Part of my job at work is to stay up to date about the options for decarbonisation of the digital sector. I listened to a podcast yestersday where there were a load of claims that seemed well.. too good to be true, even with my developing knowledge of the sector. I’ll outline them below, because if they’re true, they change my mind about a number of things.
This is just a quick notes post, so please excuse the poor prose. Here’s the link to the podcast episode in question – Episode 47 – Hydrogen data centers get real, with Yuval Bachar, ECL.
The key thing they talk about is the deployment of modular datacentres that come in 1MW ‘modules’ of generation. They can be chained together, so for a 15 megawatt datacentre, you’d have a 15 modules.
Here’s the key things that stuck out to me:
Speed of building and deployment
I’m aware that datacentres typically take years to build, and big part of this is getting planning permission, because they represent such significant loads on the grid, and increasingly, we see push back from local communities who feel that the costs and harms are imposed locally in concentrated form, but most of the benefits flow out to the owners of the facilities without much of the upside being shared back to the surrounding community. The claim that raised my eyebrows was that a multiple megawatt datacenter could be deployed in less than a year – typically around 8 months. That’s muuuch faster than normal.
The cost of energy
I was pretty glad they touched on energy costs too, as this is an area of interest for me – I’ll link to the specific past of the podcast but the general claim was that they could provide hydrogen to the fuel cells to power the entire datacentre, at prices competitive to the grid.
The figures used between 8 US cents and 50 US cents per kilowatt hour, which seemed incredibly low for hydrogen.
Actually, after making this conversion to megawatt hours, it doesn’t seem as outrageous as I first thought. If there are 1000 kilowatt hours in a megawatt hour, this works out to be – what… 80 USD to 500 USD per megawatt hour?
80 USD per megawatt sounds much more plausible than the 8 USD per megawatt hour conversion I had made in my head). This is still quite a bit higher than regular variable output renewables, which I understand you can you might get for as low as 20-30 USD per megawatt hour, depending on where you are in the world.
Co-benefits from fuel cells – water for cooling
One thing I hadn’t considered is the chemical reaction used by most fuel cells.
When you use a fuel cell, one of the outputs is usually hydrogen (or some hydrogen carrying molecule, and another is air. An output is electricity, water and some other byproduct if you’re not using pure hydrogen.
Given that water is often needed for cooling, and the local consumption of water around datacenters is a now a common criticism, the idea of a data centre that is a source of potable water rather than a sink was a surprise. The quantities mentioned sound material as well. It’s likely that the is too pure to drink straightaway, but still this is something that was new to me.
Integrating into the grid
I touched on this before, but it’s worth going into more detail – one of the advantages listed is that because you’re not waiting for a grid connection, you can deploy faster – but also, even if you CAN get a grid connection, the sheer density of energy demand can really tax the grid, because it just doesn’t have the capacity to move that much electricity over the wires (it’s a bit like running out of bandwidth with networks).
For context, a big hyperscale datacentre might have solid 24/7 power draw in the tens of megawatts these days. For context, a single family house in the UK, averaged over the year is about 300 watts, so a single megawatt datacentre might have comparable power draw to 3,000 of these family homes. I explored these numbers in this podcast episode with Molly Webb, on the Environment Variables podcast.
Having the generation local to the consumption goes some way to addressing this issue, and moreoever, once you’re connected to the grid, there’s a chance to support the grid by either feeding spare power back in, or providing all kinds of ancillary services to help maintain the ‘quality’ of the power (i.e. helping keep the grid frequency stable and so on).
There are obvious open questions about getting power to the site – transferring power with pipelines requires you to have gas pipelines that can carry hydrogen (which has its own problems, as the molecule is harder to move around than typical fossil gas). If you don’t have a pipeline, you’ll back to relying on trucks or similar.
I host a podcast exploring the decarbonisation of digital services with the Green Software Foundation – it’s called Environment Variables. If this interested you, you might find it interesting too.
Peter’s Judge’s podcast, Zero Downtime is also worth a listen – learn a lot from it, as I usually work ‘higher up the stack’ at the software level.