Why Behind The Meter Data Centers Won’t Save Us
How the Compute Heat Rate shows up behind the meter and in gas markets over time
SB Energy and SoftBank recently announced a massive new AI data center complex in Ohio. The site is slated to max out at 10 gigawatts (!!!) of compute capacity, which will be powered in large part by 9.2 GW of dedicated natural gas generation. When fully constructed, this will be the largest single-site power commitment in data center history.
It’s sometimes hard to picture gigawatts of energy demand, so what helps me is to remember that 1 GW of energy demand is about as much as a small city. Denver, Colorado, the closest city to my home, has a peak demand of about 1.5-2 GW.
This data center will use as much electricity as five to seven Denvers.
This proposed project is just wild to write about: as much electricity demand as many Denvers, popping up on a concentrated real estate footprint on DOE land, at the former Portsmouth Gaseous Diffusion Plant in Piketon, Ohio. The first phase is slated to come online as soon as 2028. The scale is more like national energy policy versus a corporate energy procurement
Obviously, the headlines have focused on the size, the billions of dollars of investment, and the geopolitics, but I’ve seen less on the impact on the energy and gas markets. What does it mean if a single site uses as much energy (electricity via gas) as 5-7 Denvers, and who can afford to pay for all this infrastructure?
The Compute Heat Rate That Pays for All of This
Earlier this year, I published the Compute Heat Rate1, a framework that quantifies the maximum electricity price AI workloads can rationally tolerate before curtailing. The blended figure across workload tiers for Q2 2026 is approximately $8,000/MWh, roughly 160 times the price tolerance implied by the gas heat rate on the supply side, and 40 to 100+ times the thresholds at which traditional industrial loads like smelters, steel, and chemicals curtail (which typically economically curtail when electricity prices are around $100-160/MWh).
Even though the CHR was originally built as an electricity market framework, nothing about it is specific to electrons delivered over wires.
When data centers move en-mass to behind the meter gas generation, the CHR effect moves with it, from the electric grid to the gas markets.
Let’s do the math. Assume a combined cycle gas plant burns about 7 MMBtu per MWh electricity produced. If a data center can tolerate $8,000/MWh before economically curtailing, the implied CHR for gas is $1,142/MMBtu. Let’s be conservative and take the utmost lower bound of the blended CHR, and call it $500/MWh CHR. This still converts to $71/MMBtu, or about 20X times the current price for natural gas.
The Piketon data center project is intended to be powered largely by behind-the-meter natural gas generation, up to 9.2 GW of electricity worth. Assuming the same 7MMBTu heat rate running at about 85% capacity, the facility would burn about 9,200 MW x 7 x 85% x 24 hours = about 1.3 billion cubic feet (Bcf) of gas per day, or about 500 Bcf/year.
That’s more gas than every household in the state of Ohio combined consumes in a single year.
So again: out of nowhere, you’re building more demand for gas than an entire state in just a few years’ time.
The Convergence Theorem Extended
In the CHR framework, the phenomenon where basically every market structure converges on the same CHR repricing outcome once AI data centers start dominating the demand stack is called the Convergence Theorem2. Depending on the market structure, it may show up differently. A grid-connected data center in a deregulated wholesale electricity market will transmit CHR price tolerance into wholesale pricing. A behind-the-meter data center powered by natural gas transmits CHR price tolerance into the natural gas market over time. It doesn’t vanish, it reroutes.
Access the foundational research at computeheatrate.com.
Energy Innovation’s Jeffrey Rissman and Eric Gimon recently published a piece in Utility Dive saying behind-the-meter data center gas plants will raise US energy bills3. Data centers that never touch the grid, they argue, still raise everyone’s electricity and heating costs because their gas plants compete for the same fuel that heats homes and sets the marginal price of power. Citing Michael Thomas and Cleanview, they note that 59 data centers have announced roughly 90 GW of behind-the-meter capacity so far.
Credit where due: a version of this extension was first put to me by Jeff Cook-Coyle of Premier Energy in a comment thread on this Substack4 back in April, including an observation that makes the dynamic worse, not better. Gas markets are structurally more volatile than power markets. Storage buffers are thinner, spikes are sharper, and recoveries are slower. Drop a price-insensitive demand block into a volatile commodity market and the spikes get taller while the mean-reversion gets weaker.
When I first drafted a similar article looking at natural gas, behind the meter, and the Convergence Theorem in April 2026, Clearview was estimating around 56 GW of behind the meter capacity across 46 sites. Now, in June 2026, it’s already up to 90 GW across 59 sites. McKinsey estimates about 25-35% of data center capacity will be run BTM and mostly on gas. If 25 GW of this gets built, that’s about 3.5 Bcf per day, which is about 10% of the total gas consumption of the entire US power sector in 2025. Other estimates suggest data center natural gas consumption will be between 5-12 Bcf per day by 2030.
At some point, we need to look at natural gas markets and ask ourselves whether $3-and-change natural gas will still be around in 3-5 years once all of this gets built.
What This Means for the Affordability Debate
Natural gas is historically one of the more price-elastic energy types. Literally the reason gas is on the margin in most electricity markets is because it can ramp up and down quickly, either meeting short term demand (peakers) or running more hours when prices are low. Industries that heavily rely on natural gas are somewhat price sensitive and can curtail some operations when prices get too high. The US produces a ton of natural gas and is largely shielded from international gas price volatility as a result. Our exports are limited to the capacity of our LNG terminals which are operating at full capacity.
But if we have tens of gigawatts of new demand getting built here, demand that is not price sensitive at all and will keep consuming billions of cubic feet of gas per day and almost never curtail, what does that mean for pricing in the mid and long-term?
Bring-your-own-power, especially if it’s gas, is not an escape hatch from the economics. The repricing happens through one market or another, over one timeline or another. The Piketon project is being pitched as ratepayer-protective, compliant with the Ratepayer Protection Pledge and allocating billions to upgrade transmission and distribution. But if gas prices increase dramatically over time, this too affects ratepayers, both on their heating bill and on their electricity bill.
Every analysis of data center cost impacts, including this one, ultimately runs into the same structural fact: this demand class has a price tolerance unlike anything the grid or the gas system has ever absorbed. How and when it gets absorbed is determined by the price tolerance of the load, and that is the variable the Compute Heat Rate was built to measure.
The next CHR extension may need to look closely at how fast the gas market reprices and who absorbs the spread first.
More soon.
Hans Royal is the originator of the Compute Heat Rate™ (CHR) framework. All views are his own and do not represent those of any employer or affiliated organization.
Royal, Hans, The Compute Heat Rate: Quantifying AI-Driven Electricity Price Tolerance
and Its Implications for Wholesale Market Repricing (February 28, 2026).
Available at SSRN: http://dx.doi.org/10.2139/ssrn.6322318
Royal, Hans, The Convergence Theorem (2026). Available at https://computeheatrate.com/research#convergence
Rissman & Giman, Behind-the-meter data center gas plants will raise US energy bills (June 8, 2026). Available at https://www.utilitydive.com/news/data-centers-raise-energy-bills-not-for-reason-you-think/822205/
Royal, Hans, Pue vs. CHR (March 21, 2026). Available at https://computeheatrate.substack.com/p/pue-vs-chr
A very welcome assessment, and thank you for the hat tip. A small update: I am now part of the team at Legend Energy Advisors. We have been focusing on this quite a bit in recent discussions. We do not have a good answer yet on how to prepare.