The constraint on new data-center load isn't fuel anymore. It's the wires. Grid interconnection now takes 3 to 5 years in the busiest US markets, so developers are turning to on-site fuel cells. Vendors quote about 90 days to energize a module, but that is a best case covering the equipment only: air and emissions permits, gas service and pipeline capacity, civil works, switchgear and interconnection of the on-site plant typically dominate the real schedule. This page covers how big that shift has become, whether on-site power actually lowers your carbon, and where green hydrogen fits.
About 2,290 GW of generation is stuck in US interconnection queues, roughly twice the entire installed fleet. A typical project waits about 4.5 years to connect, and only one in five ever makes it. For an AI data center that needs 100 MW or more now, that wait is a dealbreaker.
So the industry is routing around the grid. Solid-oxide fuel cells, running on natural gas today and hydrogen tomorrow, generate power on-site in modular blocks a vendor can energize in about 90 days once permits, gas service and site works are complete. By 2030 an estimated 38% of data centers expect to use on-site generation for primary power, up from 13% a year earlier.
Almost entirely US, almost entirely to bypass grid delays. A starter view; the market is moving fast.
| Buyer type | Technology | Scale | Note |
|---|---|---|---|
| Major cloud / AI provider | Solid-oxide fuel cell | up to 2.8 GW | Multi-year supply agreement (2026) for AI capacity |
| US power utility | Solid-oxide fuel cell | 1 GW | ~$2.65B, 20-year offtake (2026) |
| Colocation operator | Solid-oxide fuel cell | 104 MW | Primary power across ~19 data centers in 6 US states |
| AI cloud provider | Solid-oxide fuel cell | 14 MW | Primary on-site power for an AI data center |
What this means for your siting decision: on-site generation is no longer a niche backup plan. It is becoming a primary way large loads get built on time. When you weigh a location, treat "can I self-generate here, and at what carbon and cost?" as a real option next to the grid connection, not an afterthought.
The emerging playbook for a large load that cannot wait years for a connection.
On-site gas carries the site at first. The utility connection takes over as the network is built. Then the gas plant steps back to a supporting role. A common shape looks like this:
| Phase | Years | What carries the load |
|---|---|---|
| Bridge | 1 to 3 | On-site gas carries 100 percent of the load, so the site can energize almost immediately. |
| Utility ramp | 4 to 6 | The first grid megawatts arrive and ramp up as network upgrades complete. |
| Handoff | 7+ | Gas steps back to backup or a grid-supporting resource, or is redeployed to the next site. |
Often yes, but it depends entirely on how clean your grid already is.
A natural-gas solid-oxide fuel cell runs at about 55 to 60% electrical efficiency, which works out to roughly 320 gCO₂/kWh (manufacturers cite around 50% below the US grid). That sets a simple break-even at about 320 gCO₂/kWh. And not all on-site power is clean. A diesel generator runs around 680 gCO₂/kWh, dirtier than most grids.
Where the grid is dirtier than ~320:
Where the grid is already cleaner than ~320:
So a gas fuel cell is a carbon win in fossil-heavy grids and a carbon loss in clean ones. There are two routes to lower emissions at the stack. The first is biogas or hydrogen, which emit no fossil CO2 in the cell itself; whether they are low-carbon overall depends on the production pathway, upstream methane, and transport, which the cell figure does not capture. The second is natural gas paired with carbon capture, which lowers but does not eliminate emissions: capture rates are partial and upstream methane is unaffected. PowerGridIQ shows this comparison on every region's page against that grid's live carbon number, so you can check it for your exact site.
Hydrogen fuel cells emit no CO2 at the point of use. That is not the same as carbon-free: the lifecycle result depends on how the hydrogen was made and delivered. The catch is cost. Green hydrogen runs about $2 to $2.50/kg in the best regions (MENA, Australia, Chile), $3 to $6/kg more broadly, and $0.50 to $2/kg in the US with the 45V subsidy. Because a fuel cell turns about 1 kg of hydrogen into roughly 20 kWh, that works out to:
| Green H₂ price | ≈ Fuel cost | Verdict |
|---|---|---|
| $1/kg (subsidised) | ~$50/MWh | Competitive |
| $2/kg (best regions) | ~$100/MWh | Viable where a clean-hydrogen mandate applies |
| $5/kg (typical today) | ~$250/MWh | Premium: pays for carbon, not cost |
For comparison, the fuel for a natural-gas fuel cell runs only about $17 to $23/MWh, far cheaper today. So hydrogen wins on carbon, not yet on cost. The sweet spot is where power is sometimes nearly free. Grids with frequent very-low or negative price hours (Germany saw about 573 in 2025, along with Spain, Texas and California) are where you can run electrolysis on otherwise-wasted surplus and store it as hydrogen for the evening peak. As green hydrogen falls toward $1 to $2/kg this decade, the maths tips further.
Speed. Grid interconnection in the busiest US markets takes 3 to 5 years, while a fuel-cell module can be energized in about 90 days once permitting, gas service and site works are done; the full project schedule is usually considerably longer. With the AI build-out racing, time-to-power is the competitive edge.
Versus a dirty grid, yes. A gas fuel cell is about 320 gCO₂/kWh, below most fossil-heavy grids. Versus a clean grid (Nordics, France, Québec, Brazil) it is worse. The break-even sits around 320 gCO₂/kWh.
Technically yes, economically not quite. Green hydrogen is carbon-free but costs about $2 to $5/kg, which puts fuel at roughly $100 to $250/MWh. It is competitive today only with cheap surplus renewables or subsidies, and it improves as hydrogen falls toward $1 to $2/kg.