Fuel cells at gigawatt scale

Bloom Energy fuel cells are real, commercial hardware — but Project Jupiter would run them in a configuration nobody has built before. This page separates what is proven from what is still a bet.

Two different questions (do not mix them up)

Do fuel cells work?

Yes. Solid-oxide units electrochemically convert natural gas to electricity with far less combustion than a turbine. Bloom has deployed them for years at hospitals, factories, and data centers.

Has anyone run this setup at this scale?

No. Not as the sole primary power for a ~2.45 GW islanded campus with thousands of exhaust stacks and no grid connection for normal operations.

Hearing “unproven technology” in the debate? Both sides are often talking past each other. Conflating those two questions is how the argument gets louder without getting clearer.

What is proven elsewhere

Bloom deployments exist today — but almost always as one piece of a larger mix, not the entire campus running alone:

Equinix — 100+ MW across 19 data centers (supplemental / behind-the-meter)
AWS Ohio — ~73 MW, often cited as the largest single-site U.S. deployment to date
Oracle Stargate (Abilene, TX) — Bloom is ~10% of an eventual ~10 GW campus, alongside gas, grid, and renewables

In those cases fuel cells are usually bridge power or one source among several — not the only generation on an islanded

“microgrid”

Marketing calls it a microgrid, but the permit is for ~2.45 GW. Compare that to New Mexico’s grid:

~11.9 GW NM net summer plant capacity (EIA 2024) — Jupiter is about one-fifth of that
~3.5 GW average statewide electricity load — Jupiter is roughly 70% of that
~2.3 GW El Paso Electric summer peak, southern NM (2024) — about the same size as this plant alone

What would be new at Project Jupiter

~2.45 GW ceiling — roughly the size of El Paso Electric’s entire generating fleet in Texas and New Mexico
100% Bloom as primary power — an islanded
“microgrid”
Marketing calls it a microgrid, but the permit is for ~2.45 GW. Compare that to New Mexico’s grid:
- ~11.9 GW NM net summer plant capacity (EIA 2024) — Jupiter is about one-fifth of that
- ~3.5 GW average statewide electricity load — Jupiter is roughly 70% of that
- ~2.3 GW El Paso Electric summer peak, southern NM (2024) — about the same size as this plant alone
with no grid tie for normal operations (only a small substation for offices and commissioning)
Thousands of individual exhaust points — public comments and permit filings cite ~2,275 stacks, each needing pollution controls and monitoring
AI baseload reliability — continuous high-density load with no grid backup; other operators have passed on fully islanded fuel-cell designs for load-response reasons
No operating history — the 2.45 GW is a contract ceiling, not a delivered fleet with a 12-month track record

Plain English: Fuel cells are proven technology. This deployment is not. Oracle is volunteering to be first at gigawatt-scale, fuel-cell-only, islanded primary power.

What the builders say vs. what skeptics hear

Builders (Oracle / Bloom / BorderPlex)

Fuel cells were not viable at this scale until the Bloom partnership matured
Modular units can scale faster than waiting for gas-turbine delivery slots (often 2028–2029)
Electrochemistry beats combustion on local air pollutants — basis for the ~92% NOx claim vs turbines
Oracle will pay all energy costs — no impact on local electricity rates

Skeptics (including lawmakers in local press)

“New, unproven technology that’s still fossil-fuel-based” — why not solar + batteries?
Switching designs reset the permit clock after heavy opposition — looks reactive, not visionary
~10.1 million tons of CO₂e per year is still an enormous climate footprint
Manufacturing and delivering thousands of units on schedule is an open question

Which claims are proven, modeled, or still open?

Claim	Status
~92% less NOx vs old gas-turbine plan	Applicant math — NMED has not ruled. See the permit guide.
~10.1M tons CO₂e per year	Filing estimate — still enormous in absolute terms; both plans run on natural gas.
Won’t affect local electricity rates	Oracle pledge — not an air-permit enforceable condition.
Reliable 24/7 AI power at ~2.45 GW	Unproven — no operating history at this configuration.
~2,275 stacks, all controlled adequately	Under NMED review — legitimate engineering question for this permit.
Manufacturing and delivery on schedule	Unknown — single Fremont factory, massive order book.

What NMED can — and cannot — decide

NMED can review (air permit)

Whether modeled emissions meet New Mexico air rules
Whether thousands of stacks have adequate pollution controls on paper
Monitoring, operating limits, and stack-testing requirements

NMED cannot really adjudicate

Whether Bloom can manufacture and deliver thousands of units on time
Whether the campus stays online through a heat wave or supply disruption
Whether solar, grid power, or another strategy would be “better” long-term

So “unproven at scale” is a fair public concern — but it sits partly outside the permit’s statutory box. NMED is being asked: If they build what they describe, does the air math work? — not Is this a good bet?

WhutNext take: You can hold all of these at once: the fuel-cell plan may be better than gas turbines on local smog, still ~10M tons/year of climate pollution, and a first-of-kind deployment nobody has run before — without picking a tribal side. That is how you slow the reaction down.

Sources: Oracle / Bloom announcement (Apr 2026); Santa Fe New Mexican; Haussamen.com; Electron Economics analysis (May 2026). WhutNext has not independently verified applicant figures.