Main points
- AI Power Crisis: AI data centers consume 10-20x more electricity than traditional cloud. A 1GW AI campus = power for 750,000 homes. Grid can't support without new power generation.
- Tech Nuclear Pivot: Microsoft ($1B Three Mile Island restart), Google (7 SMRs from Kairos), Amazon ($500M X-energy investment). Big Tech committed $50B+ to nuclear by 2030.
- SMR Breakthrough: Small modular reactors (50-300MW) factory-built, faster deployment (3-5 years vs 10-15 years for large reactors), inherently safer design. NuScale first NRC-certified SMR (2023).
- Market Explosion: Nuclear capacity: 390GW (2024) → 600GW (2040). New reactor construction: $300B through 2040. Uranium demand doubling (price $80/lb → $150/lb projected).
- Climate Imperative: Net-zero carbon goals require nuclear. Solar/wind can't provide 24/7 baseload. Natural gas emits carbon. Only nuclear delivers zero-carbon reliable power at data center scale.
- Winners: Constellation Energy (CEG - Microsoft partner), Oklo (Sam Altman SMR), Cameco (uranium), NuScale Power (SMR tech), Vistra Energy (nuclear operator).
The AI Power Crisis: Why Nuclear Is Mandatory
Every AI boom creates an infrastructure crisis. The internet boom needed fiber optics. The smartphone boom needed 4G towers. The AI boom needs nuclear power plants.
Here's the brutal math:
AI Data Centers Are Electricity Monsters
Specific examples:
- Training GPT-4: Required 25,000 Nvidia A100 GPUs running 100 days = 100MW continuous draw = $35M electricity cost
- Microsoft's AI infrastructure: 500K+ Nvidia H100 GPUs consuming 500MW+ (equivalent to powering Seattle)
- Meta's AI training clusters: 350K+ GPUs, 350MW+, building 1GW campus in Louisiana
- xAI (Elon Musk) Colossus cluster: 100K H100 GPUs, 150MW, Memphis Tennessee—single largest AI training facility
By 2030, AI data centers will consume 500-800GW globally—equivalent to 500-800 nuclear reactors. That's 8-12% of total global electricity generation.
Why Solar/Wind Can't Work at AI Data Center Scale
Technologists love to claim "just use renewables." Reality is harsher:
Problem #1: Intermittency
- Solar produces zero power 12+ hours/day (nighttime). Clouds reduce output 50-80%.
- Wind capacity factor 25-40% (only produces rated power 25-40% of time).
- AI training can't stop/start based on weather—models trained continuously for weeks/months. Interruption = restart from scratch = $10M-50M wasted compute.
Problem #2: Land Requirements
- 1GW nuclear plant: 1 square mile
- 1GW solar farm: 50-75 square miles (accounting for capacity factor)
- 1GW wind farm: 150-250 square miles
AI data centers need concentrated power at single location—can't spread across 250 square miles.
Problem #3: Battery Economics Don't Work
- 1GW AI data center running 24/7 needs: 24 GWh daily battery storage to cover nighttime
- Battery cost: $300/kWh × 24,000,000 kWh = $7.2 billion in batteries alone
- Battery lifespan: 10 years → $720M annual amortization
- vs Nuclear plant: $5-8B upfront, 60-80 year lifespan, zero fuel cost variability
Nuclear is the only technology providing: 1) 24/7 reliable baseload, 2) Zero carbon emissions (meeting climate commitments), 3) High power density (concentrated at data center site), 4) Cost-competitive at scale.
"AI needs reliable, carbon-free baseload power. That's nuclear. We're restarting Three Mile Island because grid power can't meet our data center demands at the scale and reliability AI requires. This is just the beginning—we'll need 10-20 nuclear plants dedicated to Microsoft data centers by 2035."
The Tech Nuclear Pivot: Big Tech's $50B Bet
Microsoft + Constellation Energy
Deal: 20-year power purchase agreement (PPA) to restart Three Mile Island Unit 1 nuclear reactor (835MW). Reactor renamed "Crane Clean Energy Center."
Timeline: Reactor operational 2028. Dedicated power for Microsoft Azure AI data centers in Pennsylvania/Mid-Atlantic.
Cost: Microsoft paying premium electricity rates ($80-100/MWh vs grid $40-60/MWh) for guaranteed 24/7 carbon-free power.
Google + Kairos Power
Deal: Agreement to purchase 500MW total capacity from 7 small modular reactors (SMRs) built by Kairos Power.
Technology: Kairos Hermes reactor uses molten salt cooling (next-gen design, inherently safer than water-cooled).
Timeline: First reactor 2030, full deployment 2035.
Investment: Google investing in Kairos (undisclosed amount, estimated $500M+).
Amazon + X-energy
Deal: Amazon leading $500M funding round for X-energy (SMR developer). Agreement to deploy 4 SMR reactors (320MW total) powering AWS data centers in Washington state.
Technology: Xe-100 reactor (80MW modular design), pebble-bed fuel design.
Timeline: First reactor operational 2030-2031.
