Divergent Compute.AI Economic Think Tank

Markets / Industries / Power & Energy

Industry view · Power & Energy

AI is the grid's biggest customer and its smartest operator at once

Artificial intelligence has become a two-sided force on the power system: data centers are the fastest-growing source of electricity demand in a generation, with global data-center consumption set to roughly double from about 415 TWh in 2024 to ~945 TWh by 2030, while AI software is quietly being woven into generation forecasting, grid operations and interconnection. The money has already moved — gas turbines are sold out to 2029, nuclear plants are being restarted for hyperscalers, and PJM ratepayers are absorbing $9.3 billion in higher bills traced to data-center load.

~945 TWh
Global data-center electricity demand by 2030 (from ~415 TWh in 2024)
IEA
43%
Share of projected US power-demand growth tied to data centers (early 2030s)
NextEra
$269.92
PJM capacity price per MW-day in 2025/26 auction, up from $28.92
electricityrates.com
$21.82B
AI in Power Utilities market size, 2026 (from $17.29B in 2025)
Fortune Business Insights

01 · The thesis

The demand shock is real; the software opportunity is larger but slower

The headline story is brute-force: AI training clusters are consuming power faster than US utilities can build generation and transmission. The IEA projects data-center demand roughly doubling to ~945 TWh by 2030 (from about 415 TWh in 2024), and NextEra's management now attributes 43% of projected US demand growth in the coming years to data centers alone. This has revived gas turbines (GE Vernova's gas backlog stretches into 2029), restarted nuclear plants (Microsoft's 20-year, 835 MW Three Mile Island deal, a ~$1.6B Constellation restart investment), and pushed fuel cells and SMRs from pilot to procurement. The constraint is no longer chips — it is interconnection queues and transformers.

The quieter, more durable story is AI *inside* the system. DeepMind raised the market value of Google's wind output 20% by forecasting supply 36 hours ahead; startups like GridCARE and ThinkLabs are using AI to find stranded grid capacity and collapse interconnection timelines from years to months. The 'AI in Power Utilities' software market — $21.82B in 2026, on a ~19% CAGR — is small beside the trillions in physical capex, but it is where margin and defensibility accrue. Whoever owns the grid's operating intelligence captures value long after the turbine backlog clears.

1Generation

AI both forces new supply and squeezes more from existing assets

Data-center demand is reviving gas turbines, nuclear restarts and fuel cells, while AI forecasting lifts the realized value of intermittent renewables.

GE Vernova, Constellation, Vistra, Bloom Energy, DeepMind forecasting
2Transmission

Transformers and switchgear are the new bottleneck

Grid hardware backlogs have surged as utilities scramble to wire new load; AI is starting to optimize asset planning and routing.

GE Vernova Electrification, Siemens Energy, Hitachi Energy
3Grid operations

AI-native simulation moves from research to live circuits

Machine-learning grid simulators now process a year of hourly power-flow across 100+ circuits in minutes, enabling faster planning and dynamic operations.

ThinkLabs AI, GridCARE, utility control rooms
4Interconnection

Finding stranded capacity becomes a venture category

AI tools surface underutilized grid headroom so data centers can connect in months rather than the typical multi-year queue.

GridCARE, Base Power, hyperscaler siting teams
5Load & demand

Behind-the-meter and VPPs flex against AI's appetite

Distributed assets, batteries and virtual power plants are being marshaled by software to balance volatile data-center load.

Enphase, Lunar Energy, Base Power, Tesla
Pace of AI disruption by stage — Divergent Compute assessment

02 · Public players & exposure

Who routes through, who gets routed around

We plot the listed players on two editorial axes — how exposed each is to AI disruption, against how ready its data, brand and position are to be the answer. The figures in the table are sourced; the placement is our read.

