Estimated reading time: 5 minutes · Last updated:
Greg Abel, Berkshire Hathaway's chief executive, told CNBC that the company sees data-center power as the next structural bottleneck for AI growth and is positioning its utilities accordingly. Berkshire holds roughly $38 billion of Alphabet stock, a bet it began assembling last year because AI is already affecting its operating businesses. At the same time, Abel pointed to the firm's regulated utilities in Iowa, Nevada and across the western US as potential suppliers of power to large cloud providers — noting one Iowa utility already gets roughly 8% of its load from data centers. Berkshire, he said, will serve new AI capacity only if doing so does not push up costs for existing customers.
Key takeaways
- Berkshire's AI exposure: Berkshire holds roughly $38 billion of Alphabet stock, a position the company began building last year as a play on AI.
- Utilities as an asset: Abel says Berkshire's utilities in Iowa, Nevada and much of the western US could supply power to hyperscalers building more AI capacity.
- Existing-customer condition: Berkshire will only connect new data-center load where doing so will not raise power costs for existing customers, Abel said.
- Current data-center share: One Iowa utility already gets roughly 8% of its power load from data centers, according to Abel's comments.
Table of contents
- Key takeaways
- Why Berkshire is looking past chips and into power
- How Berkshire's utilities fit the AI buildout
- The catch: serving AI without raising bills for neighbors
- Implications for investors, communities and hyperscalers
- How the bet could pay off — and where it can stall
- What to be careful about
- Frequently asked questions
Why Berkshire is looking past chips and into power
Greg Abel framed Berkshire's AI stance around durable operational advantage rather than short-term market momentum. The company built a sizable position in Alphabet — roughly $38 billion — because its businesses are seeing AI-driven demand, Abel told CNBC. That stake gives Berkshire exposure to the software and cloud side of the boom, but he said a different bottleneck is emerging: where to get the sustained, low-cost electricity that large AI workloads consume.
Abel's point is pragmatic. Chip shortages and supply-chain constraints dominate headlines today, but AI clusters require continuous high-power feeds and local approvals. Utility-scale generation, grid capacity and the consenting process — not just silicon — can limit how quickly hyperscalers expand. For Berkshire, which owns regulated utilities, that constrains and creates an investment opportunity at the same time.
How Berkshire's utilities fit the AI buildout
Berkshire already owns utilities across Iowa, Nevada and much of the western US that serve large industrial and commercial customers. Abel highlighted that one Iowa utility now gets roughly 8% of its power load from data centers, using that as evidence of the kind of demand hyperscalers bring to a grid. Those customers want predictable, high-capacity connections and often sign long-term contracts.
That demand maps to the kinds of assets Berkshire can operate: regulated distribution networks, transmission access and generation that can be directed or expanded under state oversight. If a utility can add capacity without destabilising service or rates, it can win contracts that produce steady, long-duration revenue — the opposite of a speculative, short-cycle tech bet.
The catch: serving AI without raising bills for neighbors
Abel made clear there is a trade-off. Berkshire will pursue deals with hyperscalers only when adding data-center load does not raise costs for other customers. He framed this as a corporate and regulatory constraint: utilities must balance new large customers against rate impacts and community acceptance. That condition limits how fast utilities can monetise AI demand, even if generation can be built.
Community pushback and regulator scrutiny are material restraints. Developers often propose new generation or transmission that gets reviewed by state regulators and local stakeholders. If the approval process requires rate recovery from the general customer base, utilities face political and financial friction that slows or alters projects — and that, Abel warned, can cap the speed at which utilities turn AI demand into profit.
Implications for investors, communities and hyperscalers
For investors, Berkshire's approach mixes public-equity exposure with operational optionality: its roughly $38 billion Alphabet stake ties it to AI-driven growth, while its utilities give it a shot at supplying the physical inputs. Abel also pointed to Berkshire's willingness to buy in other sectors, citing its Taylor Morrison acquisition as a long-term play he expects to be a "very strong asset" in five to 10 years.
For communities and regulators, the message is leverage: hyperscalers bring jobs and tax revenue but also heavy electricity demands. Utilities that choose to serve those customers must show how costs are allocated and how reliability is preserved. Hyperscalers, meanwhile, may need to offer investments in local grids or long-term contracts to reduce opposition and secure permits.
| Item | Role | Relevant figure or note |
|---|---|---|
| Alphabet stake | Public-equity exposure to AI | Roughly $38 billion |
| Iowa utility | Potential power supplier to data centers | About 8% of its load from data centers |
| Taylor Morrison | Non-AI operating acquisition | Expected to be "very strong asset" in five to 10 years |
How the bet could pay off — and where it can stall
The case for
- Utilities can sign long-term contracts with hyperscalers that provide predictable, multi-year revenue once transmission and generation are in place.
- Owning regulated distribution and generation gives Berkshire a direct path to monetise local demand without relying solely on public-equity gains.
The case against
- Regulators or local communities could require rate recovery mechanisms that raise costs for others, limiting a utility's ability to accept large new data-center loads.
- If hyperscalers decide to site elsewhere because of permitting delays or lower-cost grids, the anticipated local demand may not materialise despite available generation.
What to be careful about
- Regulatory refusals or rate decisions that prevent cost recovery for new generation or transmission projects tied to data centers.
- Local opposition to new power infrastructure or large data-center campuses that delays or blocks connections.
- Hyperscalers shifting siting plans to regions with cheaper or faster permitting, leaving built capacity underutilised.
The bottom line
Berkshire's approach to AI is deliberate: a large Alphabet stake gives it market exposure while its utilities offer a route to capture the physical inputs that AI consumes. Greg Abel framed the problem simply — access to continuous, low-cost power and the approvals to deliver it are becoming as important as compute and chips. But Berkshire's condition that new hookups must not increase costs for current customers means the firm will proceed cautiously. That constraint will determine whether utilities become a steady revenue source from AI or remain a strategic option that faces regulatory and community friction.
What to watch
- Watch for filings or public notices from Berkshire's utilities in Iowa or Nevada about new large customer hookups; no date has been set.
- Watch for state regulator dockets or votes in jurisdictions where Berkshire operates that would affect rate recovery for new transmission or generation; no date has been set.
- Watch for future Berkshire disclosures about utility capital plans tied to large commercial customers; no date has been set.
Frequently asked questions
How is Berkshire positioned to benefit from AI demand?
Berkshire combines a public-equity bet — roughly $38 billion of Alphabet stock — with direct operational exposure through utilities it owns in Iowa, Nevada and much of the western US that can supply power to large data centers.
How much of Berkshire's grid load is already data-center driven?
Abel said one Iowa utility already gets roughly 8% of its power load from data centers, which shows the scale a single hyperscaler class of customer can represent on a local grid.
Will Berkshire sell power to every new data center that asks?
No; Abel said Berkshire will only serve new data-center capacity when doing so does not raise power costs for existing customers, making regulatory outcomes and cost allocation central to the decision.
Related reading
This article is information, not financial advice. Anyone acting on it should do their own checks.
