The Coming War for Electricity

The Coming War for Electricity

AI, data centers, electric vehicles, and industrial reshoring are colliding with an aging grid — and the result could reshape markets, geopolitics, and the global economy.

By Ian Gross | The Big Market Report | Last reviewed: May 2026

Not financial advice. This article is for informational and educational purposes only.

The world spent the last century obsessed with oil. Wars were fought over it. Alliances were built around it. Entire national economies were organized to secure access to it. The price of crude became a proxy for global stability, and the phrase "energy security" was synonymous with one thing: barrels per day.

That era is not over. But a new one has begun — and the commodity at its center is not oil. It is electricity.

The AI revolution is no longer just a software story. It is a power story. Every large language model trained, every inference query answered, every GPU cluster humming in a desert data center is drawing electricity at a scale that was, until very recently, difficult to comprehend. The physical infrastructure required to sustain the digital economy is colliding with a power grid that was never designed to carry this load — and the collision is happening faster than markets, policymakers, or most investors have priced in.

Key Takeaways

• Global data center electricity consumption reached 415 terawatt-hours in 2024 — roughly 1.5% of total world electricity use — and is projected to nearly double by 2030.
• U.S. electricity demand was essentially flat for two decades. It is now expected to surge by the equivalent of adding a new Texas to the grid within a decade.
• Large power transformer lead times have stretched from 7–14 months pre-pandemic to 120+ weeks today. The grid cannot be upgraded on demand.
• Constellation Energy (CEG) restarted Three Mile Island for Microsoft. Meta signed 20-year nuclear PPAs with Vistra, Oklo, and TerraPower for up to 6.6 gigawatts. The tech industry is not waiting for the grid — it is building around it.
• The next geopolitical arms race may not be over oil fields. It may be over power generation capacity.

Why Electricity Demand Is Suddenly Exploding

For most of the 21st century, U.S. electricity demand growth was remarkably flat. Efficiency gains in appliances, lighting, and industrial processes largely offset population and economic growth. Utilities planned for modest, predictable increases. The grid was sized for a world that was getting incrementally more efficient at using power.

That world no longer exists.

What is happening now is not a single demand shock. It is five simultaneous demand shocks arriving at the same moment, stacking on top of each other in a way that no utility model anticipated.

Artificial intelligence infrastructure is the most visible driver. Data centers consumed approximately 415 terawatt-hours of electricity globally in 2024, according to the International Energy Agency — roughly the annual electricity consumption of France. The IEA projects that figure will grow at approximately 15% per year through 2030, more than four times the growth rate of total electricity demand. By 2030, data centers alone could consume between 9% and 17% of all U.S. electricity generation, according to EPRI's February 2026 analysis.

Electric vehicle adoption is the second driver. As EV penetration accelerates, the load on residential and commercial distribution networks grows in ways that are geographically concentrated and temporally predictable — meaning utilities must build capacity for peak charging windows that did not exist five years ago.

Industrial reshoring is the third. The CHIPS Act, the Inflation Reduction Act, and a broader bipartisan push to bring semiconductor fabrication, battery manufacturing, and advanced industrial production back to U.S. soil have created a wave of new industrial facilities — each of which is an enormous electricity consumer. A single semiconductor fab can draw 500 megawatts or more.

Crypto mining, though cyclical, adds a fourth layer of demand that is highly sensitive to energy prices and can appear or disappear from regional grids with little warning. And the fifth driver — the broad electrification of heating, cooking, and commercial processes that previously ran on natural gas — is just beginning to show up in utility load forecasts.

None of these trends is temporary. All of them are structural. And they are all arriving simultaneously.

After two decades of essentially flat electricity demand growth in the United States, AI and data center buildout is projected to drive a surge equivalent to adding a new Texas to the grid. Historically, demand shocks of this magnitude have preceded significant repricing of energy infrastructure assets. Source: IEA, EPRI, Deloitte, Goldman Sachs.

The Grid Was Never Built for This

Here is the problem that does not get enough attention: generating more power is only half the challenge. The other half is delivering it.

The United States power grid is, in many respects, a 20th century system being asked to perform 21st century work. The average large power transformer in the U.S. is over 40 years old. The transmission network — the high-voltage lines that carry electricity from generation sources to population centers — was largely built in the 1960s and 1970s. Upgrading it requires not just capital, but permits, environmental reviews, right-of-way negotiations, and regulatory approvals that can take a decade or more to complete.