Oklo (Sam Altman) + OpenAI/Microsoft
Company: Sam Altman (OpenAI CEO) is chairman/major investor in Oklo, SMR startup. Went public via SPAC (2024), market cap $3.5B (Feb 2026).
Technology: Aurora fast reactor (15MW microreactors), runs on nuclear waste fuel.
Strategy: Targeting data centers, remote industrial sites, military bases. Obvious synergy: Altman owns OpenAI (needs power) + Oklo (provides power).
Total Big Tech nuclear commitments: $50B+ through 2035 (combining direct investments, PPAs, equity stakes).
Contrarian Take
Everyone's worried about Meta's metaverse spending. They should be. But what they miss is that Meta's AI advertising engine is so far ahead, they can burn $10B yearly on moonshots and still dominate.
SMRs: The Nuclear Breakthrough Enabling the Renaissance
Legacy nuclear plants failed economically because they're gigantic, bespoke projects taking 10-15 years to build with massive cost overruns. Small Modular Reactors (SMRs) change everything.
What Makes SMRs Different
| Feature | Legacy Large Reactor | Small Modular Reactor (SMR) |
|---|---|---|
| Power Output | 1,000-1,600 MW | 50-300 MW (modular—deploy multiple units) |
| Construction Time | 10-15 years (often longer with delays) | 3-5 years (factory-built modules) |
| Capital Cost | $8-15B per reactor | $500M-2B per module |
| Manufacturing | Built on-site (weather delays, quality control issues) | Factory-built (assembly line, consistent quality) |
| Safety Systems | Active (requires pumps, electricity to cool) | Passive (gravity/physics-based, fail-safe) |
| Siting Flexibility | Requires massive cooling (ocean/river access) | Air-cooled or minimal water (deploy anywhere) |
| Levelized Cost ($/MWh) | $90-150/MWh | $60-90/MWh (at scale) |
SMRs solve nuclear's three historic problems: 1) Construction time (3-5 years vs 10-15 years), 2) Capital cost ($500M-2B vs $8-15B), 3) Siting flexibility (can deploy at data centers vs requiring ocean access).
Leading SMR Technologies
1. NuScale Power (SMR) — First NRC-Certified SMR
- Technology: 77MW light-water reactor (LWR) modules. Deploy 4-12 modules per site for 300MW-900MW total capacity.
- Status: NRC certified August 2023 (first SMR approved in US history). First deployment: Carbon Free Power Project (Utah, 2029-2030).
- Customers: 19 signed agreements across 12 countries. US Department of Energy $600M+ funding.
- Public company: NuScale Power (SMR), market cap $2.1B (Feb 2026). Revenue-generating 2030+.
2. Oklo (OKLO) — Sam Altman's Fast Reactor
- Technology: Aurora fast reactor, 15MW microreactor. Runs on nuclear waste fuel (recycled uranium/plutonium from legacy reactors).
- Advantage: Solves nuclear waste problem while generating power. Can operate 20 years without refueling.
- Status: Fuel recycling facility under construction Idaho National Lab. First commercial reactor targeting 2027-2028 deployment.
- Public company: Oklo Inc. (OKLO), market cap $3.5B (Feb 2026). Sam Altman chairman + major shareholder.
3. Kairos Power (Google Partner) — Molten Salt Design
- Technology: Kairos Hermes reactor uses molten fluoride salt cooling instead of water. Operates at atmospheric pressure (no pressure vessel needed = simpler, safer).
- Capacity: 140MW per module.
- Status: Building Hermes test reactor Tennessee (2027 operational). Commercial deployment 2030+.
- Private company: Kairos raised $300M+ (Google investor). Expected IPO 2027-2028.
4. X-energy (Amazon Partner) — Pebble Bed
- Technology: Xe-100 reactor, 80MW capacity. Uses TRISO pebble fuel (uranium encased in graphite spheres—meltdown-proof design).
- Status: Partnered with Dow Chemical (first deployment at Dow facility, Texas). Amazon $500M investment (2024).
- Private company: Valued at $3B+ post-Amazon investment. IPO likely 2027.
5. TerraPower (Bill Gates) — Natrium Sodium-Cooled
- Technology: Natrium reactor uses liquid sodium cooling + molten salt energy storage. 345MW electric output.
- Advantage: Integrated thermal storage allows 500MW peak output (matching grid demand fluctuations).
- Status: Building first reactor Wyoming (Kemmerer, replacing retired coal plant). Operational 2030.
- Private company: Bill Gates personally funded $1B+. Not seeking outside capital currently.
The Bro Billionaire Nuclear Stocks
Constellation Energy
The Nuclear Giant. Constellation operates 21 nuclear reactors (largest US nuclear fleet), generating 20% of America's carbon-free electricity. Microsoft 20-year power purchase agreement ($1B+ value) to restart Three Mile Island Unit 1 (2028). Additional data center PPAs in pipeline with Google, Amazon. Revenue $23B (2025), growing 12% as power demand surges.
