AI in Energy: How Intelligent Infrastructure Is Reshaping the Next Growth Cycle
Updated: 4 days ago
Executive summary. AI in energy has moved from pilot projects to grid-critical infrastructure. Electricity demand from AI data centres is rising faster than new grid capacity can be built, so utilities, developers and large energy users are deploying AI to extract more throughput, reliability and flexibility from the assets they already own. Strategic Revenue Insights estimates the global AI in energy market at USD 9.20 billion in 2025, growing to USD 90.60 billion by 2035 at a 25.70 percent CAGR. This article sets out five lessons from the companies moving first, and what each one means for strategy, data, technology, talent and revenue.
For most of the past decade, artificial intelligence in energy was a pilot-project story: a forecasting model here, a predictive-maintenance trial there, usually funded from an innovation budget and rarely connected to the core of the business. That phase is over.
In 2026 the energy system faces a problem it has not seen in a generation. Electricity demand is rising faster than grids can be built, and the fastest-growing source of that demand, AI computing, arrives in blocks measured in hundreds of megawatts. Transformers, switchgear and transmission lines take years to deliver. Interconnection queues in several US regions have grown beyond what conventional planning processes can absorb. In this environment, AI has become one of the few levers that can expand usable capacity without waiting years for new steel in the ground.
The result is a market that is moving from optional technology to operational necessity. For utilities, developers, equipment makers and investors, the question is no longer whether AI in energy matters, but where value is concentrating, which capabilities decide who captures it, and how quickly competitors are moving.
Why the AI in Energy Market Is Entering a New Phase
Three structural forces are converging at once, and together they explain why this market is compounding rather than simply growing.
Load growth has become a step change. The International Energy Agency's April 2026 analysis estimates that data centres consumed about 485 TWh of electricity in 2025 and could reach roughly 950 TWh by 2030, close to three percent of global electricity demand. AI-focused facilities are growing far faster than the data-centre fleet as a whole; the IEA reports that their consumption rose by about 50 percent in 2025 alone, against roughly three percent growth in total electricity demand. In the United States, Deloitte's 2026 outlook projects that AI data centres could drive power demand to 123 GW by 2035, from about 4 GW in 2024.
The physical grid is the binding constraint. The IEA identifies grid bottlenecks and shortages of electrical equipment as the main brake on the near-term build-out. The scale of the queue problem is visible in Texas, where ERCOT's large-load interconnection queue expanded from 63 GW to 226 GW within a single year. In PJM, the 2027/2028 capacity auction cleared about 6.6 GW short of the reliability requirement, prompting FERC to accept an expedited interconnection track in June 2026. When hardware is the constraint, software that squeezes more throughput from existing assets becomes disproportionately valuable.
The technology has crossed the credibility threshold. Machine learning for demand forecasting and asset health is now in production at most large investor-owned utilities. Control-room applications are moving from advisory dashboards to systems that recommend, and in some cases execute, switching, voltage and dispatch decisions. Deloitte projects that close to 40 percent of utility control rooms will use AI by 2027. Regulated industries do not scale technology until it has proven itself inside their risk frameworks; AI in energy has now done so.
These forces explain why Strategic Revenue Insights sizes the global AI in energy market at USD 9.20 billion in 2025, rising to USD 90.60 billion by 2035 at a compound annual growth rate of 25.70 percent across the 2026 to 2035 forecast period. The growth is pulled by a system under stress, not pushed by technology hype.
Five Lessons From the Companies Moving First
The organisations that are converting AI from experiment to infrastructure share a recognisable pattern. We have distilled it into five lessons. Each is followed by a takeaway for a specific leadership function, because the market shift is not only technological; it changes strategy, data, operating models, talent and the revenue model itself.
Lesson 1: Reliability, Not Decarbonisation, Now Sets the Agenda
For a decade the energy transition was framed around emissions. In 2026 the framing has shifted to keeping the lights on. A September 2026 survey of 134 US utility innovation leaders found that 78 percent already use AI to manage rising demand, and 74 percent say AI-driven data-centre load is affecting grid reliability. The same survey reported that grid reliability has overtaken net zero as the industry's top concern.
