Berkshire CEO Greg Abel Sees Energy Opportunity in AI Power Demand

Berkshire Hathaway CEO Greg Abel has identified a constraint that reaches far beyond Google News headlines: AI data centers need vast amounts of reliable electricity. That demand is creating a significant opportunity for Berkshire Hathaway Energy, even as infrastructure costs, regulation, and grid limits threaten the opportunity.

Abel’s argument connects two unusually large Berkshire positions. The conglomerate owns regulated utilities across several western and midwestern states. It also ended June with nearly 106 million Alphabet shares, then its third-largest common stock holding.

That combination makes Berkshire more than an investor watching the AI boom from the sidelines. It has exposure to both the companies buying computing capacity and the physical networks supplying that capacity. The reversal is that electricity, not the latest model or chip, increasingly determines where AI infrastructure can expand.

Google News Highlights Berkshire’s Two-Sided AI Position

Berkshire is approaching AI through both ownership and infrastructure, giving it exposure to demand on opposite sides of the meter.

Abel discussed the strategy during a CNBC interview from Tokyo on September 2. According to the AI power report, he called Alphabet a significant AI participant and described electricity as a constraint on data center development.

Berkshire’s Alphabet investment gives it direct exposure to an AI infrastructure buyer. The company held almost 106 million Alphabet shares at the end of June, valued at approximately $37.8 billion. Apple and American Express were its only larger common stock positions at that time.

The size of the position changed sharply during 2026. Berkshire bought roughly 48.1 million Alphabet shares during the second quarter, according to the company’s disclosed portfolio. The total had stood at only 17.8 million shares at the end of 2025.

Berkshire also authorized an additional $10 billion investment in Alphabet three months before Abel’s interview. Alphabet planned to use new capital for the computing infrastructure behind its AI services, including the servers, networking equipment, buildings, and energy systems supporting them.

The connection is strategically important. Alphabet and other hyperscalers, companies operating enormous cloud-computing networks, need access to firm electrical capacity before a planned data center can become useful. Securing chips does not solve that problem when the local grid cannot serve the completed facility.

Berkshire Hathaway Energy sits on the other side of that demand. Its subsidiaries include utilities, interstate natural gas pipelines, renewable energy assets, and electricity transmission infrastructure. These are slower-moving businesses than software, but their assets can remain in service for decades.

In Iowa, data centers represented about 8% of Berkshire Hathaway Energy’s load during the previous year, Abel said. That figure turns a broad AI narrative into an operating-business issue. Data center demand is already large enough to affect utility planning, rather than remaining a distant forecast.

Abel summarized the tension directly by saying he had long viewed energy as the constraint. His emphasis matters because he spent much of his career in Berkshire’s energy operations before becoming CEO in January 2026.

The story therefore is not simply that Berkshire likes Alphabet. Nor is it only that a utility expects more electricity sales. Berkshire is positioned between the AI industry’s capital spending and the physical limits that spending encounters.

That is why the Google News framing captures only the visible part of the event. The deeper shift is the convergence of technology investment, utility regulation, and long-duration infrastructure inside one conglomerate.

AI Data Centers Turn Electricity Into the Scarce Input

AI infrastructure is moving the competitive bottleneck from computing equipment alone toward generation, transmission, and dependable grid connections.

A data center requires more than an annual quantity of electricity. It needs power at the correct location, at predictable quality, and often on a continuous schedule. Large facilities also need backup systems and enough network capacity to withstand maintenance or equipment failures.

Those requirements make power availability a site-selection factor. A developer can purchase servers globally, but it cannot instantly create a substation, high-voltage line, gas turbine, or regional transmission corridor. Many of those assets require permits, regulatory approvals, equipment reservations, and years of construction.

The national demand outlook supports Abel’s concern. A federal energy study estimated that data centers consumed about 4.4% of United States electricity in 2023. It projected a share between 6.7% and 12% by 2028.

The same study estimated that annual data center consumption rose from 58 terawatt-hours in 2014 to 176 terawatt-hours in 2023. Its 2028 scenarios ranged from 325 to 580 terawatt-hours. One terawatt-hour equals one billion kilowatt-hours.

Newer projections remain uncertain, but they point in the same direction. The International Energy Agency expects data centers to account for nearly half of United States electricity-demand growth through 2030. It also expects global data center consumption to more than double by that year.

AI is not the only source of data center demand. Cloud services, video delivery, enterprise applications, and ordinary internet traffic also require computing infrastructure. However, AI training and inference, the process of generating answers from trained models, have raised the density and urgency of new capacity requirements.

The pressure is especially intense in local markets. National electricity production can look adequate while a specific metropolitan area lacks transmission capacity. Data center developers must then wait, pay for upgrades, move to another location, or pursue their own generation.

