Can Nuclear + Batteries + Renewables Actually Support Gigawatt-Scale AI Campuses?

Photo: Jakec, Wikimedia Commons, CC BY-SA 3.0

01 The gigawatt problem

A one-gigawatt data center campus draws, continuously, the power of a mid-sized city — larger than the peak demand of most American municipalities. No single generation resource built in the 2020s serves that cleanly: a gigawatt of firm nuclear is a multi-billion-decade undertaking, a gigawatt of wind and solar is intermittent by physics, and a gigawatt of batteries stores hours, not weeks. The anchor video for this analysis poses the question bluntly: for AI campuses, what actually pencils?

The planning answer that has emerged across announced projects is architectural rather than technological: the stack is the strategy. Nuclear supplies high-capacity baseload volume, renewables add cheap energy mass, batteries bridge second-to-hour ramps and outages, and gas — usually behind-the-meter — covers what the rest cannot, for exactly as long as it has to. The question is not whether each component works, but whether the combination can be financed, permitted and scheduled faster than the AI buildout consumes it.

Analysis — not prediction. N43 and Hermes AI grounds every scenario in the documented record and verified reporting as of September 21, 2026; where evidence is incomplete we say so.

02 Nuclear: baseload, PPAs and the Three Mile Island template

Nuclear’s role in the stack is the firm, carbon-free volume. The template deal is the Three Mile Island restart: Constellation’s Unit 1 brought back online under a long-term power purchase agreement with Microsoft to serve its AI data centers — a single customer effectively funding the recommissioning of a retired reactor in exchange for decades of dedicated baseload. The structure matters more than the site: it converts a slow, capital-heavy asset into a bankable contract between two parties who each need certainty.

The small modular reactor pipeline extends the logic: factory-built units sized in the hundreds of megawatts, sited adjacent to campuses, ordered in multiples. The documented reality in 2026 is that FOAK SMRs remain just that — first-of-a-kind — with construction timelines running well behind early projections and most designs still pre-deployment. Gigawatt campuses being planned today cannot schedule an SMR as their first decade’s power; they schedule it as the back half of the contract curve, with something else covering the front.

What nuclear actually brings to the stack is price-stable volume: fuel costs are a small share of operating cost, a PPA fixes the price for twenty years, and no gas-market shock reprices it mid-decade. For a hyperscaler writing decade-long commitments to customers, that certainty is the product as much as the megawatts.

03 Batteries: the bridge that only bridges

Batteries are the smallest slice by energy and the most indispensable by function. A campus-scale battery does not power a gigawatt load for days — at that scale the storage bill approaches the load itself. What it does is bridge ramps and ride through transients: the seconds-to-minutes surges when training jobs synchronize, the minutes-to-hours gaps when a gas unit trips or a cloud front kills solar output, and the frequency response that keeps a huge, sensitive electronic load from destabilizing the supply behind it.

The documented economics are favorable precisely because the duty cycle is short. A one-to-four-hour system at campus scale is commodity hardware in 2026 — the fastest resource to permit and build, with lead times around a year — and it pairs naturally with on-site gas, covering start ramps and letting turbines run at efficient steady output. The design mistake the record warns against is asking batteries to be generation: multi-day storage chemistries exist but remain expensive and pre-commercial at gigawatt scale.

In stack terms: batteries are the shock absorber, and like shock absorbers they are judged by what they let the rest of the system do — smoother ramps mean smaller gas units, deeper renewables integration, and a grid tie sized closer to average load than to peak.

04 Renewables: volume without firmness

Renewables supply the energy mass of the stack at the lowest marginal cost available. Wind and solar PPAs remain the cheapest new generation in most U.S. markets, and hyperscalers are historically the largest corporate buyers of both. For a gigawatt campus, renewables solve the energy problem — the total number of terawatt-hours per year — even as they leave the capacity problem — having power at 3 a.m. during a regional heat wave — to everyone else.

The honest accounting is that a gigawatt of renewables contracts is not a gigawatt of campus supply. Capacity factors of 25–45 percent mean the same contracted megawatts deliver a fraction of firm equivalent, and the gap must be covered by the nuclear slice, the gas slice or grid imports. The stack works when renewables are sized for energy contribution with firming — overbuilt solar plus storage for daily cycling — rather than for nameplate headlines.

There is also a queue irony: renewables are the fastest generation to physically build and among the slowest to connect, because the interconnection bottleneck applies to them as much as to anything else. A campus can contract renewables energy into the regional market while waiting years for the wires that would deliver it directly.

05 The interconnection queue: the binding constraint

The single most documented obstacle to gigawatt campuses is not generation but the queue. Interconnection requests across major U.S. regions now routinely wait on the order of five years from application to energization, with study timelines, restudies and upgrade cost allocations stretching projects further. A data center developer can sign a land deal, install servers in eighteen months, and then wait until the 2030s for a firm grid connection sized to its load.

Behind-the-meter generation — overwhelmingly gas turbines today — is the market’s answer to the queue: on-site capacity that serves the load without waiting for the wires. It is fast by comparison and firm by physics, which is why nearly every announced gigawatt campus includes a gas component. It is also the stack’s climate contradiction: campuses whose parent companies have the most aggressive clean-energy commitments are, in their first years, gas islands. The standard defense — that contracts for nuclear and renewables balance the books over time and the gas is transitional — is only as good as the queue reform that makes it true.

06 So: does the stack pencil?

Working the arithmetic: nuclear delivers firmness and stability but not speed; batteries deliver speed but not energy; renewables deliver energy but not firmness; gas delivers speed and firmness but not the climate commitments; and the grid delivers everything — eventually. No component is sufficient. The stack pencils precisely because it is a portfolio: each resource covers the others’ defining weakness, and the engineering question becomes sizing and sequencing rather than selection.

The documented 2026 pattern is that campuses announce nuclear PPAs for the out-years, renewables contracts for energy balance, batteries on-site from day one, and gas behind the fence — with the announced mix drifting cleaner as queue positions mature and SMR timelines firm up. The honest conclusion is conditional: yes, the stack can support gigawatt-scale AI campuses, and no single piece of it can — the binding constraints are the queue, the SMR construction record, and how long “transitional” gas is allowed to stay transitional.

What to watch: signed SMR offtakes with concrete delivery dates (the first real one converts the back-half of campus contracts from aspiration to schedule), interconnection queue-reform milestones in the largest regions, and — the quiet tell — how much battery capacity per campus creeps upward, since it measures how firm the rest of the stack actually is.