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.
THE PROBLEM: AI POWER DEMAND
| Metric | 2024 | 2026 (est.) | 2030 (est.) |
|---|---|---|---|
| US Data Center Power Demand | 17 GW | 25 GW | 35-50 GW |
| Single LLM Training Run | 50-100 MW | 100-300 MW | 500+ MW |
| Global Data Center Electricity | 460 TWh | 1,000 TWh | 1,500+ 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
| Factor | Nuclear | Solar + Storage | Wind + Storage | Natural Gas |
|---|---|---|---|---|
| Capacity Factor | 93% | 25-30% | 35-45% | 85% |
| 24/7 Firm Power | Yes | No (needs storage) | No (needs storage) | Yes |
| Carbon Emissions | Zero | Zero | Zero | 400-500g CO2/kWh |
| Land per GW | ~1 sq mi | ~7,000 acres | ~60,000 acres | ~100 acres |
| PPA Duration | 20-40 years | 10-25 years | 10-25 years | 5-15 years |
| LCOE ($/MWh) | $60-120 | $25-40 (+ $50-80 storage) | $30-50 (+ $50-80 storage) | $40-70 |
| Scalable to 1+ GW on-site | Yes (SMR fleet) | Challenging | No | Yes |
EVERY HYPERSCALER NUCLEAR DEAL
| Buyer | Developer | Reactor Type | Capacity | Deal Value | Timeline | Status |
|---|---|---|---|---|---|---|
| Microsoft | Constellation Energy | Large reactor restart | 835 MW | $16B (20-year PPA) | 2027 | Active |
| Kairos Power | KP-FHR (Fluoride salt-cooled) | 500 MW | — | 2030 | Planned | |
| Amazon | X-energy / Energy Northwest | Xe-100 HTGR | 960 MW | — | 2030s | Planned |
| Amazon | Talen Energy | Existing nuclear | — | $20B+ | — | Planned |
| Amazon | Dominion Energy | SMR (type TBD) | — | — | — | Planned |
| Meta | TerraPower | Natrium SFR | 4.0 GW | — | 2032-2035 | Planned |
| Meta | Oklo Inc. | Aurora SFR | 1.2 GW | — | — | Planned |
| Meta | Vistra | Existing nuclear fleet | — | — | — | Planned |
| Meta | Constellation Energy | Existing nuclear fleet | — | — | — | Planned |
| Equinix | Radiant Industries | Kaleidos microreactor | 20 MW | — | 2028+ | Planned |
| Data Center Operators | Last Energy | PWR-20 microreactor | 1.6 GW | GBP 300M (UK portion) | 2030+ | Planned |
| Data Center Operators | Last Energy | PWR-20 microreactor | 600 MW | — | 2030+ | Planned |
| TBD (AI Data Center) | Aalo Atomics | PWR (Aalo Pod 50 MWe) | 50 MW | TBD | 2027 | Planned |
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.