This matters commercially because reliability is the one outcome that regulators, boards and customers all reward without debate. Budgets for grid management and optimisation software are easier to defend in rate cases when the business case is "fewer outages and more deliverable capacity" rather than "innovation". It is no coincidence that SRI's research identifies grid management and optimisation as the dominant application segment of the AI in energy market.
The lesson for leaders is to anchor every AI initiative to a reliability or capacity outcome that can be measured in megawatts, minutes of outage avoided, or dollars of congestion cost removed. Projects framed that way get funded, scaled and defended.
"The utilities scaling AI fastest are not the ones with the best models. They are the ones that tied every model to a reliability number the board already cares about."
Takeaway on strategy. Reframe the AI portfolio around the outcomes the system is short of: deliverable capacity, reliability and speed of interconnection. Rank initiatives by their contribution to those metrics, not by technical novelty. If a project cannot be expressed in megawatts or minutes, it belongs in research, not in the capital plan.
Lesson 2: The Fastest Gigawatt Is the One You Already Own
The most striking shift in the past two years is that AI is being used to create capacity, not just to save cost. Dynamic line rating is the clearest example. Instead of operating transmission lines at conservative static limits, operators use weather, sensor and load data, increasingly processed by machine-learning models, to calculate how much power a line can safely carry in real time.
The measured results are material. PPL Electric reports that lines equipped with dynamic rating showed more than 16 percent additional capacity compared with the static ratings used previously, and the utility attributes about USD 23 million in annual customer savings to its initial three-line deployment. Idaho National Laboratory case studies summarised in 2026 point to peak-capacity gains of around 10 percent on monitored lines. Austrian transmission operator APG has deployed the technology across roughly 15 percent of its network, and in April 2026 National Grid in the United Kingdom signed a five-year contract to extend dynamic rating across a further 585 kilometres, with most installations due by 2028.
Predictive maintenance follows the same logic. Condition monitoring and failure prediction allow operators to run transformers and lines closer to their true limits with confidence, and to schedule interventions before equipment fails. When a new substation takes five years, extending the utilisation and life of an installed asset is a revenue lever, not a maintenance line item.
Takeaway on data. Capacity creation depends on a unified, trusted data foundation: real-time sensor feeds, weather, asset condition history and market data in one place with clear ownership. Most utilities still hold these in separate operational and enterprise systems. Unifying them is the unglamorous prerequisite for every capacity gain described above, and it is where the first eighteen months of most programmes are spent.
Lesson 3: Large Loads Are Becoming Grid Participants
The second-order effect of AI demand is that the largest new customers are no longer passive consumers of electricity. Hyperscalers are negotiating flexibility in exchange for speed. In March 2026 Google announced that it had integrated one gigawatt of demand-response capability into long-term contracts with five US utilities, including Indiana Michigan Power, Tennessee Valley Authority, Entergy Arkansas, Minnesota Power and DTE Energy. Under these agreements Google curtails machine-learning workloads when the grid is stressed, in return for faster interconnection and less need for new transmission and generation.
Policy is reinforcing the trend. In March 2026 seven of the largest AI companies and hyperscalers signed a White House-brokered commitment to build, procure or fund new generation for their data centres and to pay for the grid upgrades needed to connect them. PJM is developing rules under which data-centre load that arrives without associated new supply may be curtailed ahead of standard demand response. Research from Duke University's Nicholas Institute in 2026 found that even a one to two percent reduction in data-centre peak demand could lower electricity rates by 0.5 to 2.8 percent while improving reliability.
Every one of these arrangements runs on software: forecasting when the grid will be stressed, deciding which workloads can shift, verifying delivery, and settling the commercial terms. That is a new demand pool for AI in energy that sits outside the traditional utility channel, and SRI's research already shows the fastest change in end-user mix coming from renewable developers and large industrial consumers rather than from utilities alone.
"The data centre is turning into a dispatchable resource. The companies that build the forecasting and orchestration layer for that resource will own a market that did not exist three years ago."
Takeaway on technology and operating models. Design AI systems for a grid with controllable loads, not only controllable generation. Forecasting, orchestration and settlement for flexible large loads should be treated as a product line, with its own customers, pricing and service levels. Utilities that keep control of the load-side breaker while offering faster interconnection will set the terms of this market; technology providers that make that trade-off operable will be their partners.
