Why Data Centers Are Ditching Diesel for Gas Turbines and Fuel Cells

Ask a colocation provider in Loudoun County or a hyperscaler siting a campus in West Virginia what keeps their power team up at night, and increasingly the answer isn’t the utility interconnection queue — it’s the generator yard. Diesel backup, the boring, decades-old insurance policy behind every “five nines” uptime claim, has quietly become one of the most contested line items in data center planning. Virginia’s Department of Environmental Quality has imposed a new Tier 4 emissions baseline on large data center generators effective July 1, 2026, under House Bill 507. Texas regulators are capping how many hours backup units can run for grid support. And a wave of capital — from Caterpillar’s 2 GW order for a West Virginia AI campus to Bloom Energy’s 2.8 GW fuel cell agreement with Oracle — is flowing into alternatives that didn’t exist as procurable, at-scale options three years ago. For anyone buying or planning data center capacity right now, backup power strategy has stopped being a checkbox and started being a negotiating variable.

Diesel Isn’t Disappearing, But Its Regulatory Runway Is Shrinking

Diesel generators are not going away — there are still thousands of them installed across Virginia’s data center corridor alone, representing more than 11 GW of backup capacity, according to figures cited from the state’s Joint Legislative Audit and Review Commission (JLARC). JLARC’s own analysis found data center generators account for less than 4% of regional nitrogen oxide emissions, and industry data suggests most units run only 10 to 30 minutes a month for routine testing. That’s the case operators have made for years: these are emergency assets, not primary generation.

But two things have shifted the politics. First, utilities and grid operators including PJM have increasingly floated using data center backup fleets for demand response — running them during grid stress events, not just outages. Federal rules already allow up to 100 hours of non-emergency generator testing annually, with as many as 50 of those hours usable for demand response, and some industry proposals have pushed for 50 to 100 hours of active onsite backup use. That’s a meaningfully different emissions and noise profile than “emergency only,” and it’s exactly what triggered community pushback in Virginia and prompted HB 507. The new rule sets a presumptive Best Available Control Technology standard of 0.60 g/hp-hr for large units, effectively requiring Selective Catalytic Reduction systems, diesel particulate filters, and continuous emissions monitoring on new air permits. Compass Data Centers and the Data Center Coalition have pushed back on cost, while the Southern Environmental Law Center and Piedmont Environmental Council have pushed for stricter limits still. Similar fights are playing out in Texas, where ERCOT and the Texas Commission on Environmental Quality have been weighing rules that would cap backup generator operation at roughly 10% of a facility’s primary power hours.

The practical upshot for buyers: if your build requires a new air permit anywhere with an active generator debate — Virginia, Texas, Georgia, and a growing list of other states — budget for SCR-equipped units, longer permitting review, and a real chance that your emergency power plan gets litigated by neighbors before it gets approved by regulators. If you haven’t already, it’s worth revisiting how your generator fleet is sized and specified in the first place; our data center generator sizing guide walks through the capacity and cost tradeoffs that now sit underneath these permitting decisions.

Gas Turbines and Engines Are Absorbing the Overflow — With Their Own Multi-Year Queue

The most visible response to both grid delays and diesel scrutiny has been a rush into gas-fired generation, sized for continuous or near-continuous operation rather than emergency-only backup. Wärtsilä has installed 15 gas engines delivering roughly 300 MW at a site in Ohio. Caterpillar has taken a 2 GW order of its fast-response Cat G3516 units — which can ramp from zero to full load in about seven seconds — for the Monarch Compute Campus in West Virginia, with the 2,250-acre site potentially expanding to 6 GW total. At Stargate, the Oracle/OpenAI AI campus in Abilene, Texas, mobile GE Vernova LM2500XPRESS turbines and Solar Turbines Titan 350 units are reportedly supporting up to 1.2 GW of consumption across a roughly 4-million-square-foot footprint. Compressed natural gas logistics firm Certarus is already supplying more than 120 MW to data center operations and has a 135 MW project slated to start in 2027.

