AINEWS
Prometheus Hyperscale's Reeves County plan bypasses the Texas grid queue by going 'islanded' and absorbing the fuel, permitting, and customer commitment risk a utility would otherwise carry.
Grid interconnection queues are now the binding constraint on new AI data center delivery, and one West Texas developer is responding by going around them entirely.
Prometheus Hyperscale is planning a 1.5 GW IT-load data center campus in Reeves County, Texas, with a development path to 2.5 GW. The site is "islanded," meaning it runs on its own on-site generation rather than connecting to ERCOT, the operator of most of Texas's grid, at start-up. The plan, outlined by chief energy and business officer Adam Mirick to Data Center Knowledge, is one of the more explicit attempts yet to vertically integrate power in order to control the delivery date.
The developer now carries fuel logistics, air permits, redundancy design, water use, and the customer commitment that has to clear before the first phase breaks ground.
The 1.5 GW figure is the planned compute demand, not the size of the power plant behind it. At a preliminary 1.3 power-usage effectiveness (PUE) ratio, 1.5 GW of IT load implies roughly 1.95 GW of facility demand. Mirick's illustrative model adds about 30 percent generation overbuild for redundancy, which puts installed capacity near 2.5 GW. All of those figures are projections, and the final design is not set. Prometheus has not published a capex or fuel-cost breakdown.
The first phase is small relative to the headline. Mirick says the developer can deliver 100 to 150 MW of compute in 2027, but only if a customer signs first. A stick-built, custom-built first phase would slip to mid-2028. "It's very challenging to bring anything to market in '27," Mirick told Data Center Knowledge, citing generation, modular data center (MDC) supply, and contractor capacity as the binding constraints.
Prometheus is leaning toward rich-burn reciprocating engines for the initial blocks, rather than the lean-burn gas turbines that dominate utility-scale generation. Reciprocating engines handle fuel variability more gracefully, which matters because the site plans to swing between pure natural gas and an ethane blend as supply changes. The cost is a footprint penalty: more units, more building, more maintenance hours per megawatt than a combined-cycle turbine.
Reliability for the first block is N+1. Mirick says higher reliability targets, measured in nines, are tenant-defined, and the cost climbs nonlinearly as the targets tighten.
Base fuel is pipeline natural gas, delivered through two pipeline connections with firm transportation. In Q1 2028, co-located fractionation developer Istmo Energy is targeted to bring its facility online. Once that happens, Prometheus can run an 85 percent natural gas and 15 percent ethane baseline, with the ability to fall back to 100 percent ethane if gas supply tightens or pricing moves. Onsite ethane and potentially propane storage is planned as a buffer.
The ethane path is the unusual part. Few data center developers have built their generation around a natural gas liquids byproduct stream. It works because Istmo is willing to site a fractionator next to the campus, and because the Permian produces more ethane than the petrochemical complex can absorb. The arrangement also locks the campus into Istmo's Q1 2028 in-service target, which becomes a hard external dependency on the same timeline Prometheus wants the larger build to ramp.
Cooling is closed-loop, running roughly 75 percent water and 25 percent food-grade propylene glycol. Mirick's preliminary combined power-and-hall water estimate works out to less than the equivalent of about 100 households, with no routine discharge. The final design is not set.
The campus still has to clear environmental and air permits, and Texas regulators retain oversight even without an initial ERCOT interconnection. Any future grid tie-in is tenant-directed, not company-driven, which means a customer could later choose to plug the site into ERCOT for redundancy or market exposure without changing the underlying generation.
The reciprocating engines Prometheus is selecting for the initial blocks have spent most of their working lives as backup machines. Continuous duty is a different operating regime for this class of engine. Joshua D. Rhodes has studied the failure modes that emerge when reciprocating engines are pushed into that regime.
"Continuous-duty operation of what was traditionally backup generation increases mechanical stress and can wear systems 2 to 3 times faster," Rhodes said in comments to Data Center Knowledge. He suggests layering battery storage between volatile AI compute loads and the reciprocating engines as a buffer that would also let the engines run closer to their design sweet spot.
Reciprocating engines can run continuously, but the opex bill is different in kind from a peaking-duty maintenance schedule. The wear cost is one of the variables Prometheus did not put on the page when it discussed the $100 to $125 per MWh cost estimate.
Mirick's headline number is roughly $100 to $125 per MWh at a $3.50 per MMBtu gas assumption. Without a detailed capex, financing, fuel, emissions, or maintenance breakdown from Prometheus, the figure functions as a directional estimate from the developer's own model.
Three dates decide whether this plan is real by 2027. First, whether Prometheus signs a customer in time to trigger the 100 to 150 MW first phase. Second, whether Istmo's Q1 2028 fractionator actually comes online, because the 85/15 fuel mix and the ethane fallback both depend on it. Third, whether the reciprocating fleet can be operated under continuous duty at a wear profile that the cost model still has to absorb.
The 1.5 GW headline is a permitted direction, not a delivered project. Whether it lands at scale is now a function of customer signatures, fuel infrastructure, and how fast reciprocating engines can be run without burning through their own maintenance budget.