This geographic concentration explains why Berkshire’s existing utility territories matter. MidAmerican Energy serves a large portion of Iowa, where major cloud companies have established facilities. PacifiCorp operates across western states where developers are seeking additional capacity.

Electricity demand can generate several forms of value for a utility. It can increase energy sales, justify investment in new infrastructure, and expand the regulated asset base. Regulators generally allow utilities to earn an approved return on qualifying investments, subject to customer protections and performance requirements.

However, more demand does not automatically mean better economics. A utility can spend heavily on infrastructure that becomes underused if a project is delayed or canceled. Equipment costs can rise during construction, while regulatory decisions can limit cost recovery.

AI companies also have alternatives. They can select other utility territories, build facilities near dedicated generation, sign agreements with independent power producers, or add on-site energy resources. Efficiency gains in chips and models could reduce consumption per computation, even if total usage continues growing.

The constraint is therefore commercial and institutional, not merely technical. Utilities must offer credible timelines while protecting existing customers. Developers must make commitments strong enough to support projects whose useful lives can exceed the current AI investment cycle.

For Berkshire, the opportunity begins where those commitments become enforceable contracts.

Berkshire’s Advantage Depends on Who Pays for the Grid

The central contest is not Berkshire against another utility, but hyperscaler growth against the obligation to protect ordinary electricity customers.

Abel has presented cost allocation as a nonnegotiable condition. Berkshire Hathaway Energy wants to serve incremental data center demand, but it does not want households and smaller businesses financing infrastructure built for hyperscalers.

That position appears throughout the energy subsidiary’s planning materials. PacifiCorp has added charges for unused demand and tightened customer funding obligations in Oregon, Utah, and Wyoming. It has also reduced certain construction credits and strengthened financial-security provisions.

These mechanisms address a basic mismatch. A hyperscaler can change its technology plan within a few years. A utility may build a transmission line or generating plant expected to operate for several decades.

If a data center requests substantial capacity but uses less than promised, the utility still carries much of the infrastructure cost. Without contractual protection, regulators could eventually place some of that burden on other customers.

Berkshire’s response is to demand stronger commitments before construction begins. Those protections can include upfront contributions, minimum-demand charges, credit support, and contracts that assign project-specific costs to the large customer.

The company’s energy presentation illustrates both the opportunity and the friction. PacifiCorp said it was preparing options that could make approximately 2,000 additional megawatts available in Utah and Wyoming by 2030.

In Oregon, parties were pursuing more than 7,000 megawatts of data center load, with 784 megawatts under contract. Yet negotiations on additional sites had stalled amid disputes over implementing legislation for large electrical loads.

The Oregon rules require new data centers above 20 megawatts to cover their costs without delaying state decarbonization goals. That principle resembles Berkshire’s stated approach, but agreement over implementation remains difficult.

PacifiCorp also disclosed a dispute with Amazon Web Services. AWS had alleged failures involving service obligations, while PacifiCorp disputed those claims. The utility said it would continue fulfilling existing agreements while the matter proceeded.

That conflict is a useful pressure test for Abel’s thesis. Data centers want speed, certainty, and competitive energy costs. Utilities need commitments that regulators will accept and existing customers can tolerate.

Neither side fully controls the timetable. Hyperscalers bring large balance sheets and valuable demand, but utilities operate under public-service rules. Regulators can reject contracts, alter cost-allocation terms, or require additional environmental protections.

Local opposition adds another variable. Communities can support construction jobs and tax revenue while opposing water use, transmission corridors, emissions, or higher power bills. These concerns can slow projects even after a developer selects a site.

Berkshire’s decentralized structure does not remove those barriers. Each regulated subsidiary must work through its own state commissions, resource plans, and customer groups. A successful Iowa model cannot simply be copied into Oregon or Utah.

The company nevertheless has relevant experience. It has financed large renewable projects, managed transmission networks, operated gas pipelines, and negotiated with regulators for decades. That record can help it structure complicated agreements without depending on short-term market enthusiasm.

Capital capacity is another advantage. Berkshire reported $364.7 billion in cash and equivalents at the end of June. That reserve gives it room to consider large projects while maintaining the financial flexibility central to its operating model.

Yet cash alone does not create acceptable utility returns. A project must balance construction risk, customer commitments, regulatory approval, and the required return on invested capital. Berkshire can decline projects that fail that test.

This discipline separates the opportunity from a simple race to build. The winning utility will not necessarily announce the largest development pipeline. It will secure projects whose customers bear the relevant costs and whose infrastructure remains useful under several demand scenarios.

