Alphabet Is Now a Berkshire AI Bet, but Greg Abel Says Power Is the Constraint

Alphabet has become Berkshire Hathaway’s third-largest common stock holding, despite the conglomerate’s long-standing caution toward major technology bets.

Berkshire CEO Greg Abel tied that commitment directly to Google’s position in artificial intelligence during a September 2 CNBC interview. He also identified the electricity behind AI data centers as a separate opportunity for Berkshire Hathaway Energy.

The result is more than a large portfolio adjustment. Berkshire is placing capital on both sides of the AI infrastructure buildout. It owns Alphabet shares, while its utilities can supply electricity to the hyperscale facilities supporting Google and its rivals.

That creates an unusual investment structure. Berkshire gains exposure to AI software, cloud services, and advertising through Alphabet. It can also earn regulated or contracted returns from the physical infrastructure that AI companies cannot operate without.

Abel applied a very different timeline to Berkshire’s expanding housing investments. He described Taylor Morrison as a strong long-term asset, yet said the housing market showed no sign of an immediate recovery.

These positions reveal a consistent strategy rather than two disconnected wagers. Berkshire is targeting durable demand while accepting that the payoff depends on infrastructure, execution, and time.

Berkshire Turned Alphabet Into a Core Holding

The important change is not simply that Berkshire bought Alphabet. It is the scale and stated reasoning behind the commitment.

Berkshire ended June with nearly 106 million Alphabet shares valued at about $37.8 billion. That made the company Berkshire’s third-largest common stock holding, behind Apple and American Express.

The position had grown dramatically within nine months. Berkshire held 17.8 million Alphabet shares worth $5.6 billion at the end of December 2025, according to the latest portfolio filing.

Berkshire added roughly 48.1 million shares during the second quarter of 2026. The expansion included a $10 billion investment authorized by Abel and Berkshire Chairman Warren Buffett three months before the interview.

Abel said Alphabet’s standing as a significant AI participant helped motivate that decision. His explanation matters because Berkshire historically avoided technology companies whose future economics seemed difficult to forecast.

Buffett initiated the original position in 2025. Abel said he was responsible for arranging the additional investment at a 6.5% discount to Alphabet’s stock price.

The purchase therefore combined a familiar Berkshire preference with a less familiar target. The conglomerate obtained favorable terms, but applied its value discipline to one of the largest AI infrastructure spenders.

Alphabet said the transaction formed part of a broader effort to raise $80 billion for computing infrastructure. Berkshire agreed to purchase equal amounts of Alphabet’s Class A and Class C shares.

Those investment terms connect the capital directly to Alphabet’s expanding AI capacity. They make the Berkshire Alphabet investment different from a passive purchase made quietly through public markets.

Alphabet receives funding for servers, networking equipment, and data centers. Berkshire receives equity exposure to the services that infrastructure supports, including Google Cloud and consumer AI products.

The transaction does not guarantee that Alphabet will convert AI spending into attractive profits. It does show that Berkshire considers the company’s competitive position sufficiently durable for a major commitment.

That judgment places more weight on Alphabet’s complete operating system than on any single model. Google combines custom chips, data centers, cloud distribution, research teams, consumer products, and an advertising business with global reach.

Other AI developers often depend on outside infrastructure providers or financing partners. Alphabet controls more of its stack, although that control also leaves it responsible for enormous capital requirements.

For Berkshire, that combination offers both a moat and a pressure point. Alphabet can deploy AI across established products, but it must secure enough computing capacity and energy to keep those products competitive.

The investment is consequently a wager on execution, not merely enthusiasm surrounding artificial intelligence. Berkshire is backing Alphabet’s ability to turn infrastructure spending into recurring economic value.

Alphabet’s AI Advantage Depends on Physical Capacity

Alphabet’s AI advantage matters only if the company can obtain enough computing capacity and electricity at workable costs.

Abel’s comments shifted attention from model performance to the systems underneath it. AI services require specialized chips, dense server clusters, cooling equipment, transmission capacity, and dependable electricity.

A hyperscaler is a company operating cloud infrastructure at enormous scale. Alphabet, Amazon, and Microsoft belong to that group, alongside other large data center developers.

These companies compete on models and developer services, but they also compete for power connections. A region with available land cannot support a major facility when its grid lacks generation or transmission capacity.

Abel summarized that bottleneck directly. “I’ve sort of always had the strong view that energy would be the constraint,” he said in the September interview.

That observation explains why Berkshire’s Alphabet position fits Abel’s operating background. Before becoming CEO, he led Berkshire Hathaway Energy and later oversaw Berkshire’s non-insurance businesses.

