The AI infrastructure race is becoming a power-and-grid race, as AMD’s 2.5 GW agreement and a proposed $100 billion data-center campus show. That favors electrical equipment, engineering, storage, and reliable generation over the latest chip headline alone.
The AI infrastructure debate has crossed an important line: the question is no longer simply whether hyperscalers will keep buying accelerators, but whether the grid can connect and serve the capacity they want to deploy. AMD’s agreement with Core Scientific for up to 2.5 GW of data-center capacity is a power-access deal disguised as an AI infrastructure deal. The proposed NextEra–Brookfield campus makes the same point even more explicitly, pairing a $100 billion data-center project with 2 GW of gas-fired generation and 2.6 GW of battery storage. The scarce asset is increasingly the energized site, not the server rack sitting inside it.
That distinction changes where the strongest operating leverage should appear. A chip can be manufactured and shipped, but it cannot generate revenue if a data center is waiting on interconnection, transformers, switchgear, transmission, or firm power. The AMD–Core Scientific agreement is therefore more revealing than another announcement about accelerator demand: the parties are securing the physical capacity required to run compute. The NextEra–Brookfield plan goes further by embedding generation and storage into the campus itself. In both cases, the commercial bottleneck is the ability to deliver electricity on schedule.
The grid equipment complex is already showing that demand. Eaton reported that Electrical Americas order acceleration reached 42% on a twelve-month rolling average, driven by data-center momentum, and raised its 2026 organic growth guidance midpoint to 10% from 8%. Quanta Services reported a $48.5 billion backlog at the end of the first quarter, up 10.2% from year-end 2025, while framing its opportunity across utilities, generation, and large-load markets. Those figures do not prove that every AI project will be built, but they do show where spending is landing before the full computing payoff arrives: in the equipment and construction needed to make large loads possible.
Market performance reinforces the split between enabling infrastructure and power names whose earnings remain more exposed to the timing of the cycle:
Eaton and Quanta have benefited from the immediacy of electrification and construction demand, while Constellation Energy and Vistra show that simply owning generation is not the same as monetizing the AI load today. That is not a contradiction to the thesis; it is the nuance. The winning layer may be the infrastructure that gets power connected and contracted, followed by reliable generation with a visible path to new demand. A utility without the right project timing, or with earnings moving in the wrong direction, can still lag even as the long-term power story strengthens.
Yes, chip bulls can argue that more efficient accelerators will reduce electricity consumed per unit of training or inference. They can also point out that operators are using demand response, on-site generation, and co-investment to work around regional grid constraints. But efficiency does not eliminate the need for power when total deployment is expanding, and regional workarounds still require capital equipment, storage, interconnection, and reliable supply. The constraint may prove temporary in a particular market, but solving it is itself an infrastructure opportunity. The more sophisticated the workaround, the more the AI buildout starts to resemble an energy and electrical-construction cycle.
Data-center landlords are part of this reframe, but they should not be confused with the pure beneficiaries of power scarcity. Digital Realty raised its 2026 adjusted FFO guidance and disclosed a $475 million Kansas City land purchase supporting hyperscale development for up to 2 GW of utility power. Equinix also raised its outlook on record bookings. That confirms demand is broadening into colocation, interconnection, and power-enabled real estate. Yet a headline about a campus is only valuable if the site has power, equipment, and a workable path to service. The market should therefore reward delivered capacity rather than announced square footage alone.
The closest historical parallel is the late-1990s telecom buildout. The debate eventually moved from who owned the routers to who controlled fiber, rights-of-way, and backhaul. AI is following a similar path: compute remains essential, but the scarce and difficult-to-replicate layer is the network around it. Today that means transformers, switchgear, transmission, interconnection queues, storage, and firm generation. The AMD and NextEra-linked announcements make that shift visible in real time, because both are organized around securing the conditions under which chips can actually operate.
Our view is that the next investable phase of AI infrastructure is less about picking the newest accelerator and more about identifying who can connect, power, and build the load. Eaton and Quanta offer the clearest operating read-through because equipment orders and construction backlogs arrive before a data center reaches full utilization. Generation and data-center real estate remain important, but their returns depend more heavily on project timing and the conversion of power access into cash flow.
We would change that view if chip-efficiency gains and demand-response systems consistently allowed hyperscalers to expand compute without major new generation, grid equipment, or interconnection spending. Until then, the AMD–Core Scientific capacity commitment and the NextEra–Brookfield campus point to the same conclusion: electricity is becoming the gating item in AI deployment, and the grid is moving from background utility to core infrastructure.
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.
