The AI winners are moving beyond chips and into the grid
The next leg of the AI trade may not belong to the most obvious chip names. As hyperscaler spending gets more contested, the real scarcity is shifting toward power, cooling, permitting, and data-center infrastructure — and that favors a different set of winners.
The market is still talking about AI as a compute story, but the trade is starting to behave like an electricity story. That matters because once investors begin to doubt how long hyperscalers can keep expanding capex at the same pace, the obvious winners become less obvious. Chips and networking still matter, but the harder bottleneck now is physical: enough power, enough cooling, enough interconnection, and enough equipment to bring new capacity online. This week’s setup made that shift hard to ignore, with AI infrastructure holding up better than broad semis even as policy and grid headlines underscored that the constraint is no longer just silicon.
The clean contrarian take is that AI leadership is broadening precisely because the first phase worked too well. Compute demand pulled forward the chip winners; now that same demand is colliding with the grid. Public reporting this month showed generator step-up transformer lead times stretching past 160 weeks, up from 143 weeks in 2024. That is not a normal supply-chain hiccup. It is a sign that the scarce asset in AI is moving from the server rack to the electrical system that feeds it.
That shift helps explain why the most interesting comparisons now sit outside classic semis. Vertiv is the clearest expression of the theme: revenue growth of 27.7%, EPS growth of 164.4%, and a TickerSpark Score that would reflect exactly what investors are paying for here — direct exposure to cooling and power density inside the data center. Yes, VRT is expensive at 63.61x earnings and already up 71.2% this year. But expensive and crowded are not the same thing when the company is tied to the actual choke point. If AI racks require more thermal management and power conditioning per deployment, Vertiv is selling into the bottleneck, not around it.
The broader electrification layer matters too, and that is where Eaton and Quanta Services deserve more attention than they usually get in AI conversations. ETN is not a pure-play, which is exactly why the market can still underestimate it; a 33.57x P/E on 10.3% revenue growth looks less flashy than Vertiv, but the company is directly exposed to switchgear, power distribution, and the unglamorous hardware that determines whether a data center can energize on time. PWR, meanwhile, is the construction and transmission side of the same thesis. Its 19.8% revenue growth and 45.9% YTD gain say investors are already starting to price in that connecting generation, substations, and large-load facilities is becoming part of the AI stack.
The utility angle is even more contrarian because it forces investors to think about AI as baseload demand, not just digital demand. Constellation Energy looks messy on the surface — EPS growth is negative and the stock is down 25.9% YTD — which is exactly why it belongs in the debate. A market that spent two years rewarding anything tied to compute is now being asked to price long-duration power contracts, nuclear restarts, and grid reliability. The recent 15-year nuclear power purchase agreement and the push to accelerate the Three Mile Island restart are not side stories. They are evidence that large customers increasingly want firm power, not just more servers.
The REITs make the same point from another angle: location and power access are becoming assets in their own right. Digital Realty and Equinix are not just landlords if power scarcity is real. DLR trades at 40.43x earnings with 10.0% revenue growth and 114.4% EPS growth, while EQIX sits at 38.80x earnings with a 15.0% net margin. Those are not cheap multiples for real estate, but this is no longer a plain real-estate story. If permitting tightens and local grids push back, then campuses with existing interconnection, development rights, and customer density become harder to replicate than many investors assumed.
None of this means the chip-and-networking winners are finished. Arista is still a real AI beneficiary, and its 28.6% revenue growth with a 38.3% net margin is better than most companies in the market, let alone infrastructure names. But that is also the point: ANET already reflects a lot of excellence at 22.60x sales. Investors are no longer just asking who supplies the fastest gear inside the cluster; they are asking whether the cluster can get enough power, enough cooling, and enough permits to be built at scale. That is a different question, and it naturally pushes capital toward a different cohort.
Bulls on the old leadership will argue that if AI demand remains strong, chips and networking should still capture the highest-value dollars. Fair. But that comparison ignores where delays are now showing up. A GPU shortage can ease with more supply and better product cycles; a transformer shortage measured in years, or a state-level moratorium on large new data centers, is a harder constraint. New York’s one-year moratorium on large new data centers was the clearest signal yet that AI expansion is becoming a political and physical infrastructure issue, not just a capex line item.
The market does not need to abandon the chip leaders for this thesis to work. It only needs to accept that the marginal bottleneck in AI is moving outward from silicon to the systems around it. That is why we think the next leg of the trade is more likely to reward companies tied to power quality, cooling, transmission, interconnection, and data-center land banks than investors still anchored to the first wave of AI winners.
What would change our mind? A clear reacceleration in hyperscaler capex without corresponding stress in grid equipment, permitting, or power procurement would argue the bottleneck is still mostly compute. Short of that, we would watch lead times, utility contracting, and data-center policy more closely than chip headlines. The AI buildout is no longer just about who makes the brains. It is about who can keep the lights on.
Our take, not advice. This is opinion commentary — informational only, not personalized investment recommendations. Markets carry risk. Do your own research and consider your own situation before any trade.
▌The Daily Briefing · Free
A new stock idea, every evening.
One stock worth watching each weekday, plus the analysis behind it. Free, in your inbox.
