AI Is Devouring Electricity. Tech Giants Are Racing Into Nuclear. How Can You Participate?

Every Major Tech Company Is Chasing the Same Thing

This week, Alibaba and China National Nuclear Corporation (CNNC) formed a joint nuclear energy company in Ningbo, China (registered capital: RMB 250 million). The goal: dedicated power supply for AI data centers.

This is not an isolated event. Nearly every trillion-dollar tech company is making the same move:

• Microsoft: Signed a 20-year agreement with Constellation Energy to restart Three Mile Island, 800 MW dedicated to data centers
• Google: Contracted 500 MW of SMR power from Kairos Power; invested in Elementl Power for 1.8 GW of reactor capacity
• Amazon: Investing $20B+ to convert Three Mile Island into a nuclear data center hub; acquired a $650M data center adjacent to Susquehanna nuclear plant
• Alibaba: Formed a nuclear JV with CNNC; plans to invest $53B+ in AI over the coming years

From Silicon Valley to Hangzhou, trillion-dollar companies are simultaneously betting on the same energy source. This is not coincidence. It is a structural judgment: AI compute needs 24/7 uninterrupted power. Wind and solar cannot guarantee that. Nuclear can.

Why? IEA data: global data center electricity consumption was approximately 460 TWh in 2024. By 2030 it will exceed 1,000 TWh—more than double. Goldman Sachs projects 165% growth. U.S. electricity demand will hit all-time highs in 2025-2026, breaking nearly 20 years of flat growth.

---

Don't Rush to Buy. You Might Already Be Late.

Everything above can be found with a few Google searches. "AI needs electricity, nuclear is the answer, uranium is rising" is one of the most crowded narratives of 2026.

If this information is what made you start paying attention to nuclear, spend one minute looking at how much the obvious plays have already run:

Constellation Energy is up nearly 8x in five years. If you are only now researching the "nuclear renaissance," you need to soberly recognize: the market has already priced in much of the easiest-to-understand narrative ("AI needs power → nuclear good → buy utilities").

An even more telling signal: CEG dropped 20.6% in 20 days in January 2026. Why? Because FERC (the Federal Energy Regulatory Commission) rejected a key behind-the-meter deal, and the market suddenly realized that "tech giants easily buying nuclear power" faces real resistance.

So the real question is not "Is nuclear power good?" It is: at which fracture point in the nuclear value chain can you find value that hasn't already been priced in?

The nuclear fuel cycle has four steps:

Most people discussing nuclear investment only look at Step 1 (uranium price) and the end user (utility companies that are already up 8x). They completely miss four deeper structural issues.

---

Fracture Point 1: Uranium Isn't the Bottleneck. Enrichment Is. And Russia Has You by the Throat.

Advanced reactors (especially SMRs) require HALEU (High-Assay Low-Enriched Uranium), with U-235 concentration between 5-20%, far higher than the 3-5% used in conventional reactors.

Here is the problem: Russia is the only country in the world with commercial-scale HALEU enrichment capability.

Not "the primary supplier." The only one.

The U.S. signed the Prohibiting Russian Uranium Imports Act in May 2024. But simultaneously granted waivers until 2028. Why? Because without Russian enrichment services, existing U.S. nuclear plants could face fuel shortages.

The U.S. DOE invested $2.7 billion in January 2026 to build domestic enrichment capacity. But full strategic independence is projected for 2032-2036. Centrus Energy launched its HALEU pilot cascade in October 2023. Commercial-scale operations are years away.

What does this mean? Everyone is saying "AI needs SMRs," but the fuel these SMRs need will depend on Russia for the next 6-10 years.

If oil was the geopolitical weapon of the 20th century, uranium enrichment capability is becoming the geopolitical weapon of the 21st.

Investment implication: Centrus Energy (the only U.S. company with an NRC-approved HALEU pilot operation) is not a "uranium mining stock." It is a "geopolitical security positioning stock." Most uranium ETFs focus on mining companies and have zero exposure to the enrichment segment's valuation.

---

Fracture Point 2: Tech Giants Want to Bypass the Grid. But the Regulator Just Fired a Shot.

The power purchase agreements Microsoft, Google, and Amazon are signing are not ordinary bulk procurement contracts. They are doing something far more fundamental: "behind-the-meter" connections.

Plugging data centers directly into nuclear power plants. Completely bypassing the public grid. No transmission and distribution networks. No transmission fees. No public grid capacity constraints.

But in November 2024, FERC (the Federal Energy Regulatory Commission) fired the first shot.

Amazon acquired a 960 MW data center campus adjacent to the Susquehanna nuclear plant for $650 million from Talen Energy, with a behind-the-meter power agreement. The original arrangement allowed 300 MW; they wanted to expand to 480 MW. FERC voted 2-1 to reject the expansion.