Lesson 4: Capital Is Flowing to the Grid, and Software Rides With It
US utilities have signalled grid investment plans in the region of USD 1.4 trillion over the next five years. Corporate venture arms are directing money toward intelligence rather than only toward steel. National Grid Partners, for example, has backed LineVision, whose grid-intelligence platform helps utilities expand capacity without new infrastructure, and Terragrit, a platform for simulating changes to physical operations before they are made.
The same September 2026 survey, however, reveals the constraint on this capital. Fifty-nine percent of utility innovation spending still goes to incremental improvements and only 16 percent to transformational initiatives, and 87 percent of leaders say returns on innovation investment are limited by regulatory and rate-case frameworks designed before the AI demand surge. In other words, the money exists, but the rules for recovering it are lagging.
For technology providers this creates a clear playbook. Products that attach to approved capital programmes, that can be expensed within existing cost-recovery mechanisms, and that produce auditable reliability outcomes will scale. Products that require a new regulatory category will wait.
Takeaway on talent, adoption and change management. The scarce skill is not data science; it is the combination of power-systems engineering, regulatory fluency and AI literacy in the same team. Utilities should build small cross-functional cells that own one operational outcome end to end, and measure adoption by how many decisions in the control room are informed by the models, not by how many models exist. Vendors should staff accounts with people who can read a rate case.
Lesson 5: Trust Inside the Control Room Is the Real Competitive Moat
The competitive map of AI in energy is being drawn at the operational layer. Incumbent equipment and control-system suppliers such as Siemens Energy, GE Vernova, Schneider Electric, ABB and Hitachi Energy hold a structural advantage: they already own the control systems, the protection schemes and the customer relationships, and they are embedding AI into those platforms. Their vulnerability is pace. Hyperscalers including Microsoft, Alphabet and Amazon Web Services are entering from the other direction, bringing cloud-scale AI platforms and, increasingly, their own energy procurement problems to solve. Specialist software companies occupy the space between them.
What decides the outcome is trust. A grid operator will not hand switching or dispatch decisions to a system it cannot audit, and a regulator will not approve cost recovery for a tool whose outcomes cannot be verified. The capabilities that build that trust are specific:
Integration with legacy operational technology, where most of the implementation risk lives.
Demonstrated reliability outcomes, measured in avoided outages and recovered capacity rather than in model accuracy.
Regulatory fluency, because utilities buy what they can recover in rate cases.
Partnerships that combine domain depth with AI scale.
Margins are likely to compress for standalone analytics and expand for platforms that can prove capacity gains inside live operations. The strategic question is shifting from "which AI model is best" to "which provider can be trusted inside the control room".
Companies evaluating these competitive dynamics in greater detail can use SRI's AI in Energy Market research to assess segment sizing, regional growth and the positioning of the leading providers across software, hardware and services.
Where the Opportunity Is Concentrating
Value is moving toward the intersection of software and the physical grid rather than toward generic analytics.
By component, software is the largest segment of the AI in energy market, and the most defensible positions are in platforms that sit inside operational workflows: grid orchestration, distributed energy resource management, outage prediction, dynamic rating and energy trading. Hardware demand follows, particularly sensors and edge computing that make the software useful at substation and line level.
By application, grid management and optimisation dominates today, while renewable energy integration is the fastest-growing application. Variable generation needs minute-by-minute forecasting, curtailment management and storage dispatch that human operators cannot perform at scale, and policy frameworks such as the EU Green Deal and US emissions-reduction targets keep that demand growing.
By end user, utilities dominate, but the fastest change is among renewable developers, large industrial consumers and data-centre operators who are becoming energy market participants in their own right.
By geography, North America leads on smart-grid spending and renewable integration, while Asia Pacific is the fastest-growing region as governments accelerate investment in energy management and transmission modernisation. Europe's opportunity is shaped by regulation that rewards flexibility and cross-border interconnection.
At the Horizon: What to Watch Over the Next Three Years
Three indicators will show whether the market is tracking toward the upper or lower end of its growth range.
The share of control rooms running AI in live operations. Industry outlooks suggest close to 40 percent by 2027. That is the threshold at which AI becomes standard practice rather than innovation, and at which procurement moves from pilots to enterprise agreements.