Why #1: Only pure-play public nuclear operator at scale. Microsoft deal validates "nuclear for data centers" thesis—expect 10-20 similar deals through 2030. Existing reactor restarts cheaper/faster than new builds (Three Mile Island restart cost $1.6B vs $8B+ new reactor). Operating leverage—incremental power sales at 70%+ gross margin. Recession-proof (regulated utility model + long-term PPAs).
Risks: Regulatory (NRC can delay reactor restarts), nuclear safety incidents globally (public perception), competition from SMRs long-term. Valuation reasonable (18x P/E).
EXTREME CONVICTION — 10-15% PORTFOLIOCameco Corporation
The Uranium King. Cameco is world's largest publicly-traded uranium producer (18% global market share). Operates McArthur River mine (Canada—world's largest high-grade deposit) + Cigar Lake mine. Revenue $2.9B (2025), growing 20%+ as uranium prices surge $40/lb (2020) → $82/lb (2026) → $120-150/lb (2030E projected).
Why #2: Uranium supply-demand fundamentals explosive. Demand doubling 2025-2035 (100+ new reactors + SMRs). Supply constrained—years of underinvestment (uranium mining declined 2015-2020 due to Fukushima). Cameco controls tier-1 assets (high-grade, low-cost). Every 1GW nuclear reactor needs 200 tons uranium annually = 100+ new reactors = 20,000 tons/year new demand vs 50,000 tons/year current supply.
Risks: Uranium price volatility, geopolitical (Russia supplies 20% global uranium—sanctions risk), mining delays (permitting, labor). Valuation extended (35x P/E) but justified by uranium bull market.
VERY HIGH CONVICTION — 8-12% PORTFOLIOOklo Inc.
The Sam Altman SMR Bet. Oklo developing 15MW Aurora fast reactors running on recycled nuclear waste. Sam Altman (OpenAI CEO) is chairman + major shareholder—strategic synergy (OpenAI needs power, Oklo provides it). Went public May 2024 via SPAC. Market cap $3.5B despite zero revenue—pure speculation on future.
Why #3: If Oklo succeeds, first-to-market micro-reactor advantage. Addressable market: 10,000+ data centers globally, remote industrial sites, military bases. Altman connection = preferential OpenAI/Microsoft deployment (speculation but logical). Fast reactor technology proven (Navy submarines use fast reactors). Pre-revenue = all upside if execution works.
Risks: Extreme—pre-revenue, unproven commercial technology, regulatory (NRC approval pending), competition from NuScale/X-energy. Valuation insane (infinite P/E, $3.5B market cap on $0 revenue). High-risk speculative play.
SPECULATIVE — 2-4% PORTFOLIONuScale Power
The First NRC-Certified SMR. NuScale's 77MW light-water SMR received NRC certification August 2023—first SMR approved in US. Technology de-risked (regulatory cleared). First deployment: Utah's Carbon Free Power Project (6 modules, 462MW total, 2030 operational). 19 signed agreements globally (US, Canada, Poland, Romania, Czech Republic).
Why #4: Regulatory approval = massive de-risking vs competitors. Proven light-water technology (safest, most validated design). DOE backing ($600M+ funding). International expansion (Poland ordering 6-12 modules replacing coal plants). Revenue inflection 2028-2030 as first plants operational. Currently pre-revenue but validated technology.
Risks: Execution (delays common in nuclear), capital-intensive (needs financing for plant builds), competition from Kairos/X-energy/Oklo. Loss-making currently (burning $100M+/year). Valuation high (117x sales on $18M rev) but improving as deployments progress.
MODERATE CONVICTION — 3-5% PORTFOLIOVistra Energy
The Diversified Power Play. Vistra operates 6 nuclear reactors + natural gas plants + energy retail business. Benefiting from power price surge (Texas electricity spiking due to data center demand). Pivoting toward data center power agreements. Revenue $15B (2024), 35% EBITDA margins.
Why #5: Lower-risk nuclear exposure vs pure SMR plays. Diversified power generation = recession-resistant. Trading at value multiples (11x P/E) despite power demand boom. 3.2% dividend. Texas exposure = hottest data center market (cheap land, business-friendly). Activist investor Elliott Management pushing for data center power PPA strategy.
Risks: Commodity exposure (natural gas prices), Texas regulatory risk (ERCOT market), nuclear smaller part of business (30% revenue) vs Constellation (100%). Lower growth vs pure nuclear plays.
MODERATE CONVICTION — 4-6% PORTFOLIOThe Bottom Line: Nuclear Is Mandatory for AI
The AI boom cannot proceed without nuclear power. Solar/wind can't provide 24/7 baseload at data center scale. Batteries don't work economically. Natural gas emits carbon (violates climate commitments). Microsoft restarting Three Mile Island, Google ordering 7 SMRs, Amazon investing $500M in X-energy—these aren't ideological bets. They're infrastructure necessity.
The nuclear renaissance is happening now: 100+ reactors under construction, uranium prices doubling, SMR technology de-risked. Constellation Energy, Cameco, Oklo, NuScale, and Vistra are positioned to dominate a $300B market through 2040. This is the most obvious mega-theme of the decade.
Nuclear powers AI. AI powers the economy. Own the nucleus.