Won't Get Fooled Again - Forecasting Texas/Louisiana Data Center Demand for Natural Gas

Tackling that question for the U.S. as a whole is a huge challenge, one fraught with all sorts of data quality and analytical obstacles, and we’ll eventually get there. But fortunately, there is a much smaller problem we need to solve for our energy modeling business, namely, what will happen to power demand in Texas and Louisiana. And, most important for our models, how much additional natural gas will be needed to fuel the generation required to meet that demand.

That’s a data series that we need for the Arrow Model, our proprietary analytical framework built to assess the Texas and Louisiana natural gas markets. The model carves up the region into pipeline “corridors” (aka arrows) that are used to assess changes in the region’s inflows, outflows and flows within each state via groups of pipes that serve similar markets from comparable supply sources.

The most significant model drivers are Permian and Haynesville gas production and LNG feedgas demand, but the model must consider everything that impacts Texas/Louisiana gas, including production from other basins, demand in each subregion (11 of them), pipeline capacity constraints, new infrastructure construction and, of course, in-region demand. And that’s our Arrow Model challenge when it comes to data centers. For the model to accurately assess demand, we must estimate how much demand growth data centers will bring over the next few years.

Therein lies the problem. There are a massive number of announced data center projects, some real and some no more than press releases. Many seem to be “agreements to agree in the future if a deal materializes,” to quote one of our good friends who tracks this stuff. So we had to figure out a process to separate the wheat from the chaff — the genuine projects from the smoke-and-mirrors publicity.

We tackled the problem with a four-step process (see Figure 1 below). Step 1 was to develop a “bottoms-up” database of all publicly acknowledged projects. We compiled publicly available information on data center projects across Texas and Louisiana, sourced from public announcements, government filings and other online resources. Through this research, we generated a list of approximately 50 unique data center projects and 70 unique phases of development — meaning projects that have more than one phase that has been announced.

Figure 1. Gauging Future Power and Gas Demand From Data Centers. Source: RBN

In Step 2, we scored each data center project on a scale from 1 to 3, with each point representing whether the planned data center has an offtaker, site control, or has begun construction. For example, if a planned data center has an offtaker named but does not have (or is not known to have) site control and has not started construction, it would score a 1, but once the project has site control it would score a 2. Once a project has all three, its score of 3 would reflect the highest level of certainty. Projects without an offtaker or site control and are not under construction were excluded from the bottoms-up data forecast.

Step 3 was to convert the announced power capacity into its associated natural gas demand. The math is pretty straightforward, but a lot goes into the assumptions used in the calculations. For example, is this a data center attached to the grid, or does it have dedicated generation? How much gas is used to power the average data center that is connected to the grid? What is the assumed efficiency of the power generation source, measured in terms of heat rate? What is the utilization factor of the data center facility itself — is it really 24x7, 365 days a year? And will some of the power come from solar or wind power or other sources? These are just a few examples of the assumptions we need to plug into the calculations.

Say we assume a data center requires 100 MW of power. That is 2,400 MWh per day (100 MW x 24 hours). If we then assume a 90% utilization factor, that is 2,160 MWh (2,400 x 90%). From this, each project has a specific percentage of its energy fueled by natural gas, depending on its power source and design. For example, we assume that data center projects supplied off the ERCOT grid (serving about 90% of Texas demand) get about 55% of their power from gas. That is higher than the total ERCOT load on average supplied by gas since these units are running around the clock and don’t see the full impact of wind- and solar-sourced power. We also assume that the average gas-fired power from ERCOT is from a unit with an 8,000 Btu/kWh heat rate, which is higher (less efficient) than a modern combined-cycle plant. This would translate our 2,160 MWh demand into 9.2 MMcf/d of natural gas (2,160 MWh x 55% x 8,000 Btu/kWh / 1.037 Btu/Mcf). We divide by the 1.037 cf/Btu factor to convert MMBtu/d to MMcf/d.

Texas Bottoms-Up Assessment

To understand how the numbers really stack up, let’s consider the incremental “bottoms-up” analysis for Texas. This analysis reflects the calculated power demand — and the implied natural gas burn expected to come online in the state over the next five years, based on our scored project database. Existing projects are excluded and the focus here is solely on Texas. We’ll look at Louisiana in a subsequent blog.