Alphabet Connects Berkshire’s Software Exposure to Physical Demand

Berkshire’s Alphabet holding makes the AI power thesis more concrete, but it does not establish a coordinated deal between the two companies.

The investment exposes Berkshire to the economics of AI models, advertising, cloud computing, YouTube, and digital services. Berkshire Hathaway Energy provides exposure to the electricity and infrastructure those businesses consume.

That symmetry invites speculation about strategic coordination. Publicly available information, however, does not show that Berkshire bought Alphabet shares to secure a specific energy partnership. The investment and the utility opportunity should be treated as related exposures, not one integrated transaction.

According to an investment filing summary, Berkshire’s Alphabet position grew to roughly 106 million shares during the second quarter. The holding was worth about $37.76 billion on June 30.

The expansion also marked a change in capital allocation. Berkshire had started building the position under Warren Buffett, then increased it after Abel became CEO. Abel said he and Buffett authorized the later investment together.

Alphabet’s infrastructure requirements help explain why investors are connecting the holding to Berkshire’s energy assets. AI products depend on clusters of specialized processors that generate sustained electrical and cooling loads. Their supporting campuses also require land, water strategies, fiber connections, and backup capacity.

Google has spent years improving data center efficiency and purchasing renewable energy. Efficiency, however, does not guarantee lower total demand. When the cost of computation falls, companies can run more models, serve more users, and introduce more computationally intensive features.

This is a rebound effect. Each task becomes more efficient, while aggregate electricity consumption rises because the number and scale of tasks grow faster.

Berkshire can benefit from that expansion without predicting which AI model wins. A utility earns revenue from delivered electricity and approved infrastructure, not from the relative quality of one chatbot’s answers.

That makes energy a form of picks-and-shovels exposure, but the analogy has limits. Electricity is regulated, location-specific, and politically sensitive. A utility cannot raise capacity as quickly as a cloud company can order servers.

Alphabet also has bargaining power. Large technology companies can compare several regions, negotiate special contracts, finance generation, and move future projects when utilities cannot meet their schedules.

Other hyperscalers bring the same leverage. Amazon, Microsoft, and Meta can pursue sites across multiple states. Independent power producers and infrastructure investors compete to serve them.

Competition therefore operates at two levels. Technology companies compete for AI users and computing capacity. Utilities, developers, and energy suppliers compete to provide the reliable power behind that capacity.

Berkshire is unusual because it participates financially in both layers. Still, its interests do not always align. Alphabet benefits from lower and more flexible energy costs, while Berkshire Hathaway Energy must earn adequate returns and protect other ratepayers.

That conflict is not a weakness in the thesis. It is the thesis. AI demand becomes valuable to Berkshire only when the company can convert technical urgency into durable, appropriately priced infrastructure commitments.

Readers following the story through Google News should watch that conversion, not just movements in Berkshire’s equity portfolio. The decisive evidence will appear in utility contracts, regulatory records, and completed capacity.

The AI Energy Opportunity Carries Long-Lived Risks

Berkshire’s strongest advantage, its willingness to own infrastructure for decades, also creates its most important risk when demand forecasts prove wrong.

Data center forecasts cover a wide range because several inputs remain unsettled. Model efficiency is improving. Specialized processors can perform more work per unit of energy. Workloads may shift among training, inference, and smaller models running on local devices.

Demand can also grow faster than efficiency. More users, automated agents, generated video, scientific computing, and enterprise adoption can expand total computation. Forecasting electricity needs requires assumptions about all these variables.

The global energy outlook projects data center electricity consumption of about 945 terawatt-hours in 2030. It expects the United States to remain the largest source of growth, while acknowledging bottlenecks and macroeconomic headwinds.

Utilities cannot wait for perfect certainty before planning. Transformers, turbines, transmission equipment, and substations often have long procurement periods. Starting too late risks losing development to another region. Starting too early risks creating unused assets.

Contracts can reduce that danger, but they cannot eliminate it. A customer can encounter financing trouble, redesign a campus, dispute service terms, or seek regulatory relief. A utility may still face construction overruns and legal challenges.

Generation choices bring additional risks. Natural gas can provide dispatchable power, meaning output that operators can call upon when needed. It also creates fuel-price exposure and emissions concerns.

Renewable projects can add capacity more quickly in some markets, but they require storage, transmission, or complementary generation for continuous service. Nuclear and geothermal projects attract technology-company interest, yet many proposals remain years from commercial operation.

Transmission may become the hardest component. New generation does little for a data center when congestion prevents electricity from reaching its site. Large transmission projects cross jurisdictions, require land access, and often face extended permitting processes.