The reasoning:

• Grid reliability concerns
• Cost-shifting: when large consumers go off-grid, grid maintenance costs fall on remaining users
• If approved, it would set a precedent for every tech giant to follow

American Electric Power and Exelon filed complaints opposing the deal. Talen is appealing in the Fifth Circuit Court of Appeals as of January 2025.

The essence of this trend: tech companies want to transform from customers of public infrastructure into their own power utilities. But public infrastructure won't go down without a fight.

When was the last time something similar happened? The railroad era of the late 19th century. Rockefeller built his own oil pipelines. But what happened next? The Hepburn Act of 1906 and subsequent antitrust actions. Every time private capital tries to control public infrastructure, regulation eventually intervenes.

Investment implication: utilities that own operating nuclear generation capacity (Constellation Energy, Vistra) do hold scarce resources. But their "moat" depends on how much regulators allow behind-the-meter connections. If FERC tightens, these companies' pricing power may actually strengthen (tech giants must buy through the grid, not bypass it). If FERC relents and behind-the-meter scales, pricing power shifts from utilities to tech giants sitting on trillions in cash. Who wins depends on regulation, not supply and demand.

---

Fracture Point 3: Everyone Is Discussing "Electricity." Nobody Mentions "Water."

Nuclear power plants and data centers are both massive water consumers. When they are co-located, the problem compounds.

A typical nuclear power plant consumes 13 to 24 billion liters of water annually. That is 35 to 65 million liters per day.

A medium-sized data center consumes approximately 400 million liters per year, equivalent to the annual water usage of 1,000 households. Large data centers consume 19 million liters daily, matching the water demand of a town of 10,000 to 50,000 people.

Nuclear plant + data center at the same location = double water demand concentrated on the same water source.

France has already experienced cases where climate change raised water temperatures enough to force nuclear plants to reduce output or shut down during heat waves. Chile and Uruguay have seen public protests against data centers consuming local water resources.

Investment implication: not all nuclear plant locations are suitable for AI data center hubs. Coastal nuclear plants (representing 45% of global nuclear capacity) have a natural advantage regarding water constraints. Inland nuclear plants in water-stressed regions could face community opposition and regulatory risk when paired with data centers. Geographic location and water source are variables almost nobody is evaluating when selecting nuclear investment targets.

---

Fracture Point 4: The Jevons Paradox. Does Higher AI Efficiency Mean More Nuclear Demand? Or Less?

This is the question this entire article must honestly confront.

"AI requires massive amounts of electricity" is the core assumption behind nuclear investment. But this assumption is being pulled by two forces simultaneously:

Efficiency is improving faster than anyone expected.

• AI inference costs have dropped 1,000x in three years
• Google's Gemini model achieved a 33x efficiency improvement in one year (May 2024 to May 2025), purely from software optimization
• Small, task-specific AI models consume 90% less power than general-purpose large models
• Stanford AI Index 2025: AI hardware efficiency improving 40% per year, prices declining 30% per year

If this trend continues, by 2030 the power required per AI query could be 1/100th of today. Wouldn't that undermine the entire nuclear demand thesis?

The Jevons Paradox says no. Quite the opposite.

In 1865, economist William Stanley Jevons discovered that as steam engine efficiency improved, England's coal consumption increased rather than decreased. Because efficiency made steam engines cheaper, applications exploded, and total demand far exceeded efficiency savings.

AI is replaying this pattern:

IEA's projections support Jevons: despite rapid per-unit efficiency gains, total global data center electricity consumption will still double from 460 TWh (2024) to 1,000+ TWh (2030).

Investment implication: don't naively assume "AI electricity demand will definitely explode." Monitor a leading indicator: whether the growth rate of total global AI interactions consistently exceeds the rate of decline in per-interaction energy consumption. If it does, Jevons holds and the nuclear thesis is solid. If efficiency ever catches up, part of nuclear's valuation premium evaporates.

---

Uranium Supply-Demand: The Structural Deficit Is Real, But Don't Get Brainwashed by the "Only Goes Up" Narrative

Uranium's fundamentals are genuinely strong. But you need the full picture:

Bull case data (real):

• Global uranium demand in 2025: ~68,900 metric tons. By 2040: over 150,000 metric tons
• Current production meets only 74-90% of demand. Annual deficit: ~45 million pounds
• Kazakhstan (40%+ of production) announced 10% production cuts for 2026
• Uranium price broke $100/lb. Bank of America projects peak of $135/lb
• ~15 new reactors coming online in 2026, adding 12 GW

Bear case data (also real):

• Uranium price has already risen 5x from $20 to $100. Uranium historically has extreme cyclicality: in 2007 it also ran from $7 to $136, then crashed to $40
• Kazakhstan's production cut is strategic (value over volume); they can ramp up anytime if prices are high enough
• Of 65 reactors under construction globally, many face delays and cost overruns
• Uranium ETF (URA/URNM) holdings are highly concentrated in a few miners

---

Fracture Point 5: Nuclear's Scarcest Resource Isn't Uranium. It Isn't Money. It's People.