Figure 2. Incremental Bottoms-Up Texas Data Center Demand for Power and Gas. Source. RBN

As shown in the left graph in Figure 2 above, by 2030 there are about 4 GW (that is, ~4,000 MW) of projects that scored a 3 on our scale (green layer), another 3 GW with a score of 2 (red layer) and 5 GW with a score of 1 (blue layer), for a total 12 GW. The average ramp-up over the next three years is about 4 GW per year, so this is not happening overnight. And, importantly, 40% of these projects score only a 1, which means there is some level of skepticism as to whether they will really happen.

The right graph in Figure 2 shows the conversion of gigawatts to billions of cubic feet of gas. Given our assumptions, there would be about 1.1 Bcf/d of incremental demand from Texas data centers by 2030 if all of the announced projects scoring a 1, 2 or 3 were built on schedule. Granted, that is a big volume of gas-fired power demand growth. But put in perspective of the typical LNG export facility feedgas demand — 2 Bcf/d at Freeport LNG, for example, and 2.7 Bcf/d at Cheniere Corpus Christi — it is not earth-shattering.

One other factor is worth noting: As we said, we segment Arrow Model demand into 11 distinct regions, so it was necessary to aggregate the numbers by region. A whopping 40% of the data center growth in our bottoms-up analysis comes from Northeast Texas — the region that spans Dallas/Fort Worth (DFW), south to Waco, and east to the Texas/Louisiana border. DFW is definitely the center of the data center universe in Texas. We’ll take a closer look at why this area has such a concentration of data center construction in an upcoming blog.

Texas Top-Down Assessment

Clearly, there are a number of assumptions in the bottoms-up methodology that are debatable, the most obvious being that we only include projects supported by publicly available project-level data. It’s always good to triangulate on these numbers, to make sure we are not missing anything important. That got us to Step 4 in our process.

To do that in Texas, we have a highly authoritative source, the Long-Term Hourly Peak Demand and Energy Forecast from ERCOT’s 2025 System Planning Report, dated April 8 of this year. Unfortunately, in preparing this report, ERCOT had the same problem as RBN — lots of data center demand, the question of what’s real and what’s not, plus the added complexity of a legal requirement that they consider requests from TSPs (transmission service providers) for all new load requirements. Texas’s House Bill 5066 and subsequent ERCOT rules require the grid operator to treat officer-attested TSP letters as “reasonable inputs” for transmission planning. But for this cycle, they received a wave of “large load” requests, dominated by new data centers, many of which lacked firm interconnection agreements. So to keep the numbers reasonable, they created an “Adjusted Large Load Forecast” to temper TSP-provided figures.

How to do that, you might ask? Well … they gave the numbers a haircut. A 50% haircut. But there was some science involved. For data centers that came online in 2022-24, actual peak consumption averaged 49.8% of the megawatts originally requested, so ERCOT applied that factor to new data-center additions and a 180-day project delay was also included — again based on experience. ERCOT used this adjusted forecast to avoid overbuilding on “phantom” load while still capturing the scale of potential growth.

Figure 3. ERCOT Adjusted Large Load Additions. Source: ERCOT

The adjusted (haircut) ERCOT megawatt numbers are shown in the gray bar segments in Figure 3 above and the table beneath the bar-chart legend, increasing from 2,433 MW in 2026 to 24,195 MW in 2031. All we need to do with these numbers is to compute the natural gas required to fuel the generation needed for these data centers, and to do so we’ll just use the same assumptions we used for the bottoms-up calculations — 55% gas-fired and an 8,000 Btu/kWh heat rate.

Figure 4. ERCOT Data Center Demand; Source: RBN

The top-down ERCOT numbers, even after being subject to the 50% haircut, require about twice as much gas as the bottoms-up approach: 2.2 Bcf/d (right end of purple line in Figure 4 above) versus 1.1 Bcf/d (stacked layers). Note that the bottoms-up approach starts off higher but falls behind in 2028. No doubt this includes many projects that have not been made public. It’s an open question as to how much of this capacity growth really comes to pass.

