Summary
- Era4 is pursuing a planned AI data centre at Shelford Farm Estate near Canterbury, using private-wire landfill gas generation.
- The proposed facility would sit on a waste-management site and use closed-loop liquid cooling with adiabatic assistance.
- The scheme brings grid avoidance, brownfield land, methane use, local opposition, and AI planning politics into one UK development.
Era4’s proposed AI data centre at Broad Oak near Canterbury is drawing attention to a new edge of the UK capacity debate: using landfill gas, brownfield land, and private-wire power to build compute without waiting for a conventional grid connection.
The planned Shelford data centre would be developed at the Shelford Farm Estate, a landfill and waste-management site associated with Valencia Waste Management. Planning-related trade coverage describes a facility of around 13,000 sq ft, or 1,210 sq m, on land previously used for composting operations.
The proposed site would be connected by private wire to the Valencia Energy Centre gas generation station, which is fed by methane from the landfill. The project is also reported to include rooftop solar, closed-loop liquid cooling, and adiabatic assistance. Era4, formerly Carbon3.ai, describes itself as a developer of lower-carbon AI infrastructure using brownfield sites.
Power strategy reaches the planning file
The Broad Oak proposal is small compared with hyperscale campuses, but it shows how UK developers are trying to work around grid constraints. Large-load connection queues have become a major barrier to AI infrastructure, and sites with private-wire or behind-the-meter power can move through the market differently from projects dependent on new public network capacity.
Landfill gas brings a specific energy argument. Methane from decomposing waste can be captured and used for electricity generation, reducing direct emissions from landfill and creating a local power source. It is still combustion-based generation, and its climate performance depends on what would otherwise happen to the methane, how the power is generated, how long the gas supply remains productive, and how the project accounts for residual emissions.
The appeal for a data centre is firmness. Solar and wind can support on-site generation and procurement claims, but AI workloads generally need stable power unless the compute is designed to flex around generation. Landfill gas can offer a more controllable local energy source, although gas output changes as a landfill ages. Long-term power modelling will be central to the project’s credibility.
Cooling, water, and consent
The proposed use of closed-loop liquid cooling with adiabatic assistance puts the development inside the wider debate over water and heat. AI data centres are associated with high power density and heavy cooling demand. Closed-loop systems can reduce continuous water use compared with some evaporative designs, but adiabatic systems can still require water under certain operating conditions. Planning authorities and residents are increasingly asking for clear cooling, water, noise, and heat-rejection data.
Local opposition has already formed around the proposal. A petition against the project argues that residents need clearer evidence on environmental impact and the credibility of sustainability claims. That response is becoming more common in data centre planning. National policy may describe AI infrastructure as strategic, while local communities focus on land use, traffic, power generation, noise, emissions, visual impact, and limited direct employment.
The project also shows how contested “brownfield” can become. Reusing a waste-management or industrial site can reduce pressure on greenfield land and place the data centre near existing energy infrastructure. A brownfield location does not automatically make a scheme low impact. The planning test still has to cover previous land use, environmental condition, local setting, access, noise, cooling, emissions, security, and long-term operations.
Era4’s model appears to rely on multiple UK sites tied to local or alternative power sources. That is a rational response to connection delays, but it could also create more fragmented planning disputes. Smaller AI facilities embedded near energy or waste assets may avoid some hyperscale grid constraints, yet they still need detailed scrutiny over resilience, emissions, water, cooling, noise, and community benefit.
The UK policy challenge is moving quickly. Ministers want AI capacity, but public acceptance will depend on whether projects show credible infrastructure design. Behind-the-meter generation can reduce pressure on the grid, but it shifts attention towards fuel source, local air quality, backup power, cooling, planning conditions, and lifecycle emissions.
Broad Oak is therefore more than a small local planning file. It is an early test of whether alternative-powered AI data centres can make a convincing case in the UK planning system, and whether legacy waste infrastructure can be recast as a platform for digital capacity.