Berkshire’s western utility operations carry wildfire exposure as well. PacifiCorp has invested more than $1.9 billion in prevention measures through 2025 and planned about $1.5 billion more from 2026 through 2028. Those obligations compete for capital and affect the risk regulators consider when setting returns.

Affordability presents a separate test. Public support can erode if residents believe data centers are raising household bills. Large-load contracts must therefore withstand more than a financial review. They must also survive political and consumer scrutiny.

Environmental concerns can influence approval. A new data center may increase demand during periods when utilities already struggle to meet peak usage. Additional fossil generation can conflict with state emissions targets, while rapid renewable development can require new land and transmission.

Water use matters in some regions because data centers may rely on evaporative cooling. The significance varies by design and location, but local scarcity can become a permitting issue.

There is also a concentration risk. Four large hyperscalers can create substantial demand, yet their combined bargaining power is formidable. A utility dependent on a few customers must structure contracts for low-probability but high-impact changes.

Abel’s 8% Iowa load estimate shows meaningful demand, not guaranteed profit. Investors still need to know the margin, required capital, contract duration, regulatory treatment, and effect on existing customers.

Those details are not available from a broad CEO statement. Until projects move through regulatory review, the opportunity remains partly a pipeline rather than an accomplished result.

The skeptical interpretation is straightforward. AI enthusiasm may encourage utilities to overbuild just as technology becomes more efficient or development moves elsewhere. Berkshire’s discipline must prevent projected demand from turning into stranded infrastructure.

The more optimistic interpretation is equally testable. If hyperscalers sign long contracts, fund their share of construction, and accept minimum-demand obligations, Berkshire can add infrastructure with unusually visible usage.

The outcome will depend less on headline demand forecasts than on contract quality.

Three Signals Will Show Whether Abel’s Thesis Is Working

Berkshire’s AI energy strategy becomes credible when proposed megawatts turn into protected contracts, approved investments, and measurable load growth.

The first signal is the conversion of PacifiCorp’s development pipeline into binding agreements. The company has identified approximately 2,000 megawatts of possible additional capacity in Utah and Wyoming. It has also described more than 7,000 megawatts of interest in Oregon.

Those figures measure opportunity, not completed demand. Investors should separate preliminary inquiries from signed contracts and operating facilities. The 784 megawatts already under contract in Oregon offers a more concrete baseline.

Watch for agreements that disclose customer contributions, minimum payments, security requirements, and construction milestones. Strong protections would reinforce Abel’s argument that growth can benefit Berkshire without shifting risk to other customers.

The second signal is regulatory treatment. State commissions must decide whether large-load contracts protect ratepayers and fit resource plans. Approvals will reveal whether Berkshire’s preferred cost-allocation model works outside Iowa.

Oregon deserves particular attention because its rules require large data centers to cover their costs without delaying decarbonization. Resolution of the PacifiCorp and AWS dispute would also clarify how quickly the utility can serve new projects under contested conditions.

Favorable decisions would strengthen Berkshire’s ability to deploy capital. Delays, rejected contracts, or unfavorable cost assignments would weaken the near-term thesis, even if demand remains strong.

The third signal is reported load growth in Iowa and other Berkshire territories. Abel said data centers accounted for about 8% of Iowa load in the previous year. Future disclosures can show whether that share rises and whether the change translates into approved earnings.

Load percentage alone is insufficient. Investors should examine capital spending, customer concentration, rate outcomes, and the timing between infrastructure investment and revenue. Faster growth with weak protections would not validate Berkshire’s strategy.

Alphabet’s capital spending provides supporting context, but it is not the decisive measure. More spending by Google can increase demand for computing infrastructure while benefiting a utility in another territory. Berkshire needs projects connected to its own networks.

The energy business also must maintain reliability. Data center growth will look less attractive if it contributes to service problems, exposes customers to volatile costs, or diverts investment from aging infrastructure.

For developers and enterprise buyers, the lesson reaches beyond Berkshire. AI capacity planning now includes energy availability, utility negotiations, and regional infrastructure risk. A computing roadmap that ignores those variables can fail before the first server operates.

Knowledge workers and AI users should care because infrastructure constraints influence product availability and operating costs. They can also affect where companies build services and how aggressively those services expand.

The next Google News headline may focus on another model, chip, or investment. The more consequential evidence will arrive slowly through commission filings, power contracts, construction schedules, and utility load reports.

Abel has made Berkshire’s position clear: energy is both the constraint and the opportunity. Now the company must show that it can serve hyperscalers without asking ordinary customers to underwrite the bet.

That is the standard readers should apply over the next several quarters. Are proposed megawatts becoming protected, operating demand, or are they remaining optimistic entries in a development pipeline?