Bottom Line

This is the predicament inherent in all assessments of data center forecasts of implied natural gas demand. You can’t be sure of what is real, how much gas will be needed for each megawatt of power demanded and produced, or how soon the power and gas will be needed. One thing we do know is that there is likely to be a significant ramp in gas-fired generation to power new data centers through the second half of the 2020s, including some new power plants not connected to the ERCOT grid.

Further, the ERCOT analysis reminds us that Texas also will see noteworthy incremental power demand from other large users — crypto mining (light-green bar segments in Figure 3), industrial projects (black bar segments) and the like — and that these need to be factored in, too, given that as a group they may well exceed the incremental demand from data centers. As a whole, it’s possible this incremental gas demand from all non-LNG-export sources (data centers and the rest) will equal or even exceed the demand from a good-sized LNG export terminal or two.

We’re not the only ones trying to get a handle on what’s real and what’s not regarding data centers and their need for power and gas. Utilities around the country are, too, as the Wall Street Journal discussed in an article published on Labor Day. Few are more interested in the topic than ERCOT itself, whose power planning will be greatly affected, one way or another, by how this all shakes out. The Texas grid operator got a boost of sorts from Senate Bill 6, which Governor Greg Abbott signed into law in late June. Among other things, the new law requires that entities planning new data centers and other large energy-consuming projects provide far more detail about their plans, demonstrate site control, and pony up serious money for studies and transmission connections. That should help ERCOT — and all of us — separate the sheep from the goats.

Morgan Overman contributed to the data analysis for this assessment.

“Won’t Get Fooled Again” was written by Pete Townshend and appears as the fourth song on side two of The Who’s fifth studio album, Who’s Next. Townshend originally wrote the song as a closing number for his unfinished rock opera, Lifehouse. The lyric at the end of the song, “Meet the new boss, same as the old boss,” was inspired by an incident at Woodstock where Townshend chased Abbie Hoffman off the stage for grabbing a microphone and preaching his agenda to the crowd during a break in The Who’s performance. On the song, Townshend played a 1959 Gretsch 6120 guitar (given to him by Joe Walsh) through a Fender Bandmaster amp. (We guess you could ask Jimmy Page about the magic of 1959 American guitars that had previously been owned by Walsh — the 1959 Sunburst Gibson Les Paul that Page that used in Led Zeppelin was bought from Walsh.) Townshend would use that combination of guitar and amp for many future Who recordings. Released as a single in June 1971, “Won’t Get Fooled Again” went to #15 on the Billboard Hot 100 Singles chart. It has been used in many television shows and motion pictures. On a sad note, it was the last song drummer Keith Moon played live with the band before his untimely passing in September 1978. Personnel on the record were: Roger Daltry (lead vocals, scream), Pete Townshend (guitars, synthesizer, organ, backing vocals), John Entwistle (bass) and Keith Moon (drums, percussion).

Who’s Next was recorded between April and June 1971 at Olympic Studios in London and on The Rolling Stones Mobile Studio outside of Stargroves, Mick Jagger’s manor home in East Woodhay, England. Production was handled by Glyn Johns and The Who. The front album photo shot by Ethan Russell was an irreverent nod to the monolith used in the 1968 motion picture, 2001: A Space Odyssey. The album was released in August 1971 and went to #4 on the Billboard 200 Albums chart. It has been certified 3X platinum by the Recording Industry Association of America. Three singles were released from the LP.

The Who are an English rock band formed in London in 1962 as The Detours. Their classic lineup from 1964-78 was Roger Daltry, Pete Townshend, John Entwistle and Keith Moon. The Who have sold more than 100 million records worldwide. They have released 12 studio albums, 16 live albums, 27 compilation albums, four soundtrack albums, four EPs, and 58 singles. They have a Grammy Lifetime Achievement Award and are members of the Rock and Roll Hall of Fame and the UK Music Hall of Fame. Eight members have passed through the group since its formation. Founding member Keith Moon died in 1978 and John Entwistle died in 2002. Roger Daltry and Pete Townshend continue with various musicians to record and tour as The Who.