Opinion | Power
Fuel cells and onsite gas can get AI data centres built years sooner than waiting for the grid, and Google and NTT DATA are showing how. The opportunity is real – what operators do next, from fuel choice to carbon capture, will decide whether it earns trust. Developers facing long grid queues are generating power next to their facilities, using technology that can be installed and expanded far faster than network upgrades. Two recent announcements show how far that has gone, and both contain encouraging signs.
What Google is getting right
Google will invest $17bn in its first Wyoming data centre, near Cheyenne. Its centrepiece is a 702MW natural gas fuel-cell plant with carbon capture and storage, which Google describes as the largest planned commercial-scale fuel-cell development of its kind. AEP plans to invest about $4bn under a 20-year power deal, with full operations expected by the end of 2028.
Several choices stand out here; the plant is designed to capture and permanently store a large majority of its emissions, rather than vent them, it will also use air cooling to reduce pressure on local water resources ans Google says it will pay for 100% of the power it uses – a pledge aimed at keeping costs off other electricity users. Taken together, that is a company designing around the complaints that data centre developments attract.
What NTT DATA is getting right
NTT DATA has agreed with Bloom Energy to deploy 900MW of onsite power across multiple projects, with fuel cells as primary power and initial shipments from 2027. The modular approach lets capacity be deployed in phases alongside construction, which cuts the risk of building ahead of demand or sitting idle waiting for a connection. The companies say fuel cells deliver low local air emissions, minimal water use and quiet operation, which matter to neighbours -whilst those are company claims, they are the right things to be competing on. Bloom’s systems can also run on natural gas, biogas, hydrogen or blends, so the hardware is not locked to one fuel.
A practical bridge
The UK has it’s own version of the same problem; Developers want grid connections totalling around 72GW against a peak demand of 45GW, so waiting is not always an option. The government’s recent Data Centre Covenant allows temporary on-site power for exceptionally strategic projects, subject to conditions such as energy security. One law firm called the move positive, noting that bridging solutions could make a very significant contribution to getting data centres deployed as soon as possible. Engineers say the same: behind-the-meter generation can support earlier energisation, phased development and greater flexibility. And ministers argue that pausing development would push jobs and data overseas.
That is a fair point – a data centre that opens sooner, with a credible plan to cut its emissions, is better than one that waits for years or relocates abroad.
What would make it credible
The strength of the bridge argument depends on the details. Three things to consider that could make a difference:
- A published capture rate. Google says carbon capture will store “a large majority” of the fuel cells’ emissions, but has not given a figure. A number, and a plan to verify it, would turn a promise into evidence. Capture on the gas turbines is still at the study stage.
- A named fuel. NTT’s announcement did not specify the fuel, locations or a rollout timetable. Disclosing them would let communities and investors judge the emissions properly. A pathway to biogas or hydrogen would be a strong story to tell.
- An end date. The BBC’s File on 4 reported that four planned UK data centres are considering on-site gas, with up to 4.1m tonnes of carbon a year as a ceiling if run flat out, and that some describe stopgaps lasting a decade or more. A date by which each arrangement ends or is decarbonised would answer the “temporary becomes permanent” concern.
Our view
Onsite generation deserves to be taken seriously as for many projects, it presents the fastest practical route to powering essential digital infrastructure – and adds resilience. Leaders in the space are already making better choices than the industry average: capture, air cooling, phased delivery and a commitment to pay their way.
The opportunity is to be the operator who can show its working – a data centre running on gas with carbon capture is a different proposition from one on renewables, and neither is automatically better – what builds trust is saying which one you are, and proving it.
There is also a consent angle to consider; In the US, Reuters reported that all but one of the 41 data centre ballot measures voted on last week went against data centres. Publishing clear, measured energy plans and demonstrating how, over time, this can be a cleaner solution than the alternative would be a very positive step towards building trust and winning support.
