This week’s softness in XLK is feeding the usual bubble chatter, but the more important shift is happening underneath the surface: AI spending is spreading beyond chips into networking, power, cooling, and data-center real estate. Nvidia’s latest results reinforced that capex is still durable; the debate now is where the next dollars in the stack get earned.
The cleanest read on this week’s AI wobble is not that the trade is cracking. It is that the market is finally pricing AI as a full infrastructure buildout rather than a one-stock phenomenon. Nvidia just posted $81.6 billion in quarterly revenue, including $75.2 billion from data center, and management still described demand as running at an unprecedented pace. If the capex engine is still that strong, the more useful question is not whether AI spending is over, but which parts of the stack become the next bottlenecks — and therefore the next beneficiaries.
That is why XLK lagging SPY this week matters less than the composition of leadership. When a theme matures, returns stop looking as clean as a single-ticker momentum trade. They rotate into the companies that make the core product deployable at scale. In AI, that means the spend is moving from pure compute toward the physical and network layers that let compute work in the real world: switches, power distribution, thermal management, and the buildings that can actually host dense clusters.
The macro constraint is not subtle anymore. U.S. data-center power demand is projected to rise from 31 GW in 2025 to 66 GW in 2027, and another widely cited estimate sees demand reaching 74 GW by 2028 with a 49 GW shortfall in available power access. That is the broadening thesis in one frame. If chips were the only scarce input, the bubble argument would have more force. But when the limiting factors become electricity, cooling, and siting, the value pool necessarily spreads beyond NVDA into names like VRT, ETN, DLR, and even power-exposed plays like CEG.
The market is already telling that story, even if the debate still fixates on Nvidia. VRT is up 81.1% year to date, far ahead of NVDA at 11.2%, while DLR has gained 24.5% and ETN 22.6%. Those are not random sympathy moves. They line up with where the bottlenecks are shifting. Vertiv sits in cooling and power infrastructure, Eaton in electrical equipment, and Digital Realty in the real estate layer of AI capacity. If this were simply a one-name mania rolling over, we would expect the second-derivative beneficiaries to fade with it. Instead, the tape says investors are paying for the stack.
The operating backdrop supports that rotation. Arista reported Q1 revenue of $2.71 billion, up 35.1% year over year, and raised its 2026 AI networking target to $3.5 billion. That matters because networking is no longer a sidecar to the AI story; it is becoming one of the core spend buckets. Training and inference clusters are only as useful as the fabric connecting them, and that is why ANET deserves to be discussed as infrastructure, not as an afterthought to semis. The same logic applies to cooling and electrical systems. Recent company commentary across the industrial side of the stack has been explicit that data-center demand is now driving order momentum, not merely flattering it.
Yes, bulls of the old AI trade will argue that this is still just NVDA by another name. There is truth in that: Nvidia remains the center of gravity, and its $75.2 billion in data-center revenue is still the clearest proof that AI capex has not broken. But that objection misses the economic handoff. Once demand is proven, the next incremental dollar of urgency often goes to the chokepoints around the leader, not only to the leader itself. That is exactly what happened in prior infrastructure buildouts, and it is what the current power and cooling squeeze is forcing now.
Valuation is the obvious pushback, and it is fair. This is not a hidden trade anymore.
Those multiples tell us investors are already discounting a multi-year AI infrastructure cycle across several layers. So the call here is not that every beneficiary is cheap. It is that the bubble framing is too narrow. A bubble argument implies the market is extrapolating one glamorous winner without regard to the rest of the system. What we are seeing instead is a more rational, if still expensive, repricing of the system itself. Even CEG, down 20.7% year to date, fits the conversation because power availability is becoming strategic to AI deployment whether or not utility equities have kept pace.
There is also a useful historical parallel. The late-1990s internet buildout did not stop at the obvious compute winners; it spread into routers, fiber, telecom gear, and the real estate that housed the network. AI looks similar, except more constrained by physical power and thermal limits than the internet ever was. That makes this cycle arguably more infrastructure-heavy, not less. The market is not abandoning AI leadership. It is widening the definition of what leadership means.
Our view is straightforward: the AI trade is rotating through the stack, not falling apart. Nvidia’s latest quarter kept the demand side intact, and the emerging scarcity is now downstream in networking, power delivery, cooling, and data-center capacity. That is why weakness in a broad tech ETF is a poor proxy for the health of the theme.
What we would watch from here is simple. If data-center power forecasts start easing, if networking demand stalls, or if cooling and electrical order momentum fades, then the broadening case weakens. Until then, the better read is that AI is becoming less of a single-stock story and more of an infrastructure story — which is exactly what a durable capex cycle should look like.
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.
