The Texas Grid, the Texas Water, and the Demand-Side Question

We are a Texas company, and the version of this debate happening in our own state is the most extreme in the country. ERCOT is the most-watched grid in America, and the numbers coming out of it in 2026 are difficult to absorb.

The queue is larger than the grid

ERCOT's all-time peak demand is 85 GW, set in August 2023. As of late 2025, its large-load interconnection queue — requests to connect new industrial-scale electricity consumers — stood at 226 GW, roughly quadruple the 63 GW recorded a year earlier, with about 73–77% of it data centers.

Then it accelerated again. In 2026, ERCOT reported the large-load queue reaching 410 GW — a jump of nearly 150 GW in two weeks — with approximately 87% of it data centers, against about 453 GW of total interconnection requests. By mid-2026 the operator was managing roughly 445 GW of prospective new load through 2033.

ERCOT's president framed the volatility plainly at the agency's Innovation Summit: it gives you an idea of how quickly things can change. The operator's own long-term forecast filed with the Public Utility Commission shows potential peak demand climbing toward roughly 278 GW by 2029 and 368 GW by 2032 when medium and large loads are included — more than four times the current all-time peak.

Not every queued project gets built; interconnection queues everywhere contain speculative requests, and ERCOT itself notes that while not all requests result in built projects, its data shows the majority expect to be operational by 2030. But the direction is unambiguous, and the supply side is not keeping pace: roughly 23 GW of new generation was added between 2024 and 2025, with another 9 GW slated for early 2026.

The state's response

Texas has moved. Senate Bill 6 requires standardized large-load interconnection rules, improvements to load forecasting, and requirements around data center load curtailment. ERCOT has approved an initial batch-based interconnection framework — studying projects 75 MW and larger together rather than one at a time — with provisions allowing large loads to bring their own on-site generation, and a path for customers who agree to be curtailed during local constraints. Officials were direct about why: the process built to review 40–50 large loads at a time had been outgrown.

Water, in a state that watches water

The water projection for Texas is the figure that stops conversations. A Houston Advanced Research Center and University of Houston study projects Texas data centers consuming 49 billion gallons in 2025, rising to as much as 399 billion gallons by 2030 — a drawdown the study likens to lowering Lake Mead by more than sixteen feet in a single year. Nationally, roughly 1.2 gallons of water is consumed per kilowatt-hour generated, so the electricity figures above carry a water figure inside them.

As we note in our national footprint review, total data center water use remains small beside agriculture. The issue is localization and rate of change — concentrated draw arriving faster than municipal planning cycles, in a state that has spent a century arguing about water.

Why a Texas company works on the demand side

Every number above describes supply-side strain: more generation, more transmission, more interconnection studies, more water. The demand-side question is the one we can actually work on from Flower Mound: how much of this projected interactive AI load never needed a data center in the first place?

The routing literature suggests most everyday queries do not require frontier models. Consumer hardware across Texas — in law offices, clinics, school districts, and small businesses — already draws power for other reasons and sits mostly idle on its AI capability. If a meaningful share of routine inference ran there instead, the avoided load would show up in exactly the queue described above, and the avoided water would never be withdrawn.

We will not put a number on that until we can measure it honestly. That measurement is the first thrust of AEOLI, and it is being built here.