The Plant Is Part of the Campus Now

Google's 702 MW Wyoming fuel-cell deal and NTT's 900 MW Bloom agreement put on-site generation on the same procurement path as chillers and switchgear. The IT load is still unpublished.

Two announcements landed on the same day. Google said its first Wyoming data center, a $17 billion Laramie County project, is contracted with American Electric Power for a 702 MW Bloom Energy fuel-cell plant with carbon capture. NTT DATA said it will deploy 900 MW of Bloom fuel cells across multiple data-center projects as primary power, not backup, with first shipments in 2027. Add them up and you get about 1.6 GW of data-center fuel cells in one news cycle (Google; NTT DATA).

That is not a press-release curiosity. It is a change in how power gets procured for AI campuses.

Generation moved onto the bill of materials

For most of the last decade, a data-center power plan meant a utility queue position, a substation and a PPA. On-site generation lived in the backup column — diesel, maybe some reciprocating engines for bridging. Google and NTT just put solid-oxide fuel cells in the primary-power column.

AEP says it will invest roughly $4 billion in the Cheyenne fuel-cell project under a 20-year PPA with Google. Construction is planned to start in 2026, with full operations targeted by the end of 2028. AEP controls about 92% of Cheyenne Fuel Cell LLC; Basalt Infrastructure Partners holds the rest. The cells come from Bloom under an earlier AEP agreement that covered up to 1 GW of equipment (AEP; Data Center Frontier).

Google is also clear about who pays. It said it will fund 100% of the power it uses and cover the infrastructure that its growth requires. The CO2 is supposed to go to Tallgrass’s Eastern Wyoming Sequestration Hub, lifting Google’s contracted CCS capacity to nearly 1 GW. The campus itself will be air-cooled (Google).

NTT’s framing is different and, for operators, more important. It is using Bloom as primary power to bring capacity online faster in power-constrained markets. Modular systems that can grow with the halls are the point. NTT called it Bloom’s largest international project and said the relationship is designed to reach gigawatt scale. No price, no site list and no firm-versus-framework split were disclosed (NTT DATA).

Do not confuse plant MW with IT MW

The industry keeps making the same mistake. A 702 MW fuel-cell plant is not a 702 MW IT campus. Google has not published the Wyoming IT load. DCD has separately reported a Cheyenne-area campus application pointing toward multi-gigawatt buildout over time, but the fuel-cell figure is the generation side of one contracted package, not the rack count (Data Center Frontier).

Treat plant MW, contracted utility MW and critical IT MW as three different lines on the same spreadsheet. Mixing them is how press-release gigawatts become underwriting errors.

Manufacturing slots are now a schedule risk

Once generation is on the procurement path, the factory becomes a critical path item. Caterpillar’s third-quarter backlog hit $78 billion, almost double a year earlier, with power-generation sales up 30% and a lot of that from data centers. The company is expanding production of large reciprocating engines and Solar turbines (Utility Dive).

Bloom is now attached to both of this week’s flagship on-site announcements. That means fuel-cell manufacturing slots belong on the same risk register as interconnection studies and transformer lead times. A schedule that assumes catalog delivery on large recips, turbines or fuel-cell blocks is already late.

Carbon capture is the other unfinished piece. Google and AEP describe CCS as part of the Wyoming design. Capture rates, equipment configuration and storage arrangements still need to prove out at commercial scale. Do not underwrite near-zero emissions until the capture train is contracted and permitted, not just announced.

Gas, quiet and the community hearing

Bloom’s pitch — and NTT’s language — leans on non-combustion, low local air emissions, minimal water and quiet operation. That is not marketing fluff in a year when ballot measures are writing gallon and megawatt triggers into local law. A fuel-cell block that avoids a cooling-tower plume and a stack opacity fight is easier to defend at a planning commission than a large evaporative campus next to a residential subdivision.

It still needs gas. Solid-oxide fuel cells on natural gas are an electrochemical bridge, not a renewable plant. Underwriters should diligence lateral capacity, commodity supply and any CCS train as carefully as they diligence the Bloom slot. AEP’s $4 billion Cheyenne structure, with Google as offtaker outside AEP’s traditional retail territory, also shows how utilities are willing to build generation as an unregulated or lightly regulated venture when a hyperscaler signs the long PPA. Copycats will appear. Not all of them will have AA+ offtakers.

What to watch

First Bloom shipments to NTT sites in 2027, and whether any sites are named.
AEP construction start at Cheyenne and whether the 2028 COD holds.
Whether other hyperscalers copy the utility-owned, customer-offtaken fuel-cell structure outside traditional service territory.
Fuel-cell and recip manufacturing capacity adds — who actually expands plants, not just order books.

Why it matters for builders, buyers and investors

Behind-the-meter generation is no longer a science project for AI campuses. It is a way to buy down queue risk. That shifts competition toward whoever can lock manufacturing slots, gas supply, permits and credit for the plant as well as the halls. Buyers should ask for plant MW, IT MW and COD as three separate answers. Investors in $BE and the engine makers are underwriting factory throughput as much as demand headlines.

Not investment advice.

Sources

Report a problem with this page ↗