This is a follow-on post from an earlier post I dashed out – Hydrogen datacentres – is this legit? – where a podcast interview caught my attention about one approach being sold to address demands on the grid caused by new datacentres. This post will make more sense if you have read it.
Basically ECL is one provider selling modular hydrogen powered datacentres, and they’re a good example for understanding the complexities of getting off fossil fuels by switching to hydrogen. The quote below is from a datacentre dynamics piece sharing an update about plans for a colossal new datacentre project.
Data center startup ECL claims that it will build a 1GW AI data center on a more than 600-acre site east of Houston, Texas, powered by hydrogen.
The TerraSite-TX1 will initially have a capacity of 50MW, coming next summer, at a cost of around $450 million. AI cloud provider Lambda will be the first customer, but is not taking the whole 50MW.
For context, most massive hyperscale datacentres are around 100MW, and this would final site would be ten times that. This also gives an idea of how they might get around the constraints on the grid, because drawing a gigawatt of power 24/7 would likely require lots of new grid infrastructure – they’re basically relying on gas pipelines:
ECL launched its first modular hydrogen data center back in May, with a small deployment at its site in Mountain View, California. Each module supports 1MW, and can cool up to 75kW per rack.
For the TerraSite, ECL said that it has three pipelines of hydrogen that will feed the facility, with an energy cost of 0.08-0.12/kWh.
The company repeatedly touted the project as producing zero emissions, but the hydrogen will come from natural gas using steam methane reformation – a process that causes carbon emissions. Update: ECL told DCD that it was referring to data center operations, not the end-to-end process.
The point above is important, because the kind of hydrogen being referred to here will not be green hydrogen. It’ll be hydrogen that comes from a takes methane (usually from fossil sources, so fossil gas), and combines it with water, to produce hydrogen and carbon dioxide, which is usually vented into the sky. The page on wikipedia is pretty good if you want to dive in deeper about the process.
Anyway, the costs here of 0.08-0.12/kWh, equate to 80 to 120 USD per megawatt hour. This is more than bog standard grid power, delivered in the USA, but if there’s no combustion, then in theory it could be clean – if it weren’t for how the hydrogen is made, which currently still emits CO2, and is arguably worse, because you’re losing energy when you convert the methane to usable hydrogen, and again when you’re converting it to electricity onsite in the fuel cells that the datacentres use.
Is there a way to make it cleaner?
You could use green hydrogen, made by splitting water into hydrogen and oxygen using electricity is likely to multiples of that cost – my intuition is probably 4-5x the cost easy. So, you’re looking at maybe 400-500 USD per megawatt hour. That’s much, much more than the grid in most cases, especially if it’s someone’s job to buy power professionally using some kind of hedged contract (above a certain scale, it makes sense to have professional buyers of power like this).
You could also use blue hydrogen, which is likely what will happen if it’s in Texas. Blue hydrogen is essentially hydrogen created using the process above, but the carbon is captured and stored somewhere. This is still pricey, but assumed to be less pricey than green hydrogen for the most part. That’s likely to increase the price again. A ballpark figure might be in something like 200-300 USD per megawatt.
In various parts of the world, there are massive subsidies available for blue hydrogen. Subsidies from the US Inflation Reduction Act are what make some of these projects financially viable for example.
There are also all kinds of open questions about how much methane leaks end up making this much less clean than we first hoped, because when methane leaks out of pipelines, it’s much, worse climate wise than carbon dioxide. This matters, because the same mass of methane causes something in the region of 40-80x more global warming as the same mass of carbon dioxide – you only need a little bit leaking and your emissions are worse than coal.
What about water?
One of the issues coming into the public discourse aroud datacentres is how thirsty they are – because significant amounts of water are used to cool the servers down, usually from local acquifers. This is often water that even when returned isn’t suitable for use for agriculture or drinking anymore, and understandably, local communities when they learn about it, often take a dim view of this level of water consumption.
One interesting aspect of datacentres running on fuel cells here is that because water is a byproduct of using hydrogen fuel cells to generate electricity, it means that the water footprint of facilities can go from being sink, to a source of water. As ever, there is nuance. The water still needs to be treated before it’s drinkable, but given water use is a common talking point about AI, it’s likely to come up.
Yes, these are looking more legit, if incredibly expensive
Anyway, I wrote this to come back to later, because I can see the use of hydrogen fuel cells being used like this as a sort of gotcha when people raise concerns about AI datacentres. In some ways, using tech like this is an improvement on on-site gas generation, or placing a massive new load directly onto the grid, , but it still brings its own complications.
As you can see it’s extremely expensive, so it’s likely to rely on public subsidies to be viable, and there lifecycle emissions to think about.
If you’re curious here’s the full piece – ECL says it will build a 1GW hydrogen-powered AI data center in Texas, with Lambda as its first tenant.
Leave a Reply