Energy: the power grid fell short, and the backlash began

Energy is the industry where AI's demand for electricity has outrun the supply, and the fight over who pays has now begun. In the six weeks since we last looked, the largest US power market failed to secure enough electricity for a third year running, federal regulators ordered a rewrite of how data centres connect, and a US state began taxing the power they draw. The scarce thing is no longer cheap electricity. It is electricity you can actually get.

The supply actually fell short

When Era Haus last looked at this industry in AI became the biggest new buyer of electricity, the story was in the price: a record capacity auction and rising household bills. A capacity auction is the yearly market where a grid operator pays power plants to promise they will run when demand peaks. The newer development is sharper. The power got dearer, and now some of it is not showing up at all.

On 14 July 2026 PJM, the operator that runs the largest US power market, held its latest capacity auction, for the year from mid-2028 to mid-2029. It cleared at the regulator's price ceiling and still fell about 6,800 megawatts short of the safety cushion the grid is meant to hold for reliability, the third year running it has come up short (PJM, July 2026). Almost no new power plants turned up to close the gap. An independent monitor of the auction put roughly $6.3 billion of the cost, about 38%, down to data centres (Utility Dive, July 2026).

Firm power, the kind you can switch on at the peak, is now worth more than it has been in a working lifetime, because the system is short of it.

The rules started to change

Two moves in a month show the reckoning arriving. On 20 June 2026 the US Federal Energy Regulatory Commission, the body that oversees the interstate grid, ordered the six largest regional grid operators to justify or rewrite, within 60 days, the rules for how large new users such as data centres connect. The goal it named was "speed to power" while shielding ordinary households from the cost.

Ten days later a US state put a price on the load directly. Virginia, the largest data-centre market in the world, signed a budget on 30 June 2026 taxing all the electricity its data centres consume at just over one US cent per kilowatt-hour, in force the next day and expected to raise around $600 million a year (Williams Mullen, June 2026). It is the first US tax of its kind, and it applies even to power a data centre generates for itself, closing the obvious way around it.

The backlash is local too

The pressure comes from the ground up as well. A tracker run by the research firm 10a Labs counted more than 75 data-centre projects worth around $130 billion blocked or delayed across the US in the first three months of 2026, the most since it began counting, with organised opposition now active in nearly every state (Fortune, June 2026). A summer heatwave has hardened the mood: the case for a power-hungry data centre nearby is harder to make when the air conditioning is already straining the grid (Al Jazeera, July 2026).

Outside the US, the wall is harder

The limit is physical everywhere, and often blunter beyond the US. In Britain, the data-centre company Nscale was told this year that grid power would not arrive in time to open its planned Essex site, and is now looking at making its own electricity on-site with gas fuel cells rather than waiting for a connection. In Brazil, a large campus planned for Rio de Janeiro, billed as Rio AI City, is being built to run on its own renewable power and waterless cooling, designed around the local grid rather than leaning on it. Across markets the rule is the same: firm, connected power sets the terms, and the buildout bends toward wherever it already exists.

What it means for you

For an energy operator at small or mid scale, the change since spring is that the constraint moved from price to sheer availability. Value has shifted to whoever can supply firm, already-connected power. If you own generation, a plant that runs on demand is worth more each month this shortage lasts. If you hold a place in the interconnection queue, the waiting list to plug a new project into the grid, that place is now an asset other developers would pay for. If you develop projects, on-site and behind-the-meter generation, making power where it is used, now has real demand behind it.

The risk sits beside the opportunity. The demand raising your asset's value is also raising household bills, and the backlash lands on whoever is visible. The rules are being rewritten as you read this, so terms you lock today may sit inside a different cost structure within a year.

What to do about it

Two grounded moves, and one thing to watch. First, if you hold generation or a grid connection, get it valued against today's market rather than last year's. Firm power and a queue position have jumped in worth within weeks, and a contract on old assumptions is priced too low. Second, follow your own regulator on how large users connect and who pays for the load they add. The US decisions this summer are the template other regions are watching; elsewhere, watch whether your grid makes the big user pay or spreads the cost across households.

What to watch rather than bet on is the largest demand forecasts. The $130 billion of projects already stalled this year shows that announced demand and delivered demand are not the same thing. Sign for the load you can see and connect.

The pattern underneath

Energy still runs the opposite way to the rest of this series. In law, property and accounting, AI made a professional skill cheap, and value moved to the judgment and trust that do not copy. In energy, AI makes a physical input, electricity, scarce, and value moves to the asset itself, the point we made in The model wasn't the moat. What has changed since June is which part of that asset is scarcest. A few months ago the scarce thing was cheap power; now it is power you can actually connect. The operator who locks in supply and connection terms early is building the advantage that lasts. The one still planning around cheap, available electricity is planning for a market that has already gone.