Positioning — editorial assessment, not a sourced metric. Bubble = approximate relative scale.
CompanyStanceThe sourced fact
NextEra EnergyNEEDemand winnerNextEra plans to deliver about 15 GW of new generation to data-center power hubs by 2035, roughly 6 GW of it new gas-fired capacity, and attributes 43% of projected US power-demand growth to data centers.
GE VernovaGEVPicks-and-shovelsQ1 2026 orders rose 71% and backlog reached $163B, with $2.4B of electrification equipment orders tied to data centers — exceeding all of 2025's data-center intake.
Constellation EnergyCEGNuclear restartSigned a 20-year deal with Microsoft to restart the 835 MW Three Mile Island Unit 1 (a ~$1.6B Constellation restart investment) targeting 2028.
VistraVSTFleet monetizerPart of Meta's January 2026 nuclear deals (with Oklo and TerraPower) enabling up to 6.6 GW by 2035; Meta will help fund Vistra's Ohio and Pennsylvania nuclear plants.
Talen EnergyTLNColocation betJune 2025 deal to sell Amazon up to 1,920 MW of nuclear power from its Susquehanna plant through 2042, a ~$18B grid-connected PPA model.
Bloom EnergyBEFuel-cell challengerOracle intends to procure up to 2.8 GW of Bloom's fuel cells; separately a $5B partnership with Brookfield for AI 'factories'.
Siemens EnergyENRTurbine supplierScaling heavy-duty gas turbine production amid record demand; one of three firms supplying roughly two-thirds of global turbine demand.
Enphase EnergyENPHEdge pivotAnnounced an IQ Solid-State Transformer for 800 VDC AI data-center power, with demos targeted in 2026 and volume shipments later in the decade.
GridCAREPRIVCapacity-finderRaised an oversubscribed $64M Series A (May 2026) to find stranded grid power, cutting data-center interconnection from years to months.
Legacy coal utilitiesCOALSqueezed & exposedPJM's market monitor found data centers drove roughly two-thirds of the rise in capacity prices to $269.92/MW-day, pushing about $9.3B in costs onto ratepayers and inviting regulatory backlash.
The map is Divergent Compute’s editorial positioning, offered as a lens, not a measurement. Every figure in the right-hand column is drawn from a named source — see Sources.

03 · The two clocks

The spend, and the payoff

Three clocks are running against the power sector at different speeds — and they are not synchronized.

IEA, Energy and AI (2025); demand roughly doubles by 2030, with AI the dominant driver.

The demand clock is fastest. The IEA projects global data-center electricity use growing roughly 15% a year through 2030 — more than four times faster than total electricity demand — with AI-driven accelerated servers the dominant contributor. That is load arriving in years, not the decades utilities are accustomed to planning around.

The supply clock is slowest. Lead times for a large gas turbine have stretched to several years, with GE Vernova's gas backlog reaching into 2029 (roughly 80 GW by the end of 2025). Transformers and switchgear are now the binding constraint, with GE Vernova's electrification backlog rising sharply — to about $42B by Q1 2026, from $9B at the end of 2022.

The cost clock is the political one. PJM capacity prices rose nearly tenfold to $269.92/MW-day, and analysts have warned of roughly $100 billion or more in cumulative ratepayer costs over the coming years absent intervention — a bill that will shape who is allowed to build, and where.

04 · Private flagships

The AI-native challengers

The companies attacking this industry AI-first, with disclosed funding where available:

GE Vernova

The arms dealer of the AI power buildout

Sells the gas turbines, transformers and grid systems everyone needs and no one can get quickly; data-center orders now outpace its own forecasts.

Public (NYSE: GEV); Q1 2026 backlog $163B

Constellation Energy

Nuclear baseload for hyperscalers

Restarting Three Mile Island for Microsoft reframed dormant nuclear assets as premium, 24/7 carbon-free supply for AI.

Public (Nasdaq: CEG); 20-year Microsoft PPA, ~$1.6B restart investment

Bloom Energy

Off-grid power when the queue is too long

Fuel cells let data centers generate on-site and bypass interconnection delays entirely; deep partnerships with Oracle and Brookfield.

Public (NYSE: BE); up to 2.8 GW Oracle procurement, $5B Brookfield partnership

GridCARE

Software that conjures grid capacity

Uses AI to locate underutilized headroom on the existing grid, compressing data-center interconnection from years to months.

Private; $64M Series A (May 2026)

ThinkLabs AI

AI-native grid simulation

Nvidia-backed; trains models per circuit in minutes and processes a year of hourly power-flow across 100+ circuits in minutes.

Private; $28M raise, backers incl. Nvidia

Base Power

Distributed batteries as a virtual power plant

Building behind-the-meter storage at residential scale to flex against grid strain and AI-driven peaks.