The transformer shortage alone is a crisis hiding in plain sight. Before the pandemic, lead times for large power transformers ran between 7 and 14 months. By early 2026, lead times for high-voltage 500kV units had stretched to 120 weeks or more — over two years. The U.S. could face a 30% shortfall in power transformers by 2025, according to industry estimates, with distribution transformer deficits compounding the problem at the local level.

Transmission capacity must expand by 50% to 100% over the next decade just to meet projected demand, according to federal analysis. By 2050, it may need to quadruple. The permitting system that governs this expansion was not designed for speed. A new high-voltage transmission line crossing multiple states can take 10 to 15 years from proposal to energization.

The result is a paradox: the U.S. has abundant energy resources — natural gas, nuclear, solar, wind — but an increasingly constrained ability to move that energy from where it is generated to where it is needed. Data center developers are already running into this wall. Dozens of utilities received requests for 700 gigawatts of new power connections in 2025 alone — more than the entire installed generating capacity of the United States.

The modern economy runs on electrons. But the grid still operates like a 20th century system.

The average large power transformer in the U.S. is over 40 years old. Lead times for new high-voltage transformers have stretched from 7–14 months pre-pandemic to 120+ weeks today. The grid cannot be upgraded on demand. Source: Power Magazine, Berlin CWIEME, Electrical Trader.

AI Changes the Equation — Permanently

Artificial intelligence is not just another demand driver. It is a demand driver with a compounding growth rate, a 24/7 load profile, and an insatiable appetite for power that scales with every model generation.

A typical hyperscale AI data center draws 100 megawatts or more — enough to power 100,000 average American households. A single large-scale AI training run for a frontier model can consume tens of millions of kilowatt-hours. And unlike most industrial loads, AI inference — the process of running a trained model to answer queries — runs continuously, around the clock, with no seasonal variation and no off-peak window.

The cooling requirements compound the problem. AI GPU clusters generate enormous heat. Liquid cooling systems, which are becoming the standard for high-density AI compute, require additional power and water infrastructure. The physical footprint of an AI data center is not just a building full of servers — it is a power substation, a cooling plant, and a fiber network, all of which must be co-located and co-built.

NVIDIA (NVDA) sits at the center of this ecosystem as the dominant supplier of AI compute hardware. But the more interesting investment story is in the physical infrastructure that powers and cools those chips. Vertiv Holdings (VRT) reported a $15 billion backlog in early 2026, driven almost entirely by data center power and cooling demand. Its Q4 2025 revenue grew 23% year-over-year. Eaton Corporation (ETN) reported that its data center backlog equals eleven years of 2025 construction levels, with data center orders accelerating 200% in Q4 alone.

Every breakthrough AI model carries a physical energy cost. That cost is being paid in copper, steel, transformers, and kilowatt-hours.

A single hyperscale AI training cluster consumes as much electricity as 100,000 average American households. This comparison makes the scale of AI's energy demand tangible — and illustrates why the grid was not designed for it. Source: IEA, Consumer Reports.

The New Winners

The investment thesis here is not complicated, but it requires looking past the obvious. NVIDIA is already priced for dominance. The more durable opportunity may be in the infrastructure layer — the companies building the physical systems that make AI compute possible.

| Category | Representative Companies | Why They Win | |---|---|---| | Power management & cooling | Vertiv (VRT), Eaton (ETN), Trane Technologies (TT) | Every AI cluster needs power distribution and thermal management | | Grid construction & engineering | Quanta Services (PWR), MYR Group | Transmission expansion requires boots on the ground | | Power generation | GE Vernova (GEV), Vistra (VST) | Gas turbines and dispatchable power fill the gap renewables can't | | Nuclear energy | Constellation Energy (CEG) | 24/7 carbon-free power is exactly what hyperscalers want | | Copper & materials | Freeport-McMoRan (FCX) | Every wire, transformer, and motor requires copper |

The nuclear renaissance deserves special attention. Constellation Energy restarted the Three Mile Island nuclear facility — renamed Crane Clean Energy Center — under a 20-year power purchase agreement with Microsoft. Meta announced in January 2026 that it had signed deals with Vistra, Oklo, and TerraPower to unlock up to 6.6 gigawatts of nuclear capacity. These are not speculative bets. They are 20-year contracts signed by the largest technology companies in the world, betting that nuclear is the only generation source that can provide the scale, reliability, and carbon profile that AI infrastructure demands.

PJM capacity auction prices — the wholesale market for power in the mid-Atlantic and Midwest — jumped from $29 per megawatt per day for the 2025–2026 period to $329 for 2026–2027. That is an 11-fold increase in one year. The market is beginning to price the scarcity.

