Tech giants are turning to small modular reactors (SMRs) because artificial intelligence data centers demand constant, carbon-free electricity that solar and wind alone cannot provide. Microsoft, Amazon, Google, and Meta have collectively committed to over 9.8 gigawatts of nuclear capacity in the last eighteen months, nearly all of it tied to AI infrastructure. A single hyperscale AI campus under construction in 2026 can draw 1 to 2 gigawatts continuously, enough to power a mid-sized American city. Grid operators in Virginia, Georgia, Texas, and Arizona have already warned that combined AI buildout will outstrip planned generation additions before the decade ends.

What Exactly Is a Small Modular Reactor?

An SMR is a nuclear reactor built well below the roughly 1,000-megawatt scale of a conventional plant, designed around standardized components that can be manufactured in a factory and shipped to a site rather than custom-built over a decade. The "modular" part is the real innovation: instead of one enormous bespoke construction project, an SMR plant can be built by combining multiple identical, factory-fabricated units, the same approach that made submarine reactors dramatically more predictable to build than one-off megaprojects.

Most SMR designs cluster between roughly 50 and 470 megawatts electric, compared to 1,000 to 1,600 megawatts for a conventional large reactor. A sub-category called microreactors goes even smaller, under 20 megawatts. To put scale in perspective, roughly 300 SMRs could generate enough electricity to power 68 million homes, about half of all US households, with carbon emissions avoided equivalent to taking 100 million cars off the road.

Why Can't AI Data Centers Just Use Solar and Wind?

AI training and inference require firm, 24/7 power, not intermittent generation. Nuclear offers carbon-free baseload power with capacity factors above 90 percent, the reliability figure solar and wind cannot match. US data center energy demand is projected to grow from 17 gigawatts in 2022 to 35 gigawatts by 2030, with global data center electricity demand projected to reach 1,300 terawatt-hours by 2035.

Hyperscalers are under shareholder pressure to hit net-zero commitments and cannot fill a multi-gigawatt gap with natural-gas peaker plants without breaking those pledges. SMRs add a second advantage over both renewables and traditional nuclear: they can be sited closer to transmission lines and, in some proposed configurations, co-located directly with the data center itself, bypassing some of the grid interconnection queue entirely.

Which Companies Are Actually Building These Reactors?

SMR is an umbrella term covering genuinely different technologies with different coolants, fuels, and use cases. Several designs have real orders, real construction, or real hyperscaler customers behind them as of 2026:

• GE Vernova Hitachi BWRX-300: A 300-megawatt boiling water reactor, the furthest along of any Western SMR design. Under construction at Ontario Power Generation's Darlington site in Canada, targeting commercial operation around 2030. Also selected by SaskPower for Saskatchewan's first nuclear plant.
• NuScale Power VOYGR: A 77-megawatt light water reactor, the first SMR design to receive full NRC design certification. A 12-module VOYGR plant totaling 684 megawatts is estimated at roughly 3 billion dollars, or about 4,385 dollars per kilowatt.
• X-energy Xe-100: An 80-megawatt gas-cooled reactor with TRISO fuel. Amazon led a 500 million dollar financing round for X-energy and separately invested 700 million dollars for rights to up to 12 Xe-100 units.
• Kairos Power KP-FHR (Hermes): A 75-megawatt fluoride salt-cooled reactor. Google signed the first US corporate SMR fleet deal with Kairos Power, targeting 500 megawatts total capacity. The Hermes demonstration reactor is under construction at Oak Ridge.
• TerraPower Natrium: A 345-megawatt sodium-cooled fast reactor with 500 megawatts of storage. Meta's largest single nuclear commitment of 2.8 gigawatts is tied to eight planned Natrium plants.
• Oklo Aurora: A 15 to 75 megawatt sodium fast reactor in the microreactor class. Publicly traded on the New York Stock Exchange with a market cap of roughly 12.9 billion dollars. Broke ground at Idaho National Laboratory in September 2025 as part of Meta's 1.2 gigawatt Aurora campus deal.
• Rolls-Royce SMR: A 470-megawatt pressurized water reactor, the leading UK-developed design progressing through UK regulatory assessment, with first units targeted for the early 2030s.
• CNNC Linglong One (ACP100): A 125-megawatt pressurized water reactor under construction in Hainan, China, on track to become the world's first land-based commercial SMR, with commercial operation expected in the first half of 2026, ahead of every Western competitor.

How Are Tech Giants Dividing Up the Nuclear Pie?

Thirteen named deals represent roughly 9.8 to 10 gigawatts of committed capacity, and every major hyperscaler is pursuing a genuinely different strategy. Microsoft agreed to buy the entire output of the restarted Three Mile Island reactor, securing 835 megawatts with the earliest delivery timeline among major hyperscalers. Amazon's 700 million dollar investment in X-energy and 500 million dollar financing round represent bets on next-generation SMR technology alongside existing nuclear power-purchase agreements. Google's fleet deal with Kairos Power targets 500 megawatts total capacity. Meta's up-to-6.6 gigawatt nuclear agreement announced in January 2026 includes eight planned Natrium plants and a 1.2 gigawatt Aurora campus deal with Oklo.

How to Evaluate SMR Economics and Timeline Feasibility

Understanding whether SMRs can actually deliver on their promises requires examining three key dimensions of the technology and its deployment:

• Manufacturing Learning Curve: Most SMR developers believe committing to production of at least 6 to 10 units of the same design would unlock cost reductions of up to 40 percent compared to a first-of-a-kind build, the same learning-curve effect already demonstrated in small reactor manufacturing for submarines.
• Construction Timeline Reality: The core value proposition of SMRs revolves around the fact that by making nuclear power projects smaller, they become far less likely to encounter serious cost or construction time overruns as is common in the traditional nuclear industry, according to IDTechEx's 2026 research.
• Grid Integration Advantage: SMRs can be sited closer to transmission lines and, in some proposed configurations, co-located directly with data centers, bypassing some of the grid interconnection queue that typically delays large-load projects by up to a decade.

The race between hyperscalers to secure SMR capacity reflects a fundamental mismatch between AI's power demands and the grid's ability to supply them. With data center energy demand projected to double by 2030 and global demand reaching 1,300 terawatt-hours by 2035, SMRs represent the only technology that can deliver the firm, 24/7, carbon-free electricity that AI infrastructure requires at scale. Whether these reactors can be built fast enough and cheaply enough to meet that demand remains the defining question of the next five years.