He is evaluating AI through an industrial lens. The important questions concern capacity, lead times, capital allocation, and long-term customer demand, not only benchmark results.

The International Energy Agency expects global data center electricity consumption to reach about 945 terawatt-hours in 2030. That would represent more than twice the 2024 level.

Its base case expects data center consumption to rise about 15% annually between 2024 and 2030. Accelerated servers, which support demanding AI workloads, account for almost half the projected increase.

The United States holds a central role in that forecast. Data centers are expected to produce nearly half of American electricity demand growth through 2030, according to the IEA’s energy outlook.

Yet the development schedules do not match neatly. A data center can become operational within two or three years, while major energy projects often require longer planning and construction periods.

Transmission lines face permitting, equipment, and interconnection delays. New generation also requires regulatory approval and long-term financing.

This timing mismatch changes what counts as an AI advantage. A strong model cannot serve growing demand without available compute, and compute cannot expand without a reliable power supply.

Alphabet has tried to address the problem through long-term energy procurement and investments in different generation technologies. Its competitors are pursuing similar approaches because the constraint affects the entire hyperscale market.

Microsoft, Amazon, Meta, and Oracle are also spending heavily on AI infrastructure. Each must translate that spending into products customers will use and revenue that justifies the capital.

That creates the central tension in Berkshire’s bet. Alphabet’s integrated infrastructure makes it a credible AI leader, but the same strategy exposes shareholders to large and persistent construction requirements.

Abel appears willing to accept that tension. Berkshire is not treating electricity as an inconvenient cost outside the investment thesis. It sees the constraint as another place where the conglomerate can participate.

AI Data Center Power Demand Creates Berkshire’s Second Bet

Berkshire can benefit from AI demand even when it is not operating the models or selling cloud services.

Berkshire Hathaway Energy owns utilities, transmission assets, renewable generation, natural gas infrastructure, and other energy businesses. Those operations give the conglomerate a direct route into AI infrastructure spending.

Abel estimated that data centers produced about 8% of Berkshire’s electricity load in Iowa during 2025. That figure turns a broad industry narrative into an observable demand source inside Berkshire’s system.

The company is willing to serve additional hyperscale customers, according to Abel. However, supplying them requires careful commitments because data centers can impose unusually large and concentrated loads.

A utility must have enough generation to meet peak demand. It also needs transmission and distribution equipment capable of delivering electricity continuously.

Large customers can support investments through long-term agreements. Their demand can make new generation or transmission projects economically viable when contracts allocate costs appropriately.

That qualification matters for existing customers. Utilities cannot simply build expensive infrastructure for a speculative project and spread every cost across households and small businesses.

Data center developers sometimes announce facilities before securing every permit, power agreement, or construction milestone. Utilities must distinguish credible demand from projects that will be delayed or abandoned.

The attraction for Berkshire rests on duration. A completed hyperscale data center represents a large customer that expects dependable service for many years.

Such facilities also operate continuously. Their workloads may change, but they cannot tolerate frequent interruptions or uncertain capacity.

The opportunity extends beyond selling additional kilowatt-hours. Berkshire can invest in generation, substations, transmission lines, storage, and grid upgrades tied to the new load.

Berkshire Hathaway Energy reported $891 million in second-quarter earnings, compared with $702 million one year earlier. Its first-half earnings reached $2.005 billion, according to Berkshire’s quarterly results.

Those results do not isolate data center contributions. They show that the energy business already operates at a scale where new industrial demand can influence capital planning.

Berkshire’s structure is particularly suited to long-lived infrastructure. The parent company can retain earnings and deploy capital without promising investors a quick exit.

That patience can help utilities fund assets whose returns arrive over decades. It also matches data center customers seeking dependable partners for projects with long operating lives.

Still, the electricity opportunity is not exclusive to Berkshire. Utilities across Virginia, Texas, Ohio, Arizona, and other development regions are responding to similar requests.

Independent power producers and infrastructure funds are also competing for contracts. Technology companies increasingly evaluate nuclear, geothermal, gas, renewables, and storage rather than relying on one supply route.

Berkshire’s advantage lies in combining financial capacity with regulated operations and infrastructure experience. It does not automatically possess the best location for every proposed facility.

Google’s data center geography will not always overlap with Berkshire’s utility territories. The equity investment and energy opportunity should therefore be understood as related exposures, not a guaranteed bilateral partnership.

No disclosed agreement establishes Berkshire Hathaway Energy as the preferred supplier for Alphabet’s future facilities. Abel’s comments described an industry opportunity, not a confirmed exclusive arrangement.

