipment.

A lower ratio generally indicates less overhead from cooling and electrical systems. Comparisons still require caution because climate, facility age, and measurement practices differ.

Investors should also examine revenue per energized megawatt and cash flow per megawatt. These measures help reveal whether new capacity improves economics.

No single metric works across the entire theme. Equipment companies require analysis of orders, margins, inventories, and manufacturing capacity.

Utilities require attention to regulatory approvals, rate cases, and capital recovery. Operators require deeper analysis of occupancy, leases, electricity, and financing.

Power producers require fuel and contract analysis. Server manufacturers require examination of component availability, customer mix, and working capital.

This segmentation prevents a common error. A favorable forecast for data center electricity demand does not make every company attached to that forecast attractive.

It can even create opposing outcomes. Higher demand raises utility investment while weakening an operator whose contract prevents energy-cost recovery.

The Google News story is useful because it forces this distinction. It should start due diligence, not replace it.

Readers following several companies can maintain a structured record of earnings claims, construction milestones, and regulatory decisions. A searchable AI knowledge base can help connect those updates across reporting periods.

The goal is not to predict every electricity-price movement. It is to identify which business models remain viable when power becomes more expensive or delayed.

Three Signals Investors Should Watch Next

The next phase will be decided by interconnection progress, contract disclosure, and evidence that AI revenue supports infrastructure spending.

The first signal is actual grid connection progress. Investors should watch utilities, regional grid operators, and data center developers for completed interconnection agreements.

They should also track transformer deliveries, substation construction, and confirmed energization dates. These milestones carry more weight than conceptual campus announcements.

Faster connections would strengthen the view that current infrastructure pipelines can become operating assets. Repeated delays would weaken revenue forecasts tied to near-term capacity.

EIA’s server energy outlook adds another reference point. It projects rising server electricity use across commercial buildings, with stronger growth at standalone facilities.

The second signal is clearer cost allocation. Earnings calls and regulatory filings should explain whether developers, hyperscalers, utilities, or general ratepayers fund required upgrades.

Contracts that pass electricity costs to customers can protect operator margins. They do not eliminate risk because aggressive pricing may reduce customer demand.

Regulators may also require large users to post financial security before utilities build dedicated infrastructure. Such rules would reduce stranded-asset risk for other customers.

Weak disclosure should remain a warning. Management teams that promote megawatts without explaining power costs leave investors unable to model returns.

The third signal is the relationship between hyperscaler spending and AI revenue. Infrastructure growth ultimately depends on customers earning acceptable returns from AI products.

Investors should watch cloud growth, accelerator utilization, AI service revenue, and management commentary about capital efficiency.

Continued spending supported by rising usage would strengthen the long-term demand case. Spending cuts or persistent underutilization would pressure suppliers across the chain.

These signals should be evaluated together. A signed customer without power is incomplete, while secured power without profitable demand is equally fragile.

Google News will continue surfacing large data center announcements because their scale attracts attention. The harder work begins after the headline.

Ask whether the project has deliverable electricity, a binding customer, protected margins, and financing that does not overwhelm future cash flow.

AI infrastructure remains one of technology’s largest investment cycles. Energy does not end that cycle, but it changes where value can accumulate.

The companies worth following are those that turn power scarcity into a defensible service or product. The vulnerable ones treat electricity as an unlimited input.

As new earnings reports arrive, compare promised capacity with energized capacity and contracted revenue. That comparison will reveal whether the AI buildout is producing durable businesses or expensive waiting rooms.