The important framing: this is not a stock pump thesis. It is a macro observation. The companies listed above are not being recommended — they are being identified as the infrastructure layer of a structural shift. Do your own diligence. Understand the valuations. But understand the trend.

The AI power buildout creates a multi-layer investment ecosystem. The infrastructure layer — power, cooling, grid, and materials — may represent the most durable opportunity as software valuations already reflect significant optimism.

The Geopolitical Dimension

Energy has always been geopolitical. What is changing is which form of energy sits at the center of the competition.

The United States and China are engaged in a technology race that is, at its foundation, an energy race. China has been building data center capacity and power generation infrastructure at a pace that dwarfs most Western projections. It has invested aggressively in domestic semiconductor production, grid modernization, and renewable energy capacity — not primarily for environmental reasons, but for strategic ones. A nation that can power its own AI infrastructure without dependence on foreign energy or foreign technology has a decisive long-term advantage.

The United States has its own advantages: abundant natural gas, a head start in frontier AI model development, and a private sector that is moving faster than any government program. But the permitting bottleneck, the transformer shortage, and the aging transmission network are structural vulnerabilities that cannot be solved with capital alone. They require regulatory reform, supply chain investment, and a national-level recognition that electricity infrastructure is now a defense priority — which is exactly how the Trump administration framed it in a July 2025 executive order.

The Middle East is playing its own angle. Gulf sovereign wealth funds, flush with oil revenue and acutely aware that the hydrocarbon era has a horizon, are deploying capital into AI infrastructure, data centers, and power generation at scale. Saudi Arabia's NEOM project, the UAE's AI ambitions, and Qatar's data center investments are not just economic diversification plays — they are attempts to position petrostates as nodes in the global AI infrastructure network before the oil era ends.

The next geopolitical arms race may not be over oil fields. It may be over power generation capacity, transmission corridors, and the ability to host the physical infrastructure of artificial intelligence.

The Real Bottleneck

Strip away the hype, the earnings calls, and the venture capital narratives, and the AI revolution comes down to a simple physical constraint: you cannot run a data center without electricity, you cannot deliver electricity without infrastructure, and you cannot build infrastructure faster than the supply chain allows.

The bottleneck is not chips. Semiconductor capacity is expanding. The bottleneck is not software. Models are improving rapidly. The bottleneck is not capital. There is no shortage of money chasing this opportunity.

The bottleneck is electricity. Transmission. Cooling. Transformers. Copper wire. Permitting timelines. Utility interconnection queues that stretch years into the future.

The world can print money faster than it can build transformers.

Goldman Sachs projects that data center power consumption will rise 165% from 2023 to 2030. Deloitte estimates that AI data center power demand in the U.S. alone could grow more than thirtyfold by 2035, reaching 123 gigawatts. The infrastructure market required to support this buildout is expected to exceed $1 trillion in annual investment. The IEA estimates that $1.4 trillion in AI data center electrification investment will be needed by 2030.

These are not projections from fringe analysts. They are consensus estimates from the IEA, Goldman Sachs, Deloitte, EPRI, and S&P Global. The debate is not whether the demand surge is real. The debate is whether the infrastructure can be built fast enough to meet it.

What This Means for the Decade Ahead

The electricity bottleneck is not a temporary problem that will be solved by the next quarterly earnings cycle. It is a decade-long structural constraint that will shape capital allocation, corporate strategy, geopolitical competition, and market returns in ways that are only beginning to be understood.

For investors, the implication is that the physical infrastructure layer of the AI economy — power generation, grid equipment, cooling systems, transmission construction, nuclear energy — may represent a more durable and less crowded opportunity than the software and semiconductor layer, where valuations already reflect enormous optimism.

For policymakers, the implication is that permitting reform, grid investment, and supply chain resilience for critical components like transformers are not environmental issues or infrastructure issues in the traditional sense. They are national security issues.

For the broader economy, the implication is that electricity costs are going to rise. The $100 billion in grid upgrades that U.S. utilities are planning will ultimately be paid by ratepayers. The scarcity premium in wholesale power markets will flow through to industrial and commercial electricity bills. The companies and regions that secure reliable, affordable power will have a structural competitive advantage over those that do not.

Markets still underestimate how physical the digital economy has become. The cloud is not weightless. It runs on steel, copper, concrete, and kilowatt-hours. The companies building and powering that physical layer are not supporting characters in the AI story. They may be the most important characters of all.

The next decade may belong not just to the software companies — but to the industries capable of powering them.

Ian Gross is the founder and editor of The Big Market Report. This article is for informational purposes only and does not constitute financial advice. Always conduct your own research before making investment decisions.