Energy: AI became the biggest new buyer of electricity

Energy is the industry AI is reshaping from the demand side: data centers running AI now compete with homes and factories for electricity, and they are winning. For a power producer, grid operator, or energy retailer, that demand has turned electricity from a slow, predictable business into the scarcest input in the AI economy. The result is higher wholesale prices, multi-year waits to connect new supply to the grid, and a premium on anyone who can deliver firm power on a schedule the AI buildout can count on.

Where AI actually stands in energy today

This is the first time Era Haus has looked at this industry, so start with the honest baseline. AI shows up in energy two very different ways.

The first is what most people mean by "AI in energy," and it is the smaller story: software inside operations. Utilities and producers use machine learning to forecast demand and renewable output, to spot failing equipment before it breaks, and to schedule crews and trades. Independent forecasters such as Amperon sell AI price forecasts to traders, and large operators such as NextEra run predictive-maintenance models that flag a turbine likely to fail. These tools make the work more efficient; the business itself stays the same.

The second matters more, and runs the other way: AI is the biggest new buyer of electricity in a generation. Data centers training and running AI models are adding load to grids that stayed flat for twenty years, reshaping prices, generation and regulation at once.

The demand shock is already in the price

Start with the clearest signal, because it is in the numbers. In the US, the regional grid operator PJM runs the power market for 67 million people across thirteen states, and each year it auctions "capacity": commitments to have power on hand when demand peaks. Its December 2025 auction cleared at the regulator's price ceiling for a record total of 16.4 billion dollars, and PJM's own market monitor attributed roughly 40 percent of that cost to data center demand (Utility Dive, January 2026). For households in that US region, PJM expects bill rises of about 1.5 to 5 percent for the twelve months beginning June 2026.

For an operator the meaning is plain: the value of firm, dispatchable power, the kind you can switch on when the grid runs short, has jumped.

The generation rush, and where a smaller operator fits

That price signal is pulling a wave of supply deals: through 2026, large technology firms have signed long-term contracts with power producers to lock up electricity. Meta alone signed deals for up to 6.6 gigawatts of nuclear power in January 2026, enough for several million homes (CNBC, January 2026). US producer Vistra expanded its fleet in June 2026 and moved to restart a retired nuclear plant, explicitly to serve AI demand (company announcements, June 2026).

These headlines are about giants, but the opening reaches further down. When one buyer locks up a multi-gigawatt site, everyone else competes harder for what is left, and any reliable megawatt becomes more valuable. A mid-size independent producer, a developer with a battery-storage project, or an operator holding an existing grid connection now owns something the market wants. On-site generation, making your own power instead of waiting years for a grid hookup, has moved from niche to mainstream for the same reason.

Regulators are deciding who pays

The third front is policy, moving fast. As AI demand lifts bills, regulators are rewriting who pays for the grid. By May 2026, twenty-three US states had approved "large-load tariffs," rules that make very large customers such as data centers cover their own connection and capacity costs instead of spreading them across every household. More than 300 data-center bills were filed in US statehouses in early 2026, and Pennsylvania's regulator advanced ratepayer protections in May. These rulings are the rulebook for the next decade of demand, and they are being written now.

The shock is global, with local limits

The same demand is hitting every region, with different limits. In Europe the bottleneck is the grid itself: a new data center waits seven to thirteen years for a connection in the busiest markets, and industrial power costs far more than in the US, around 111 dollars per megawatt-hour in the UK against 28 in May 2026 (CNBC, May 2026). That gap is steering some AI investment toward cheaper electricity elsewhere.

Latin America is competing for it. Argentina and OpenAI announced Stargate Argentina in late 2025, a 25-billion-dollar data center in Patagonia to be built on clean power with local firm Sur Energy, and comparable hubs are taking shape in Mexico and Brazil. For an operator anywhere with spare, affordable, low-carbon power, AI demand is now a real buyer.

What it means for you

For an energy operator at small or mid scale, the core change is blunt: what you generate or move became scarce and valuable in a way it has not been in your working life. If you generate power, a signed long-term contract with a creditworthy buyer is now an asset others will pay for. If you develop projects, storage and on-site generation have real demand behind them. If you trade or retail, rising volatility rewards good forecasting and punishes a loose hedge. If you run a utility, you are negotiating with the most demanding customer class in your history while regulators watch who absorbs the cost.

The risk runs alongside it: the demand lifting your asset's value also lifts your costs and your customers' bills, and the backlash lands on whoever is visible. The operator who locks in supply and connection terms early gains; the one still planning around cheap, flat power is preparing for a market that has already left.

What to do about it

Three grounded moves. First, if you hold generation or a grid connection, get it valued against today's market; the worth of firm power and of a place in the interconnection queue has moved sharply in a year. Second, treat AI forecasting and maintenance tools as the efficiency layer they are, worth piloting on one site and measuring before you scale. Third, follow your regulator's large-load and cost-allocation rulings closely, because the 2026 decisions on who pays for the grid will set your margins for years.

Watch, but do not bet the company, on the most aggressive demand forecasts. AI power demand is real and large, yet some announced data centers will never be built. Sign for the demand you can see, not the demand in a press release.

The pattern underneath

Energy breaks the pattern the rest of this series has traced. In law, real estate, marketing and accounting, AI made a professional skill cheap and value moved to the judgment and trust that do not copy, the case Era Haus made in Defensibility in the AI era. Energy runs the opposite way: here AI makes a physical input, electricity, scarce and valuable, and the moat becomes the asset itself, a power plant, a grid connection, a signed contract. We argued in The model wasn't the moat that the AI model would slide toward commodity and value would settle elsewhere. In energy, it settles on the megawatt.