That distinction prevents the thesis from becoming circular. Berkshire is not necessarily funding Alphabet so that Alphabet can purchase Berkshire electricity.

Instead, it is positioning itself around the same structural demand from two directions. Alphabet offers upside if AI services grow, while Berkshire’s energy assets can serve whichever hyperscalers build within their markets.

The Real Risk Is Turning AI Spending Into Returns

Power demand can grow while the financial returns from AI infrastructure remain uncertain for both technology companies and utilities.

Alphabet plans enormous infrastructure investment because AI capacity has become necessary for competition. Necessity, however, does not establish the eventual return on every dollar spent.

Major cloud companies do not fully disclose the sales and profits produced specifically by AI data centers. Investors must infer the relationship through cloud growth, capital spending, depreciation, and management commentary.

That gap complicates Berkshire’s thesis. Alphabet can remain a leading AI company while earning disappointing returns on incremental infrastructure.

Competition may force providers to reduce model prices. Hardware performance can improve faster than expected, making some installed equipment less valuable before its planned retirement.

AI applications may also consume more computing resources than customers are willing to fund. High usage does not always translate into strong margins.

Alphabet has meaningful defenses. Its advertising operation generates cash, Google Cloud has established enterprise customers, and AI features can strengthen existing products.

Yet those advantages create another obligation. Google must defend search and advertising while financing a new computing layer that competitors are also building.

The company faces pressure from Microsoft’s cloud and AI distribution, Amazon’s infrastructure footprint, Meta’s open model strategy, and specialized model developers.

Berkshire’s decision suggests that Abel and Buffett judge Alphabet’s position favorably within that field. It does not settle which company will capture the best returns.

The energy side carries its own risks. Utilities could build for demand that arrives later than promised or changes location after contracts are discussed.

Equipment shortages and permitting delays can raise project costs. Communities may oppose new generation, transmission corridors, water consumption, or higher bills connected with data center development.

The electricity mix creates additional tension. Hyperscalers often announce carbon goals, while fast load growth can increase reliance on natural gas or extend the operation of fossil-fuel facilities.

The U.S. Energy Information Administration found that data centers are helping revive electricity growth after years of relatively flat demand. Its analysis also warned that faster growth can increase fossil generation.

Between 2020 and 2025, U.S. electricity demand grew about 1.7% annually. The comparable rate from 2005 through 2019 was only 0.1%, according to the agency’s power demand analysis.

That acceleration supports Abel’s infrastructure argument. It also raises questions about who pays, how quickly supply expands, and whether environmental commitments remain achievable.

Utilities need enforceable contracts and realistic construction schedules. Regulators need to examine whether costs remain aligned with the customers creating them.

Technology companies need enough flexibility to manage changing hardware and model requirements. Communities need evidence that new projects will deliver benefits alongside their demands on local systems.

Berkshire faces less technology-selection risk through its energy assets than through its Alphabet shares. Electricity supports many AI platforms regardless of which model leads a benchmark.

However, energy assets carry location and regulatory risks. Demand can be strong nationally while individual projects fail because they selected the wrong market or assumed an impossible connection date.

The strongest version of Berkshire’s thesis therefore depends on disciplined contracting. It does not depend on accepting every data center proposal.

Abel’s record as an operator makes that filter central to the story. Berkshire must convert a broad demand wave into projects with protected economics and credible counterparties.

Housing Shows the Other Side of Abel’s Time Horizon

Berkshire’s housing investments follow the same preference for durable demand, but Abel does not expect the cycle to improve quickly.

Berkshire agreed on May 31 to acquire Taylor Morrison for approximately $6.8 billion. The transaction closed on July 24 after shareholder and regulatory approvals.

Taylor Morrison develops residential communities and provides mortgage, title, escrow, and insurance services. It adds another large housing platform to Berkshire’s existing operations.

The conglomerate already owns Clayton Homes. Other subsidiaries sell paint, flooring, insulation, building products, brokerage services, and related goods.

Berkshire also held investments in Lennar and D.R. Horton during the second quarter. Its collection of assets gives it exposure across several stages of home construction and ownership.

That breadth does not make the business immune to interest rates or affordability pressure. Higher mortgage costs reduce purchasing power, while inflation continues to strain household budgets.

Abel said Berkshire saw no sign of an immediate housing recovery. He called the near-term path bumpy and described consumers as having to stretch each dollar further.

Single-family housing starts fell in July to their lowest level since November 2022. That decline reinforced Abel’s caution about expecting a fast cyclical rebound.

