Two AI bets, one reachable claim: Berkshire's split play on the compute cycle

When Greg Abel, Berkshire Hathaway's new CEO, laid out the ways his conglomerate plans to cash in on AI, he gave two answers — and they are not the same kind of bet. The first is a portfolio position: a roughly $38 billion stake in Alphabet, now Berkshire's third-largest holding, built from scratch in under a year. The second is an operating build: selling electricity to the data centers every AI company is racing to power. One of those claims lands in Berkshire's stock portfolio, marking up book value. The other is what actually reaches its earnings — and it moves on a clock Berkshire does not set.

The distinction matters right now because Berkshire has been sitting out the AI rally. Its shares have lagged the S&P 500 by more than 10 percentage points this year, the cash-heavy, old-economy portfolio disconnected from a market priced on artificial-intelligence capex. The question investors keep asking is how a company that does not make chips, software, or models participates at all. Abel's two answers are the first explicit accounting of it.

The equity leg is a bet on someone else's execution

The headlines chase the equity route, and Abel is candid about why it exists. Berkshire, he has said, has "a lot of visibility from within our companies as to how we're using AI, what type of benefits it's delivering" — and that internal view drove it to conclude Google is a significant player in the space. The mechanics are pure Berkshire: a $10 billion private placement at a 6.5% discount to Alphabet's market price, plus more bought in the second quarter, until the position settled as one of the three largest in the portfolio.

But for all its size, this leg only cashes if Alphabet executes. Berkshire is buying a ticket on a hyperscaler's delivery — the AI customer, the capex, the monetization — and that profit accrues to the equity portfolio, not to Berkshire's operating businesses. It is also a bet made more in hope than in moat. Apple, still the company's largest single holding by far, has been slower than its Magnificent Seven peers to flesh out its AI strategy. The two stock pillars most people assume make Berkshire an AI story are one fast-moving hyperscaler and one architectural laggard.

The earnings leg is the grid, and the grid is the bottleneck

The answer that does not get the headline is energy. And it is the one tied to actual operating results. Berkshire's Iowa utility, MidAmerican Energy, already draws roughly 8% of its power load from data centers. Across its U.S. utilities, Berkshire Hathaway Energy holds about 32,400 net megawatts of generation in operation or under construction — physical capacity positioned for the one input every AI buildout needs and cannot make.

That is the load-bearing observation, and Abel states it plainly: "I've sort of always had the strong view that energy would be the constraint." The interesting part is what he names as the constraint — it is not capital. Berkshire can write any check for turbines or substations. The bottleneck is "how long it would take to get the sites prepared and being in a position they could serve the data centers." Permitting, siting, and interconnection timelines — the physics and politics of hooking gigawatts of new load to a grid — decide when this leg reaches revenue.

And Abel has attached a condition that most utility CEOs would not. Berkshire will only serve new data centers if it does not push up power costs for existing customers — demand that can be met with "no negative impact on other customers' rates; in fact, there must be a benefit." In an industry where ratepayers routinely fight data-center deals, this is the tightest possible spin on the trade: growth that must also be a discount for the people already on the grid.

See the loop in the two ways together. The Alphabet position is a claim on the very hyperscaler capex — Alphabet has guided to $175–185 billion in 2026 capital spending — that later shows up as load growth on Berkshire's own grid. Berkshire is simultaneously buying the customer and supplying the customer's power, so the same demand is being monetized twice: once at market prices in the stock portfolio, once at regulated rates on its utilities.

One fast bet, one slow one

That split is what separates the two ways for an investor. The Alphabet leg is fast and immediate — a discounted position marked to market today — but it reaches book value, rides on another company's delivery, and one of its assumed partners (Apple) is not pulling its weight on AI. The grid leg is slow and durable — it is the only path into Berkshire's operating earnings — but its schedule is set by regulators, local communities, and Abel's own promise that existing ratepayers come out ahead.

The practical read is that Berkshire can afford any amount of generation; it cannot speed up permission. For a company entering the AI story late, that makes the durable exposure the part investors are least patient about — the utility queue that compounds on regulators' clocks, not Abel's. The equity leg buys speed. The grid leg buys the actual constraint. They are not substitutes, and only one of them ends up on the income statement.