The catch is that this market is now capacity-constrained too. Major OEM gas turbine backlogs stretch to the end of the decade and beyond, which is exactly why so much of the recent activity has centered on smaller, trailer-mounted mobile units (Dynamis Power Solutions, for instance, offers 8 to 70 MW trailer-mounted turbines, including a 24 MW unit at 13.8 kV) rather than large frame turbines that would otherwise take years to deliver. Even packaged boilers for steam-cycle plants are running roughly a one-year lead time. Omdia analyst Shen Wang has projected that AI data centers will need another 60 GW of new power capacity every year through 2030 — a number that dwarfs what any single generation technology can deliver on its own, which is precisely why buyers are seeing hybrid solicitations that blend grid power, gas, and fuel cells rather than betting on one horse.

Fuel Cells Are the Fastest Thing You Can Actually Deploy Right Now

If speed is the constraint, fuel cells currently have the best story. Goldman Sachs Research, in a November 2025 note from Michele Della Vigna, estimated fuel cells could meet 6% to 15% of the incremental power demand data centers will need through 2030 — translating to roughly 8 to 20 GW of fuel cell capacity, out of an expected 25 GW of total behind-the-meter generation supplying a quarter to a third of the 730 terawatt-hours of new data center demand Goldman projects between 2024 and 2030. The appeal is deployment speed: Goldman puts fuel cell installation at under a year, versus five-plus years for new gas turbines, with 10% to 30% better efficiency and none of the NOx or CO output diesel and gas combustion produce.

The deals underway back that thesis. Bloom Energy signed an agreement with Oracle for up to 2.8 GW of fuel cell capacity, announced in April 2026, and separately crossed 100 MW with Equinix. In October 2025, Bloom struck a $5 billion strategic partnership with Brookfield, whose global head of AI infrastructure, Sikander Rashid, called behind-the-meter power “essential to closing the grid gap for AI factories.” On the hydrogen side, Microsoft has piloted a 3 MW hydrogen fuel cell system with Plug Power in Latham, New York, and a smaller 250 kW system with ESB in Dublin, while Innio’s Jenbacher engines now run on hydrogen-natural gas blends up to 10 MW, with 1 MW hydrogen-only units already available. Hydrogen still faces real hurdles — the EPA’s mid-2024 final greenhouse gas rule dropped an earlier proposal that would have required 30% hydrogen co-firing by 2032, and infrastructure like the roughly 300 km of repurposed hydrogen pipeline in the Netherlands remains the exception rather than the norm. Natural-gas-fed fuel cells are the near-term workhorse; hydrogen is the option being kept warm for later.

What This Actually Changes in a Buyer’s Contract and Timeline

None of this is academic for anyone signing a colocation agreement or planning a self-build. Three things are worth pressure-testing in any active deal. First, ask what backup and behind-the-meter technology is specified, not just the total MW — a fleet of older diesel gensets in a state moving toward Tier 4 rules is a future capital expense and permitting risk you’re implicitly inheriting, whether or not it’s itemized in your contract. Second, treat gas turbine and fuel cell lead times as part of your delivery-date risk, not just the provider’s problem; a 2027 commissioning date built around a turbine order placed into a backlogged OEM queue is a date worth confirming, not assuming. Third, in jurisdictions with active generator litigation — Virginia and Texas being the clearest examples right now — expect permitting timelines and community engagement requirements to lengthen further before they stabilize, and factor that into any site-selection timeline alongside the grid interconnection delays discussed elsewhere on this site.

None of this makes diesel obsolete overnight; it remains the cheapest, most field-proven emergency backup available, and most existing fleets aren’t going anywhere. What’s changed is that “we’ll just run diesel gensets” is no longer a default answer that clears regulatory and community review without friction, and it’s no longer the only procurable option for operators who need firm, fast power now. The operators moving first on gas turbines and fuel cells aren’t doing it out of environmental idealism alone — they’re doing it because in several key markets, diesel-only power plans are becoming the slower, riskier path to energization, not the safer one.

If you’re evaluating capacity commitments or negotiating power terms in the next 12 months, run your own diligence checklist against the generator mix, permitting status, and delivery timeline before you sign — our Data Center Buyer’s Toolkit has the framework to help you do exactly that.

Written by

Raajeev Ratra

Data Center Infrastructure Expert | 15+ Years in DC Design, Operations & Project Management

Raajeev is a seasoned data center professional with hands-on experience in hyperscale facilities, colocation design, power & cooling infrastructure, and global DC operations. He shares practical insights to help engineers and IT leaders build better infrastructure.

Connect on LinkedIn →