DEEPTECH.INTEL
CROSS-SECTOR ANALYSIS // ENERGY × AI

Nuclear Power for AI: Why Every Tech Giant Is Going Nuclear

As of March 2026, every major tech hyperscaler — Microsoft, Google, Amazon, and Meta — has committed to nuclear power for AI data centers. Across 13 announced deals, over 9.8 GW of nuclear capacity is now committed to powering the next generation of artificial intelligence infrastructure. This represents more corporate investment in nuclear energy than any prior decade in US history. The thesis is simple: AI training and inference workloads need 24/7, carbon-free, firm baseload power at a scale that only nuclear can deliver. Solar and wind cannot do it alone — not without grid-scale battery storage that doesn't yet exist at the required scale.

9.8 GW Nuclear Committed
13 Deals Tracked
7 Buyers
2027 First Power

THE PROBLEM: AI POWER DEMAND

Metric20242026 (est.)2030 (est.)
US Data Center Power Demand17 GW25 GW35-50 GW
Single LLM Training Run50-100 MW100-300 MW500+ MW
Global Data Center Electricity460 TWh1,000 TWh1,500+ TWh
% of US Electricity to Data Centers~4%~6%~8-12%

AI power demand is growing faster than any energy source can be deployed. A single frontier model training run on 100,000 GPUs consumes as much power as a small city. Inference workloads — serving models to billions of users — are growing even faster. The IEA projects that data centers will consume more electricity than the entire country of Japan by 2030.

WHY NUCLEAR WINS FOR AI

FactorNuclearSolar + StorageWind + StorageNatural Gas
Capacity Factor93%25-30%35-45%85%
24/7 Firm PowerYesNo (needs storage)No (needs storage)Yes
Carbon EmissionsZeroZeroZero400-500g CO2/kWh
Land per GW~1 sq mi~7,000 acres~60,000 acres~100 acres
PPA Duration20-40 years10-25 years10-25 years5-15 years
LCOE ($/MWh)$60-120$25-40 (+ $50-80 storage)$30-50 (+ $50-80 storage)$40-70
Scalable to 1+ GW on-siteYes (SMR fleet)ChallengingNoYes

EVERY HYPERSCALER NUCLEAR DEAL

Meta5.2 GW
4 deals
TerraPower
Natrium SFR · 4.0 GW · 2032-2035
Oklo Inc.
Aurora SFR · 1.2 GW
Vistra
Existing nuclear fleet
Constellation Energy
Existing nuclear fleet
Data Center Operators2.2 GW
2 deals
Last Energy
PWR-20 microreactor · 1.6 GW · 2030+
Last Energy
PWR-20 microreactor · 600 MW · 2030+
Amazon960 MW
3 deals
X-energy / Energy Northwest
Xe-100 HTGR · 960 MW · 2030s
Talen Energy
Existing nuclear
Dominion Energy
SMR (type TBD)
Microsoft835 MW
1 deal
Constellation Energy
Large reactor restart · 835 MW · 2027
Google500 MW
1 deal
Kairos Power
KP-FHR (Fluoride salt-cooled) · 500 MW · 2030
TBD (AI Data Center)50 MW
1 deal
Aalo Atomics
PWR (Aalo Pod 50 MWe) · 50 MW · 2027
Equinix20 MW
1 deal
Radiant Industries
Kaleidos microreactor · 20 MW · 2028+
BuyerDeveloperReactor TypeCapacityDeal ValueTimelineStatus
MicrosoftConstellation EnergyLarge reactor restart835 MW$16B (20-year PPA)2027Active
GoogleKairos PowerKP-FHR (Fluoride salt-cooled)500 MW2030Planned
AmazonX-energy / Energy NorthwestXe-100 HTGR960 MW2030sPlanned
AmazonTalen EnergyExisting nuclear$20B+Planned
AmazonDominion EnergySMR (type TBD)Planned
MetaTerraPowerNatrium SFR4.0 GW2032-2035Planned
MetaOklo Inc.Aurora SFR1.2 GWPlanned
MetaVistraExisting nuclear fleetPlanned
MetaConstellation EnergyExisting nuclear fleetPlanned
EquinixRadiant IndustriesKaleidos microreactor20 MW2028+Planned
Data Center OperatorsLast EnergyPWR-20 microreactor1.6 GWGBP 300M (UK portion)2030+Planned
Data Center OperatorsLast EnergyPWR-20 microreactor600 MW2030+Planned
TBD (AI Data Center)Aalo AtomicsPWR (Aalo Pod 50 MWe)50 MWTBD2027Planned

For detailed reactor specifications and deployment timelines, see smrintel.com/nuclear-data-center-deals →

BOTTOM LINE

The nuclear-for-AI thesis has moved from speculation to execution. With 9.8 GW committed across 13 deals, hyperscaler nuclear procurement is now the largest driver of new nuclear capacity in the United States. Microsoft will be first to receive power (Crane Clean Energy Center, 835 MW, 2027) by choosing an existing reactor restart. Meta has placed the largest bet (up to 6.6 GW) by diversifying across both new-build SMRs (TerraPower, Oklo) and existing fleet PPAs (Vistra, Constellation). Google and Amazon are backing next-generation reactor designs (Kairos KP-FHR and X-energy Xe-100) that could define the data center nuclear standard for the 2030s. The winners will be companies that combine both strategies: existing nuclear for near-term power, and new-build SMRs for long-term scalable supply. Nuclear will not be the only answer to AI's energy challenge — but it is the only firm, zero-carbon, scalable power source available at the GW scale that data centers require.

FREQUENTLY ASKED QUESTIONS

RELATED INTELLIGENCE

The AI Hardware Revolution: From Chips to InfrastructureDeep Tech Stocks: Complete Investor GuideEvery Deep Tech Company: The Complete Directory