His investment horizon is longer. Abel said Taylor Morrison should become a strong asset over five to ten years as more Americans pursue homeownership.

That position resembles the energy strategy in one important respect. Berkshire is acquiring capacity before the demand environment becomes comfortable.

The differences are equally important. AI data center power demand is accelerating now, while housing remains constrained by financing conditions and consumer affordability.

A hyperscaler can sign a long-term electricity agreement based on corporate demand forecasts. A homebuilder depends on thousands of households making individual purchasing decisions.

Berkshire can influence operating efficiency and capital allocation at Taylor Morrison. It cannot determine mortgage rates, household formation, land costs, or local construction rules.

The company plans to unify its site-built homebuilding operations over time. That approach represents a modest shift from Buffett’s traditional preference for highly autonomous subsidiaries.

Consolidation can create purchasing, financing, land management, and administrative efficiencies. It can also introduce integration risks and weaken local decision-making if applied too aggressively.

Abel therefore faces a management test beyond selecting assets. He must demonstrate that a more active operating model creates value without damaging Berkshire’s decentralized culture.

The housing wager should not be presented as evidence that a rapid recovery is approaching. Abel expressly rejected that interpretation.

Instead, it shows how Berkshire uses permanent capital. The company can buy during a difficult period and wait for demographic need and supply constraints to support returns.

That patience also appears in the Alphabet position. Berkshire is not betting that every quarter of AI spending will produce immediate profit.

It is backing a company and an infrastructure need that it expects to remain relevant across a much longer cycle.

What to Watch After Berkshire’s Alphabet Commitment

Three signals will determine whether Berkshire has built a coordinated AI strategy or simply accumulated two forms of expensive exposure.

The first signal is Alphabet’s conversion of infrastructure spending into durable business growth. Investors should watch Google Cloud revenue, operating margins, and management’s discussion of AI-related demand.

Capital spending alone is not validation. The thesis strengthens when expanding capacity supports sustained customer usage without persistent margin deterioration.

It weakens if depreciation and infrastructure costs rise faster than associated revenue. That outcome would suggest that competition, rather than customer economics, is setting the investment pace.

The second signal is Berkshire Hathaway Energy’s contracting activity. Future earnings reports should clarify whether data center loads are translating into approved capital projects and dependable returns.

The most useful disclosures will concern committed capacity, project timing, customer protections, and regulatory treatment. Announced demand without those elements provides limited evidence.

Berkshire’s Iowa load offers a starting benchmark. If the data center share rises from the 8% Abel cited, investors should also examine how the utility funds generation and grid expansion.

The thesis strengthens when hyperscalers accept long-term commitments that protect other customers. It weakens when speculative requests dominate interconnection queues or costs shift toward residential ratepayers.

The third signal is execution at Taylor Morrison. Berkshire’s next reports should show how it integrates the builder with Clayton’s site-built operations while the housing market remains weak.

A quick rebound is not the standard Abel established. Progress should instead appear through disciplined land spending, stable operations, and credible integration milestones.

The housing thesis strengthens if Berkshire improves the combined platform without relying on favorable mortgage-rate movements. It weakens if consolidation produces disruption before demand recovers.

Together, these signals test Abel’s larger capital-allocation model. He is combining public equities, wholly owned businesses, and infrastructure investment around needs he considers durable.

The strategy differs from buying a single AI winner. Berkshire can participate through electricity even when another hyperscaler captures part of the market.

It also differs from a short-term housing trade. Berkshire controls the acquired business and can hold it through a difficult cycle without depending on a near-term sale.

The structure reduces reliance on one outcome, but it does not eliminate execution risk. Each layer requires different expertise, contracts, and performance measures.

For developers and enterprise buyers, the energy issue can affect product availability and cloud economics. Capacity constraints influence where services launch, how quickly they scale, and what providers charge.

Knowledge workers also have a stake in the outcome. The AI features entering search, productivity software, and enterprise systems depend on infrastructure decisions made years before users see them.

Berkshire’s alphabet bet is therefore useful as a market signal, not as proof that the AI investment cycle will succeed. A famously patient allocator now considers Google’s position credible enough for a core holding.

Abel’s sharper insight concerns the bottleneck underneath that holding. Models attract attention, but electricity, transmission, construction, and contracts determine how much computing capacity reaches customers.

Watch Alphabet’s operating returns, Berkshire Energy’s signed commitments, and Taylor Morrison’s integration. Those three indicators will show whether Abel’s patient capital is meeting durable demand or merely waiting on it.