Private; raised over $1B across recent rounds

05 · Signals

What moved, and what to watch

Jun 2025

Talen-Amazon ~$18B nuclear PPA

Talen agrees to sell Amazon up to 1,920 MW from its Susquehanna nuclear plant through 2042, cementing the colocation model for the data-center era.

Oct 2025

Bloom-Brookfield $5B fuel-cell pact

Brookfield commits $5B to deploy Bloom fuel cells across AI data centers globally, validating on-site generation as a grid-bypass strategy.

Dec 2025

PJM capacity auction hits record

Capacity costs reach record highs as the grid falls short on supply; the market monitor names data centers the primary driver of price increases.

Jan 2026

Meta's 6.6 GW nuclear push

Meta signs deals with Vistra, Oklo and TerraPower for up to 6.6 GW by 2035 to power its Prometheus AI supercluster in Ohio.

May 2026

GridCARE's $64M Series A

An oversubscribed round funds a new 'power acceleration' category — AI that finds stranded grid capacity for AI infrastructure.

06 · The exposure read

Who’s defensible, who’s at risk

AI rewards clean, structured advantage and punishes friction. The line runs through who owns the data, the brand and the customer — and who is merely a step the technology can route around.

Defensible

  • Independent power producers with nuclear and gas fleets — Constellation, Vistra and Talen are monetizing baseload assets at premium 20-year terms hyperscalers will sign because uptime matters more than price.
  • The turbine-and-transformer oligopoly — GE Vernova, Siemens Energy and Mitsubishi control roughly two-thirds of gas-turbine supply and enjoy multi-year backlogs and pricing power while demand outruns capacity.
  • Interconnection and grid-intelligence software — GridCARE and ThinkLabs turn the grid's worst bottleneck, the connection queue, into a venture-scale software category with structural pull from hyperscalers.
  • Behind-the-meter and fuel-cell players — Bloom Energy and Base Power win when the grid simply can't connect load fast enough, selling speed-to-power rather than cheapest electrons.

At risk

  • Ratepayers and the utilities serving them — PJM customers face about $9.3B in higher costs and double-digit monthly bill increases in some zones, inviting regulatory intervention that could constrain who gets to build.
  • Legacy coal and laggard utilities — exposed to cost-allocation fights and carbon scrutiny while lacking the clean-baseload or grid-software assets hyperscalers actually want to buy.
  • Pure capex bulls ignoring the queue — multi-year turbine lead times and transformer shortages mean money committed today may not become megawatts for years; physical constraints, not demand, now gate the buildout.
  • Speculative AI-grid software with no utility traction — the ~$22B software market is real but crowded, and incumbents are absorbing the most credible startups, leaving thin air for the rest.
The power sector's AI moment is bimodal: a physical supply crunch that rewards whoever owns scarce generation and grid hardware, layered over a slower software transformation that will decide who operates the grid intelligently for the next two decades. The near-term winners sell electrons and turbines; the durable winners will sell the operating system for an electricity system that AI made both larger and more volatile. Watch the regulators — the ratepayer bill is the variable that can throttle the entire buildout.

Sources

Where this comes from

IEA — Energy demand from AI (Energy and AI report)  ·  IEA — Data center energy consumption set to double by 2030 to 945TWh (DCD)  ·  Fortune Business Insights — AI in Power Utilities Market  ·  NextEra's AI bet: data centers ~43% of US demand growth — Latitude Media  ·  GE Vernova Q1 2026 orders & backlog — GE Vernova News  ·  GE Vernova gas turbine backlog into 2029 — Utility Dive  ·  Microsoft–Constellation Three Mile Island restart (~$1.6B) — Utility Dive  ·  Meta's 6.6 GW nuclear deals (Vistra, Oklo, TerraPower) — Utility Dive  ·  Talen–Amazon ~1,920 MW nuclear PPA — POWER Magazine  ·  Bloom Energy–Oracle up to 2.8 GW fuel-cell deal — investor release  ·  Bloom Energy–Brookfield $5B partnership — Bloom investor release  ·  GridCARE $64M Series A — BusinessWire  ·  DeepMind — Machine learning can boost the value of wind energy (20%)  ·  PJM capacity prices & ratepayer costs — IEEFA