Data center fuel cells explained: what Bloom's gas-fed cells do, and why the carbon capture is the open question

Data centers need power faster than most grids can deliver it. One answer is to put a power plant next to the building. For Google's new Wyoming campus, announced November 5, 2026, a large part of that plan is a 702 MW development of natural gas fuel cells made by Bloom Energy, run by American Electric Power (AEP). If you have searched for how data center fuel cells work, the short version is that they turn gas into electricity without burning it, and the long version has several caveats worth knowing.

What a fuel cell does differently

A turbine burns gas to spin a generator. A fuel cell converts the chemical energy in a fuel into electricity through an electrochemical reaction, with no flame. TNW's November 5 report describes the Wyoming plant in exactly those terms: fuel cells that generate electricity from natural gas without burning it. Bloom's cells are solid oxide fuel cells, a type that runs at high temperature and can take natural gas directly, according to Data Center Frontier.

The practical points for a data center are modularity and speed. Fuel cells come in repeatable blocks, so capacity can be added in steps, and the site does not wait for a transmission line. That is why Bloom and AEP's original arrangement in November 2024 was for up to 1 GW of equipment, starting with 100 MW and options for 900 MW more.

The Wyoming numbers

Data Center Frontier's November 4 report gives the project's shape:

• About $4 billion of AEP capital for 702 MW.
• A 20-year power purchase agreement with Google, pricing undisclosed.
• Construction expected to start before the end of 2026, with full operation targeted for late 2028.
• Ownership through Cheyenne Fuel Cell LLC: AEP about 92%, funds advised by Basalt Infrastructure Partners the rest.
• An estimated capital intensity of about $5.7 million per MW, which the outlet warns should not be compared directly with conventional gas plants.

Google's own post calls the development the largest planned project of its kind and says the plant will include carbon capture and storage, with captured carbon sent to Tallgrass Energy's Eastern Wyoming Sequestration Hub.

The part nobody has quantified

The headline claim is near-zero-emissions power from natural gas. Here is what the public record supports.

Bloom says its fuel cells produce a more concentrated carbon dioxide stream than combustion, which could simplify capture, according to Data Center Frontier. Bloom and Chart Industries partnered in February 2025 on capture, aiming to prepare carbon dioxide for industrial use or permanent geological storage.

But the same report says AEP has not disclosed the capture rate, the capture equipment supplier or the arrangements for transporting, using or storing the carbon dioxide. TNW adds that Google says capture will remove "a large majority" of the plant's emissions and has given no figure. So the design intent is documented. The performance is not.

That gap is the right place to focus. Natural gas fuel cells without capture still emit carbon dioxide. With capture, the emissions depend entirely on the capture rate and on whether the storage hub is permitted and operating by 2028.

How it fits the wider power picture

Fuel cells are one of several ways AI campuses secure power outside the grid queue. Our data center power tracker logs the gas turbines, nuclear deals and on-site generation across sites, and Wyoming is an unusual case because it stacks several of them: Black Hills grid service, a private microgrid and Tallgrass turbines on top of the AEP fuel cells.

Data Center Frontier cautions that the AEP 702 MW and Black Hills' 2.1 GW of third-party resources should not be added together, because it is unclear whether they overlap. The same caution applies to anyone summing fuel cells with turbines to estimate the campus draw.

What to check before believing a fuel cell claim

• Capacity versus output. Megawatts of nameplate capacity are not megawatts delivered around the clock. Ask for availability figures.
• Capture rate and basis. A percentage with no basis, such as share of stack emissions versus lifecycle emissions, tells you little.
• Storage readiness. A capture unit with nowhere to send the carbon is a cost, not a climate benefit.
• Price. The Google contract price is undisclosed, so claims about cost competitiveness cannot be tested.
• Timeline. Late 2028 is a target, not a commitment.

Fuel cells beside turbines

Wyoming does not rely on fuel cells alone. Tallgrass's power hub would use gas turbines, and Google says it is studying capture for them too. Fuel cells and turbines solve different problems: the cells are modular and quick to add in blocks, while large turbines supply bulk capacity at lower capital cost per megawatt. Data Center Frontier's estimate of about $5.7 million per MW for the AEP plant is a reminder that speed and modularity carry a price, and the outlet cautions against comparing it directly with gas plants.

Bloom also promoted an 800V direct-current-native power design in September 2026, but AEP and Google have not said whether the Wyoming plant will use it. For the wider argument about generating power on a campus instead of drawing from the grid, see our explainer on behind-the-meter power for data centers.

Our take

Fuel cells are a serious tool for getting a data center powered on time, and the Wyoming project is a real, financed example. The low-carbon label is a promise, not a measurement. When AEP or Google publishes a capture rate with a stated basis and a storage permit, this explainer should be updated. Until then, describe the plant as gas-fed fuel cells with planned capture, which is what the documents say.