Everyone discusses uranium supply, construction costs, regulatory approvals. Almost nobody mentions the most fundamental bottleneck: who will build and operate these reactors?

U.S. nuclear industry survey, 2024: 63% of nuclear power manufacturing employers found hiring qualified workers "very difficult." Nuclear construction employment is projected to grow 9.2%, but talent supply is far behind.

The employment outlook for nuclear engineers is surprisingly bleak: the Bureau of Labor Statistics projects only 1% growth in nuclear engineering jobs from 2023 to 2033. But approximately 700 job openings per year arise from retirements alone. On one hand, the industry expects explosive growth. On the other, the talent pipeline is running dry.

New technologies make things worse. SMRs and advanced reactors require specialized talent in safety engineering, systems integration, digital reactor monitoring, and advanced materials. These are not positions that can be filled through quick training programs. Developing a qualified nuclear engineer takes 8-10 years.

France plans to build 6 EPR2 reactors (potentially 8 more), requiring tens of thousands of specialized workers. The UK, India, and China are simultaneously building 65 reactors. The entire world is competing for the same talent pool.

Investment implication: the workforce shortage is nuclear's least sexy but most lethal constraint. It won't stop nuclear development, but it will systematically extend timelines and inflate costs. Every time you see an SMR company say "commercial by 2030," mentally add 3-5 years. Because building reactors requires people, and there aren't enough.

---

SMR's Troubles Are Real. Don't Be a Nuclear Salesman.

NuScale's flagship project failed. The Carbon Free Power Project (CFPP), America's first commercial SMR project. Original budget: $3.6 billion for 720 MW. By 2023: $9.3 billion for 462 MW. Cost per MWh: $58 to $130 (+124%). Canceled November 2023 after securing only 20% of required capital.

IEEFA's verdict: SMRs are "too expensive, too slow, and too risky" to meaningfully contribute to the energy transition within 10-15 years.

Nuclear waste controversy: some studies suggest SMRs could produce up to 30x more radioactive waste per unit of electricity than conventional reactors.

Renewables remain primary: IEA projects renewables will meet 60% of incremental data center electricity demand by 2035.

NRC approval: a new reactor design takes 6 years and $500 million to certify. NuScale is the only SMR to clear this bar.

These counterarguments don't mean nuclear has no value. They mean: if you can't be bothered to distinguish risk profiles across different value chain segments, you aren't investing, you're chasing a narrative.

---

The Fracture Point Method: Four Paths to Nuclear Investment

Knowing "nuclear is a trend" is useless. The key question: at which fracture point in the value chain do you bet?

---

What Today's Market Data Is Telling Us

XLU (Utilities ETF) trend score is 100. Maximum. Volume at 3.5x. This is not a random sector blip. This is the utilities sector being repriced under AI electricity demand expectations.

But remember the principle from this article's framework: trending data is a micro-anomaly detector, not a macro-trend validator. One day of elevated volume in XLU does not prove nuclear's structural value thesis. It is a trigger to investigate further. And we already did.

---

Two Questions to Think About

Question 1: Constellation Energy is up 793% in five years. If the nuclear narrative is this crowded, why can you still make money? Use this article's framework: which segments are already priced in (CEG, uranium price), and which are still being overlooked (enrichment, workforce, water)? Among the overlooked segments, which is most likely to be discovered by the market first?

Question 2: Google's Gemini achieved a 33x efficiency improvement in one year. If efficiency keeps improving at this rate, by 2030 the energy cost per AI query could approach zero. The Jevons Paradox says "usage will compensate." But what if efficiency wins this time? If AI electricity demand ultimately grows only 50% instead of 165%, which targets in the nuclear investment chain get hurt first?

---

Disclaimer

This article is for investment education and analytical framework sharing only. It does not constitute investment advice. Any assets, ETFs, companies, or financial instruments mentioned are used as teaching examples and do not represent buy or sell recommendations. Investing involves risk; please exercise caution.

---

Trend Reports · Reading markets through data and frameworks Sources: IEA · Goldman Sachs · U.S. DOE · NucNet · World Nuclear Association · Sprott · BNEF · Third Way · IEEFA · ekx.ai