A 2 MW absorption chiller consumes between 20 and 25 kW of electricity. A comparable mechanical chiller draws 500 kW or more. The difference is roughly a 90 percent cut in the electrical load needed to deliver the same cooling capacity. For data centers building around onsite generation, that single ratio rewrites the facility electrical plan.
The FM Industry analysis lays out the physics. Onsite generation from gas turbines, fuel cells, or reciprocating engines converts 35 to 50 percent of fuel input into electricity. The rest leaves as waste heat. That heat stream has been treated as a loss line on the energy balance sheet for decades. Combined cooling and power architecture turns it into a cooling asset.
The pressure behind this conversation is the same pressure reshaping every other corner of the cooling industry. Utility interconnect timelines have stretched from 24 months to 48 months in multiple US regions. Neocloud projects are stalling because the grid allocation cannot meet the compute schedule. Some forecasts put roughly one-third of new data centers operating as fully onsite-powered campuses by 2030. That shift is not driven by sustainability messaging. It is driven by interconnect queues.
Once a facility commits to onsite generation, the question of what to do with the thermal output is not optional. A 100 MW campus with gas turbine generation is rejecting something in the neighborhood of 100 to 120 MW of thermal energy continuously. If that heat is vented to the atmosphere, the operator is paying twice: once for the fuel that produced it, and again for the chillers that remove the heat from the IT load. If that heat is captured, it can drive absorption cooling at a fraction of the electrical cost.
The chemistry is not new. Absorption chillers use a working fluid pair, typically lithium bromide and water or ammonia and water, and a heat source in place of a mechanical compressor. The heat boils the refrigerant out of solution, the vapor condenses, the liquid evaporates in the cooling zone and absorbs thermal energy, and the cycle repeats. The only moving part of any size is a solution pump, which is why the electrical draw is so low compared to vapor-compression chillers with their motor-driven compressors.
The efficiency economics are specific to the heat source. Turbine exhaust at 400 to 500 degrees Celsius drives double-effect absorption chillers with coefficients of performance above 1.2. Fuel cell waste heat at 250 to 300 degrees Celsius runs single-effect units at COP near 0.7. Both beat the effective COP of an electric chiller when the electricity is being produced onsite at 35 to 45 percent generation efficiency.
Combined cooling and power is not for every facility. The sweet spot is a site that already has firm reasons to run onsite generation. Grid-constrained campuses, facilities in regions with unreliable utility power, sites that need to meet resiliency standards beyond what the local grid can provide, and increasingly, AI training clusters that cannot wait for interconnect studies. For all of those, the waste heat was already there. Absorption chillers turn it from a cost into a capacity play.
There are catches. Absorption chillers need more cooling tower capacity than mechanical chillers to reject the additional thermal load from the generator side. In water-constrained regions, that shifts the bottleneck from electricity to water, which means the architecture pairs best with dry-cooled or hybrid heat rejection. Retrofit is harder than greenfield because piping runs and tower placement are decided years before the chillers get specified.
The other direction this architecture runs is district heating. In Germany, Denmark, and parts of Sweden, regulations now require data centers to deliver recoverable waste heat to municipal heating networks. Onsite generation plus absorption cooling plus heat export is the full-stack answer, and it is already being built. The American conversation has not arrived there yet. It will, once the first state utility regulator credits a data center for delivering industrial waste heat to a neighboring manufacturing process and the economics get published.
The operators who treat onsite generation as a backup plan will miss this. The ones who treat it as the electrical architecture for the next decade will design cooling around the waste heat stream from day one. A 90 percent reduction in cooling electrical draw is not a refinement. It is a different thermodynamic footprint for the whole facility, and it opens power headroom that goes directly into GPU capacity. That is the math operators should be running on every campus entering siting review this year.