The AI revolution has a dirty secret: it runs on staggering amounts of electricity. A single hyperscale AI training cluster now consumes 100MW or more — equivalent to powering 80,000 homes. With over 500 data center projects in various stages of construction across the US alone, total incremental power demand from AI infrastructure is projected to reach 45-60 GW by 2030. That’s roughly the output of 45 nuclear reactors — and the commodity markets aren’t priced for it.

Overview

The explosion of generative AI, large language models, and enterprise AI adoption has triggered a data center building boom without historical precedent. Microsoft, Google, Amazon, Meta, and Oracle collectively committed $260+ billion in data center capex for 2025-2027. But the true demand shock extends beyond Big Tech — sovereign AI initiatives, defense applications, and a new generation of AI-native startups are all racing to secure compute capacity.

US data center power consumption is forecast to double from 35 GW in 2024 to 70+ GW by 2030, according to Goldman Sachs and McKinsey estimates. Globally, the figure could reach 130 GW — representing approximately 4-5% of total world electricity generation, up from under 2% today.

The math is simple but the implications are profound. An NVIDIA H100 GPU consumes 700W under load. A single training cluster with 25,000 GPUs draws 17.5 MW just for compute — before cooling, networking, and facility overhead push the total to 35-50 MW. Next-generation clusters built around NVIDIA’s B200 and GB300 platforms are targeting 100-300 MW per facility. Microsoft’s planned Stargate complex in Wisconsin is designed for 5 GW — larger than many countries’ total grid capacity.

This isn’t future speculation. Permits are filed. Substations are being built. Gas turbines are being ordered. The commodity demand is arriving now.

Key Impact Channels

Primary: Natural Gas — The Bridge Fuel That Became Permanent

Natural gas is the immediate winner of the AI power surge. Despite the green energy rhetoric from hyperscalers, the reality is that renewables cannot deliver the 24/7 baseload reliability that data centers require. Power Purchase Agreements (PPAs) for wind and solar help with carbon accounting, but the electrons hitting the rack come overwhelmingly from gas-fired generation.

Metric	Current (2025)	Projected (2028)	Delta
US data center gas demand	4.2 Bcf/d	8.5 Bcf/d	+102%
Share of US gas demand	4.8%	8.9%	+4.1pp
New gas plant capacity for DC	12 GW	35 GW	+192%

Key data points driving the thesis:

• Amazon signed a deal to purchase the 960 MW Talen Energy Susquehanna nuclear-adjacent gas plant in Pennsylvania for direct data center power
• Dominion Energy reports that data center load in Northern Virginia (the world’s largest data center market) grew 32% in 2025 and is accelerating
• ERCOT (Texas grid) has received 65 GW of large-load interconnection requests, the majority data center-related — against a current grid capacity of 85 GW
• Henry Hub natural gas futures have climbed to $4.20/MMBtu, partly reflecting growing baseline demand from the power sector

The gas demand impact creates a virtuous cycle for producers. Unlike seasonal heating demand, data center load is flat 24/7/365 — the highest-quality demand profile in the energy market. This structural floor under gas prices benefits EQT Corp (EQT), Coterra Energy (CTRA), and Antero Resources (AR) disproportionately.

Secondary: Uranium — The Nuclear Renaissance Accelerator

If gas is the bridge, nuclear is the destination. Every major hyperscaler has publicly committed to nuclear power for data centers, and the uranium market is tightening fast.

Company	Nuclear Initiative	Capacity Target
Microsoft	Three Mile Island restart (Constellation deal)	835 MW
Google	Kairos Power SMR fleet contract	500 MW by 2030
Amazon	Talen Energy nuclear PPA + X-energy SMR investment	960 MW + SMR pipeline
Oracle	Planning 3 SMR-powered data centers	1 GW+
Meta	RFP for nuclear generation (2025)	1-4 GW

The combined committed and planned nuclear capacity for tech companies now exceeds 10 GW — representing approximately 20 million lbs of annual U3O8 demand once operational. Current global uranium production is 145 million lbs against demand of 180 million lbs, with a 35 million lb deficit already being filled from secondary supplies (government stockpiles, underfeeding, recycling) that are rapidly depleting.

Spot uranium prices at $82/lb are well above the $50-60 restart threshold but still below the $100+ levels that many analysts see as necessary to incentivize sufficient greenfield mine development. The Sprott Physical Uranium Trust (SRUUF) and producers like Cameco (CCJ), Kazatomprom, and NexGen Energy (NXE) are positioned for a multi-year bull run.

Small Modular Reactors (SMRs) are the key inflection point. NuScale, X-energy, Kairos, and TerraPower are all targeting first commercial deployments in 2029-2031. If SMRs deliver on cost and timeline promises, data center uranium demand could triple by 2035.

Tertiary: Copper — The Wiring of the AI Economy

Every megawatt of data center capacity requires approximately 20-40 tonnes of copper — for power distribution, busbars, cabling, cooling systems, transformers, and grid interconnection. At the projected build-out rate, AI data centers alone could add 1.5-2.5 million tonnes of cumulative copper demand through 2030.

This arrives into a copper market already facing a structural deficit. Global refined copper demand is projected at 28.5 million tonnes in 2026 against supply of 27.8 million tonnes — a 700,000 tonne gap. Data center demand wasn’t even in most forecasters’ models two years ago.

