AI/Data Center Electricity Demand Elasticity Model
3/8/26
SEQH Capital Research
AI Data Center Electricity Demand Elasticity Model
Tear Sheet – March 7, 2026
Why This Report Exists
The AI revolution is producing a structural break in U.S. and global electricity demand unseen since post-war industrialization. Data center electricity consumption is projected to more than double from 460 TWh in 2024 to 945–1,587 TWh by 2030. This report constructs a quantitative demand elasticity framework demonstrating that AI/data center power demand is functionally price-inelastic, creating a persistent rightward shift in the electricity demand curve that structurally elevates wholesale power prices and creates an asymmetric advantage for nuclear energy as the only generation source capable of delivering firm, carbon-free, 24/7 baseload power at the scale data centers require.
The AI Power Surge
U.S. faces a projected 20% increase in total electricity consumption by 2030, with data centers accounting for 30–50% of all net-new demand.
U.S. data center demand could reach 134.4 GW by 2030, up from 61.8 GW in 2025 (per 451 Research).
A single high-end AI GPU consumes over 400 watts. AI-capable racks exceed 100 kW per rack vs. 7 kW for standard enterprise racks.
DeepSeek-R1 and OpenAI’s o3 consume over 33 watt-hours per long prompt, more than 70x lightweight models.
PUE gains have plateaued at 1.54–1.58 since 2014, meaning demand growth translates near-linearly to grid load growth with no efficiency offset.
The Elasticity Framework
Data center electricity demand operates under a fundamentally different elasticity regime than traditional consumers. SEQH’s model assigns data center demand an effective short-run price elasticity approaching zero (ε ≈ 0) and long-run elasticity of −0.05 to −0.10.
Why data center demand is near-perfectly inelastic:
Revenue per MWh consumed dwarfs the power cost by orders of magnitude.
Multi-year colocation agreements and SLAs make load reduction operationally impossible.
24/7/365 operational mandate, no “off-peak” for AI inference serving global users.
Hyperscale data centers represent 500M–2B+ dollars in sunk capital, marginal power cost always lower than idling cost.
AI compute is existential competitive infrastructure for hyperscalers, power cost is a rounding error.
The mathematical consequence: when perfectly inelastic demand shifts rightward, the entire burden of equilibrium adjustment falls on price.
Merit Order Dynamics
When data center demand shifts the aggregate demand curve rightward, clearing prices jump nonlinearly as dispatch moves from 45 dollars per MWh CCGT costs into 80–150+ dollars per MWh peaker territory.
Federal Reserve Bank of Dallas: mid-capacity data center build-out could increase wholesale electricity prices by 20–30% by 2030.
ERCOT (Texas) faces a projected 45% wholesale price increase in 2026.
Structural supply-demand gap of 17–35 GW by 2030.
The Nuclear Thesis
Capacity factor dominance: U.S. nuclear fleet operates at 92.5% capacity factor vs. 56% for natural gas CCGT, 35% onshore wind, and 25% utility-scale solar.
Hyperscaler commitments: Every major hyperscaler has committed to nuclear power. Total announced commitments exceed 10 GW, with Meta alone securing up to 7.7 GW. Microsoft is paying approximately 110 dollars per MWh for TMI restart power, roughly double regional wind PPAs, the quantified market price of reliability.
Risk-adjusted LCOE: When backup storage costs and carbon risk premiums are incorporated, nuclear LTO and reactor restarts deliver the lowest risk-adjusted LCOE of any firm power source at approximately 40 dollars per MWh.
Revenue sensitivity: For an 800 MW nuclear plant, a 10 dollars per MWh increase in wholesale prices translates to approximately 65M in incremental annual revenue. Under the bull scenario’s 35 dollars per MWh price increase, a single large reactor captures an additional ~228M annually.
Uranium Supply Chain & HALEU Bottleneck
WNA projects uranium demand rising 28% by 2030 and doubling by 2040.
Current global reactor requirements: ~179M lbs U₃O₈ vs. primary mine production of 140–150M lbs, creating a 30–40M lb annual supply gap.
No commercial-scale Western HALEU production facility exists, a critical-path constraint for SMR buildout.
ASP Isotopes (ASPI) through QLE is developing laser-based enrichment capable of producing HALEU, positioned at a chokepoint with significant pricing power.
Key Risks
AI efficiency breakthroughs could moderate the demand trajectory.
Nuclear regulatory delays could slow capacity additions.
Interconnection bottlenecks may limit pace of grid connection.
Renewable + storage cost declines could compress wholesale prices.
Upside risks: faster AI adoption, renewable deployment shortfalls, geopolitical uranium supply disruption.
Want the Full AI Data Center Electricity Demand Elasticity Model?
[READ THE COMPLETE DEMAND ELASTICITY FRAMEWORK]
The full report includes a proprietary quantitative model and investable framework unavailable elsewhere:
Complete demand shock quantification: 460 TWh to 945–1,587 TWh build-up with Goldman Sachs and 451 Research projections, GPU power density analysis, and PUE plateau evidence
Full elasticity framework with empirical NBER and Duke University parameters, SEQH’s data center near-zero elasticity model, and mathematical proof of the price transmission mechanism
Merit order dynamics and price transmission modeling: supply curve convexity, inelasticity amplifier effect, and the Fed Dallas wholesale price impact analysis
SEQH Capital wholesale electricity price model (2025–2030): base, bull, and bear scenarios with wholesale price trajectories from 47 to 82 dollars per MWh
Nuclear capacity factor dominance analysis with equivalent capacity calculations to match 100 MW nuclear reliability across gas, wind, and solar
Hyperscaler nuclear commitment tracker: 10+ GW of announced deals across Microsoft, Meta, Amazon, Google with PPA pricing vs. renewables
Risk-adjusted LCOE comparison across all generation sources including backup storage costs and carbon risk premiums
Nuclear revenue sensitivity model: per-plant incremental revenue at each 10 dollars per MWh wholesale price step
Uranium supply-demand deficit model with Goldman Sachs cumulative deficit projections and HALEU bottleneck analysis
ASPI/QLE positioning at the HALEU chokepoint: Western HALEU supply vs. SMR demand gap and enrichment requirement modeling
AI data center demand is near-perfectly inelastic and growing exponentially. The grid can’t keep up. Nuclear is the only answer at scale. This report gives you the quantitative framework to see why.
Available exclusively to SEQH Capital Research paid subscribers. Upgrade to access →