Regulatory reform of cost recovery. Most innovation returns are limited by rate frameworks designed before the AI demand surge. The jurisdictions that modernise cost recovery first will see the fastest software adoption, and vendors should weight their go-to-market investment accordingly.
The behaviour of large loads. If data-centre operators move decisively toward flexible interconnection and behind-the-meter generation, demand for AI-driven trading, forecasting and flexibility platforms will grow outside the traditional utility channel. Technical validation is still in progress; a 96 MW facility in Manassas, Virginia, developed by EPRI, NVIDIA, Emerald AI and partners, is intended to test workload flexibility at scale in late 2026.
"The next phase of competition will not be about who has the smartest algorithm. It will be about who can prove, in a regulator's language, that the algorithm delivered capacity."
The Strategic Revenue Opportunity
For technology providers, the immediate task is to identify which applications and regions are converting from pilot to procurement, and to align product roadmaps with where capital is actually being approved. For utilities and developers, the priority is AI investment that produces measurable capacity and reliability gains, because those are the outcomes that regulators and boards reward. For investors, the opportunity lies in platforms that bridge AI capability with operational trust.
Market intelligence turns these questions into decisions. Segment-level sizing shows where spend is concentrating, competitive benchmarking shows where incumbents are exposed, regional analysis shows which markets are opening first, and customer analysis shows which end users are converting fastest. Used well, it reduces market-entry risk, sharpens pricing and shortens the path from a product roadmap to booked revenue. This is the path from market signals to strategic insight to revenue impact.
Takeaway on revenue. Treat the AI in energy market as four distinct revenue pools with different buyers and sales cycles: utility operations software, renewable integration and trading, flexible-load orchestration for large consumers, and sensing hardware at the edge. Each needs its own evidence base. Enter where the buyer already has an approved budget line and a reliability metric to defend.
The Next Leg: Preparing for the Journey Ahead
The defining shift in the AI in energy market is that intelligence has become infrastructure. Grids cannot be built fast enough to meet AI-driven demand, so AI is being deployed to make existing grids do more. That reframes the market from a technology category into a reliability imperative, and it rewards providers that can operate inside the physical and regulatory constraints of the energy system.
The next three years are the window in which operational positions will be established. Decision-makers should anchor their AI agenda to capacity and reliability outcomes, build the data foundation that makes those outcomes measurable, design for a grid in which large loads are participants, and invest in the regulatory and engineering fluency that turns a model into a trusted operational system.
Companies evaluating growth opportunities in this market can use Strategic Revenue Insights' AI in Energy Market analysis to assess market size, competitive dynamics, regional opportunities and strategic growth priorities.
Related analysis: Sodium-Ion Batteries: How 175 Wh/kg Cells Open a USD 6.70 Billion Market; Methanol Bunkering: Why Ports Are Racing to Fuel a USD 10.97 Billion Market; Anaerobic Digestion: How Biomethane Is Turning Waste Into an Energy Business.
Frequently Asked Questions
How big is the AI in energy market?
Strategic Revenue Insights estimates the global AI in energy market at USD 9.20 billion in 2025, rising to USD 90.60 billion by 2035 at a 25.70 percent compound annual growth rate.
Why is AI in energy growing so fast?
Electricity demand from AI data centres is rising faster than new grid capacity can be built. Utilities, developers and large energy users are deploying AI to get more throughput, reliability and flexibility from the assets they already own.
Has AI in energy moved beyond pilot projects?
Yes. For most of the past decade AI in energy meant isolated forecasting or predictive maintenance pilots funded from innovation budgets. It has now become grid-critical infrastructure connected to the core of the energy business.
What should energy companies prioritise?
The companies moving first treat AI as an operating capability rather than an experiment, with clear priorities across strategy, data, technology, talent and revenue, and focus it on extracting more value from existing grid and generation assets.
About the authors: This article was prepared by the Strategic Revenue Insights research team, which combines market intelligence, competitive analysis and strategic advisory for growth-stage, mid-market and enterprise organisations across the energy and power sector. Sector analyst: Rohan Mehta, Principal Strategist, Finance, Chemicals and Market Risk. Our market estimates combine primary research (interviews, surveys and expert panels) with secondary data (industry reports, databases and public disclosures); read more about how Strategic Revenue Insights works. To discuss this market with our analysts, contact our team.
Comments