Critical copper pressure points:

• Transformers: Lead times for large power transformers have stretched to 3-4 years, each requiring 15-30 tonnes of copper windings
• Grid upgrades: Utilities are spending $100B+ on transmission and distribution upgrades to serve new data center loads
• Cable demand: High-voltage DC (HVDC) submarine and land cables for renewable integration use 2-3x more copper per MW than AC alternatives

Copper futures at $4.85/lb are near multi-year highs, and the structural deficit supports prices above $5.00/lb through 2027. Freeport-McMoRan (FCX), Southern Copper (SCCO), and Ivanhoe Mines (IVN) are the primary equity beneficiaries.

Winners

Tier 1 — Direct Power Demand:

• EQT Corp (EQT) — Largest US natural gas producer, Appalachian Basin proximity to Virginia data center corridor. Trading at 5.8x forward EV/EBITDA vs. historical average of 7.5x.
• Cameco (CCJ) — Premier Western uranium producer, locked in long-term contracts with upside repricing. Cigar Lake and McArthur River operations are tier-1 assets.
• Constellation Energy (CEG) — Owns the largest US nuclear fleet (21 GW). The Microsoft TMI deal is just the beginning — every reactor in their portfolio is a potential hyperscaler PPA target.

Tier 2 — Infrastructure Build-Out:

• Freeport-McMoRan (FCX) — Copper exposure to both data center build-out and broader electrification. Grasberg ramp-up provides volume growth.
• Quanta Services (PWR) — Largest US electrical contractor; directly builds the grid infrastructure connecting data centers to the power system.
• Eaton Corp (ETN) — Power management and distribution systems are inside virtually every data center. Backlog at record levels.

Tier 3 — Enablers:

• Sprott Physical Uranium Trust (SRUUF) — Pure uranium price exposure without mining operational risk.
• Vistra Energy (VST) — Texas power generator benefiting from ERCOT data center load growth. Stock already +180% in 18 months but power market re-rating has further to run.
• NexGen Energy (NXE) — Rook I project in Saskatchewan’s Athabasca Basin is the highest-grade undeveloped uranium deposit globally.

Losers

Tier 1 — Grid Stress Casualties:

• Residential ratepayers — Data center power demand is pushing electricity prices higher across major markets. Virginia residential rates are up 18% since 2023, with data centers consuming 30%+ of Dominion’s load. Regulatory backlash is building.
• Renewable-only utilities — Companies like NextEra (NEE) that rely heavily on intermittent generation face the awkward reality that hyperscalers want baseload, not weather-dependent power.

Tier 2 — Resource Competition Losers:

• Traditional copper consumers — Auto manufacturers, construction companies, and appliance makers face higher copper input costs and longer delivery times as data centers absorb transformer and cable production capacity.
• Non-AI industrial electricity users — Manufacturers in data center corridors face rising power costs and potential reliability concerns as grids strain under new loads.

Tier 3 — Stranded Narrative:

• Pure-play solar/wind developers — While PPAs with hyperscalers generate headlines, the 24/7 matching requirements increasingly favor nuclear and gas. Solar/wind’s capacity factor disadvantage (25-35% vs. 90%+ for nuclear/gas) means 3-4x the nameplate capacity is needed per actual MW delivered.
• Grid battery storage (short duration) — 4-hour lithium batteries cannot bridge overnight gaps for 100MW+ loads. Long-duration storage remains uneconomic at scale.

Trading Note

The AI-data-center-commodity nexus is early innings — most commodity analysts haven’t fully incorporated hyperscale build-out into their demand models. The IEA’s World Energy Outlook 2025 added a “digital demand” scenario for the first time, projecting 2x higher electricity growth than their base case.

Where the edge remains:

1. Gas basis differentials: Watch Appalachian (Dominion South) gas prices vs. Henry Hub. As Virginia/Pennsylvania data center demand absorbs local supply, basis premiums could widen significantly — benefiting Appalachian-focused producers like EQT and Antero.

2. Uranium contract cycle: 40% of global utility uranium contracts expire in 2025-2027. Recontracting at $75-80/lb vs. legacy $45-55/lb prices represents massive margin expansion for Cameco and Kazatomprom. Tech company demand adds a new buyer category competing for limited supply.

3. Copper transformer bottleneck: Large power transformer lead times are the binding constraint on data center deployment. Companies that control transformer copper supply chains — or the transformer manufacturers themselves (Hitachi Energy, Siemens Energy) — have structural pricing power.

4. Pair trade: Long CEG (nuclear fleet owner) / Short AES (emerging market + renewables-heavy utility) captures the baseload vs. intermittent power divergence driven by data center demand preferences.

Risk to the bull case: AI model efficiency improvements (smaller models, better inference chips) could moderate per-query power consumption. A major data center accident or grid failure could trigger regulatory moratoria. China’s data center build-out may slow if US export controls limit access to advanced GPUs. And a sharp economic downturn could cause hyperscalers to defer capex — though the competitive dynamics of AI make unilateral pullbacks unlikely.

Bottom line: The commodity markets are underpricing the most significant new demand driver since China’s industrialization. AI data centers will consume more electricity by 2030 than most G20 nations use today. Natural gas, uranium, and copper are the three pillars of this build-out — and the demand curve is just beginning to inflect. Position for a multi-year structural demand shock, not a cyclical trade.

ai data-center copper natural-gas uranium energy

Share X / Twitter LinkedIn Email

Methodology

How to read this Impact Map

CommodityNode Signal Reports combine directional sensitivity, supply-chain structure, category overlap, and linked thematic context. Treat the percentages and correlations as research signals designed to accelerate deeper diligence, not as financial advice. Read our full methodology.