The Ultimate Guide to Nuclear Power and Small Modular Reactors

Small modular reactors, or SMRs, have been promised for over a decade as the technology that finally makes nuclear power fast and cheap to build. In 2026, that promise is being tested against poured concrete for the first time. Several companies have moved from paper designs to actual construction, driven largely by one customer: hyperscale data centers that need firm, round-the-clock power and are willing to sign long-term contracts to get it.

What actually makes a reactor "small" and "modular"

A traditional nuclear plant is built almost entirely on-site, with a single large reactor generating 1,000 megawatts or more. An SMR is designed differently: smaller individual units, typically in the 50 to 300 megawatt range, built with standardized components that can be manufactured in a factory and shipped to the site for assembly, the way a modular home is built. The pitch is that factory manufacturing brings the cost curve down the way it did for solar panels and batteries, and that smaller units can be added incrementally as demand grows instead of requiring one enormous upfront investment.

Who's actually pouring concrete

Four projects have moved past the planning stage in the US as of 2026. TerraPower's Natrium plant in Kemmerer, Wyoming — a sodium-cooled Gen IV design, not technically an SMR but part of the same advanced-reactor wave — received its construction permit and broke ground in early 2026, backed by the Department of Energy's Advanced Reactor Demonstration Program.

Kairos Power has two projects running in parallel in Oak Ridge, Tennessee: the 35-megawatt-thermal Hermes demonstration reactor, where construction began in 2025 and the company is now preparing an operating license application, and Hermes 2, a 50-megawatt-electric commercial demonstration unit that broke ground in April 2026 and is expected online around 2030. X-energy's Xe-100 reactor, partnered with Dow Chemical in Seadrift, Texas, is still in NRC construction-permit review rather than under construction. A combined TVA and Holtec effort, backed by up to $800 million in federal cost-sharing, is in the same regulatory-review stage for sites in Tennessee and Michigan.

Kairos, Google, and TVA's power purchase agreement made TVA the first US utility to sign a contract with an advanced nuclear plant — a template other utility-hyperscaler pairings are likely to follow.

Why the timelines matter more than the announcements

Nearly every SMR company has, at some point, announced a reactor "online by [year]" that later slipped. NuScale received NRC approval for an uprated SMR design in 2026 — a real regulatory milestone — but has no construction timeline attached to it yet. The gap between "design approved" and "reactor producing power" has historically run five to ten years in the advanced-nuclear industry, and Hermes 2's own 2030 target reflects that reality even for a company that's already broken ground.

Why hyperscalers are the ones actually funding this

The customer driving nearly all of this construction isn't a traditional utility looking to replace an aging coal plant — it's data center operators. AI training and inference need power that's available continuously, doesn't depend on weather, and can be contracted years in advance at a predictable price, which nuclear offers in a way solar and wind, on their own, don't. That's the same power-hungry data center demand driving Microsoft, Amazon, Google, and Meta's existing nuclear deals and the record pace of battery storage buildout happening alongside it.

What could still go wrong

Cost overruns and schedule slips are the industry's historical norm, not the exception — Vogtle, the last large US reactor project, ran roughly $17 billion over budget and years behind schedule, and SMR advocates have yet to prove the modular, factory-built approach avoids the same trap at scale. Supply chains for reactor-grade components remain thin, and the workforce needed to build and operate a fleet of new reactors has to be rebuilt largely from scratch after decades without major US nuclear construction. None of the four active projects have produced commercial electricity yet; 2026 is the year they moved from announcements to construction sites, not the year SMRs proved themselves.