The Invisible Trillion · The Energy Question, Honestly
But Globally Hides Locally
The “1 to 1.5% globally” number is true. It is also, on its own, misleading, in a way that the rest of the chapter will spend some time correcting.
Here is the problem with global percentages. The Earth is large. A small percentage of a very large total is still an enormous amount, but more importantly, the small percentage is unevenly distributed. A handful of regions host most of the world’s data centers, and in those regions, the share of electricity going to data centers is not 1.5%. It is closer to 20%, 30%, or 40%, depending on the specific region and how far the buildout has gotten.
This matters for two reasons. First, because the people living in those regions experience the buildout’s effects regardless of what the global average is. Second, because the financial cost of expanding the grid to serve the data centers is mostly paid by the residents and businesses in the regions where the buildout is happening, not spread across the global population whose AI queries the data centers are serving.
Let me show you what this looks like in numbers, using the United States’s largest regional electricity market.
PJM Interconnection is the grid operator covering all or part of thirteen states and the District of Columbia, from Illinois east to New Jersey, south to Virginia and North Carolina. Roughly 65 million Americans get their electricity through PJM’s grid. PJM runs an annual capacity auction in which it pays generators to be available for future demand. The prices set in those auctions feed directly into residential electricity bills across the entire region.
For the 2025–2026 delivery year, PJM’s capacity prices were the highest in the system’s history. PJM’s own Independent Market Monitor — the office whose statutory job is to evaluate the market for the grid operator — concluded that data center demand was the primary cause of the price increase. The Monitor’s analysis attributed $9.3 billion in additional capacity costs to data center load, an increase of 174% over what prices would have been without that load. This is not an activist claim. It is the grid operator’s own analyst, applying the grid operator’s own methodology, saying clearly that residential and commercial electricity customers across thirteen states are paying billions of additional dollars because of data centers.
Retail electricity prices in PJM and adjacent markets have reflected this. Between March 2021 and March 2026, average residential electricity prices rose:
- 94% in Washington, D.C.
- 74% in Maryland
- 73% in Maine
- 58% in New York
- Substantial increases in essentially every state across PJM, ISO-NE, and NYISO
Nationally, electricity prices rose nearly 40% over five years — the fastest sustained increase in American history. Utilities requested $31 billion in rate increases in 2025 alone, double the 2024 figure. About half of the 2025 requests had not yet been approved by early 2026, which means the largest wave of rate increases is still working its way through the system.
Not all of this is data centers. Some is general inflation. Some is the cost of grid maintenance that was deferred for decades. Some is the transition away from older generation. But a substantial portion — by the grid operator’s own analyst, the primary portion of the recent capacity-cost increase — is data center load.
The structural feature of this that I most want you to understand is the who pays part. The grid is a shared system. When a new very large customer connects to it — say, a 500-megawatt data center — the cost of the new substation, the new transmission lines, the new generation capacity all get distributed across every customer in the region, according to formulas set by utility regulators. The data center pays its share. Its share is calculated as if it were a customer rather than as the cause of the new infrastructure. The rest of the customers — meaning ordinary households and small businesses, mostly — pay the difference.
The principle that “the customer causing the cost should pay for the cost” is well established in utility economics. It is called cost causation. Cost causation, properly applied, would put most of the grid expansion cost on the data centers requiring the expansion, rather than on residential ratepayers. In practice, this has mostly not happened, because the data centers and their parent companies have substantial legal and political resources to participate in rate-case proceedings, while residential ratepayers do not.
Some states are starting to push back. Virginia’s Senate Bill 253, introduced in 2026, would direct the state’s utility commission to shift more capacity costs onto data centers specifically. The analysis projected residential rates would drop about 3.4% and data center rates would rise about 15.8%. By mid-2026, at least eighteen states had introduced bills creating special rate classes for very large energy users, with the explicit goal of preventing the cost shifts described above. The fact that these bills are being introduced at all tells you that the current rate structure is widely understood to be unfair. The fact that they have mostly not passed tells you who has the political resources to fight them.
So: globally, data centers are 1–1.5% of electricity. Regionally, in the areas hosting the buildout, the share is much higher, the prices are much higher, and the costs are falling on people who mostly do not use what the data centers produce. The global number is real. The local experience is also real. Both are true. The local experience is the one most people in PJM are actually living with.
All 19 chapters of The Invisible Trillion · Library · All courses