AI
Furthermore, data centres can choose to adapt their actual computing workloads to relieve pressure on the grid, or rely entirely on onsite microgrids during peak events.
“ The way you design your data centre can essentially ensure you actually don’ t have to reduce your workload at all, but that is ultimately a business decision, right?” Anuja explains.“ You don’ t have to reduce your workload if you have an onsite portion of generation power that you can then use to power periods when grid power is not available.”
By categorising workloads by priority at the individual rack level, operators can seamlessly orchestrate energy usage. However, for critical infrastructure where workload flexibility is not an option, the focus turns to onsite alternatives. This route requires patience, particularly when dealing with clean baseload technologies like nuclear energy.
As Anuja notes,“ long development cycles, like all the permitting applications and approvals and environmental reviews – compared to a typical, say, renewable project development – it is a little bit longer”.
Europe’ s heat laws and ultra-efficient partnerships In Europe, the sustainability push is heavily driven by legislation. Germany, for instance, has introduced strict energy efficiency laws requiring data centres to reuse at least 20 % of their waste heat by 2028, often by routing it into local district heating networks. But, practical bottlenecks remain. In
major digital hubs like Frankfurt, there are not enough residential or commercial buildings close to the data centres to absorb the immense volume of thermal energy produced. This has sparked interest in converting waste heat directly back into electricity for self-consumption, though the physics of this process present distinct hurdles.
“ In data centres, there is high-quality, consistent heat as well as low-quality heat, which is inconsistent,” Mandar explains.“ The problem is that data centre workloads are very dynamic, so trying to have them produce meaningful heat that you can use for power generation is somewhat inefficient.
96 August 2026