Startup plugs AI datacenters into biogas-powered energy
Sidestepping the grid led to 44% cheaper electricity and 70% fewer emissions, CEO says
A UK datacenter startup realized it could have to wait until the late 2030s for power grid connection dates, and has instead turned to modular facilities located at the site of renewable energy sources.
Chainergy, which was founded in 2022, is putting AI datacenter infrastructure inside enclosures that are similar to shipping containers. These modules can then be deployed on the same site or adjacent to a biogas generator facility, directly drawing power from it at a lower cost than being connected to the grid, the company claims.
"We've largely come from the energy sector, and we recognized the impacts of the lagging infrastructure and that the energy grid connecting projects was taking a very long time," co-founder and CEO James Behan told The Register.
This is a widespread problem, but particularly in the UK, where developers have complained about available energy supplies as a major impediment to datacenter projects.
"We developed various renewable energy projects in the past, and actually seeing connection dates get further and further away was becoming a problem. So we would have grid connection offers for the late 2030s, which obviously made things very challenging," Behan said.
A "grid connection offer," if you're wondering, is something those seeking to connect to the National Grid must obtain from Britain's high-voltage electric power transmission network before they can put in orders for "works to be undertaken," or even find out pricing, terms, and other important info.
That prompted the startup to look at how to integrate at the same location as where energy is generated, and develop modular datacenter infrastructure that would be simpler to deploy there.
Chainergy says it discovered that some generator sites had excess capacity they couldn't sell to the grid, making it a resource that was just waiting to be used.
"The reason they weren't able to sell it to the grid is that they had a restriction on how much power they could actually export, again down to infrastructure constraints. So we developed a small pilot project just outside of York, just to prove our concept on a very small scale," Behan said.
This was a site producing biogas, where food waste or agricultural waste is fed into a vat and broken down by microorganisms in a process of anaerobic digestion. The resulting biomethane can then be used to power generators to create electricity, which can be sold to the grid.
According to the National Grid, 109 biogas plants are currently in operation across the UK, with the first one to feed power into the national transmission system getting connected in 2020.
"In this instance, they had 15 percent more energy they can generate than they can actually sell to the grid. So we took that excess electricity to our small modular datacenter," Behan explained.
He claims that by eliminating grid costs and utility margins this way, Chainergy is able to get power 44 percent cheaper and emit 70 percent fewer CO2 emissions than a standard bit barn. That pilot project ran throughout the second half of 2023 up until the end of Q3 last year.
The company will be putting together its first commercial-scale deployment this year, and aims to initially offer "GPU as a service" to customers. It hopes to target two kinds of customer: one is what Behan calls "Neo clouds" offering pay-as-you-go compute, and the other is privately connected corporate clients.
"There's a lot of cloud repatriation going on right now," Behan claims. "Maybe some of the hyperscalers haven't really provided what the corporates were looking for in terms of costs and at the same time, there's concerns around data privacy, sovereignty, etc. We're saying these facilities could become privately connected to a local corporate, essentially being an extension of their on-premises datacenter."
While some corporates may be looking to repatriate workloads on-premises again, they don't necessarily have the energy capacity or cooling infrastructure to do energy intensive workloads like AI, he asserts.
- Mega UK datacenter greenlit, but we still don't know who's moving in
- Microsoft throws more cash at its carbon guilt by replanting Brazilian rainforest
- Real datacenter emissions are a dirty secret
- Looming energy crunch makes future uncertain for datacenters
This chimes with what Canalys chief analyst Alastair Edwards stated last year – that the public cloud is too costly for AI workloads at scale, but few organizations want to build their own datacenter anymore.
IDC also noted that cloud repatriation is a trend for specific workloads, which can include AI, but it is not a wholesale mass migration away from the public cloud.
Chainergy's first commercial project will be relatively small-scale at 400 kW of infrastructure, and it is going to be inside a barn, Behan told us, "so it's going to be a bit more than a container in a field."
The company will also have engineers on site, and is aiming to be "at a minimum a tier two datacenter." Because it is a smaller project, this will be built with standard rear-door cooling, but future projects are expected to feature liquid cooling as standard, with direct-to-chip the favored technology.
We asked Chainergy if it was considering other renewable energy sites such as those based on solar or wind, but Behan told us these are currently too intermittent for its requirements. Biogas can deliver 95 percent plus uptime, he claimed, which means that it only needs to fall back on biofuel-powered backup generators for a small period of time.
Chainergy's business model means it may find its growth limited by the amount of excess power available at the biogas sites it chooses to locate alongside, and it may need to draw some additional energy from the grid like other operators. Behan also concedes that not all energy sites will meet its criteria, which includes having fiber network connectivity not too far away.
However, with demand for GPU-powered AI servers high, and the difficulty in expanding datacenters or building new ones because of power constraints, we could see more niche startups such as this, taking advantage of available energy sources wherever they are. ®