Follow the Carbon, Data Centres, The Digital Economy Still Runs on Physical Energy

Artificial intelligence may operate in the digital world, but the infrastructure supporting it is intensely physical. Every query, model-training run and cloud service ultimately requires electricity. That electricity becomes computation, heat, cooling demand and, depending on the cooling system and electricity source, water consumption and carbon emissions. As data-centre capacity expands, perhaps we should stop looking at the data centre simply as an electricity consumer and start looking at the complete physical system. Follow the Energy The conventional boundary is often: Grid → Data Centre → Computing But the physical pathway is considerably longer: Primary energy → electricity generation → transmission → data centre → IT equipment → heat → cooling → surroundings At every stage there are conversion losses, infrastructure requirements and opportunities for energy recovery. This is why renewable electricity is important, but the percentage of renewable electricity contracted by a data centre does not by itself describe the performance of the complete energy system. We should also ask: How much primary energy was required? When was the electricity generated relative to when it was consumed? What firming, storage and network infrastructure was required? What happens to the heat produced by the servers? How much water is consumed? And what happens when renewable generation is unavailable? Now Follow the Carbon Carbon accounting requires a parallel pathway. Where electricity is generated from fossil fuels, the carbon pathway may be: Geological carbon → extraction → fuel → power generation → CO₂ → atmosphere Renewable generation changes that pathway substantially because there is no continuous fossil-carbon feedstock entering the generation process. But when gas engines, turbines or other combustion technologies provide backup or firm power, we should continue following the carbon. Was the carbon newly extracted from geological reserves? Was the CO₂ captured? Was it permanently stored? Was the carbon recovered and reused? Or was it released to the atmosphere? These are materially different carbon pathways even when they ultimately deliver the same unit of electricity to a server. Net Zero, Decarbonisation and Defossilisation Are Not Identical These terms are often used interchangeably, but they describe different questions. Net zero asks about the balance between emissions and removals within a defined boundary. Decarbonisation measures the reduction of emissions associated with an activity or economy. Defossilisation asks a more fundamental material-flow question: How much newly extracted fossil carbon continues to enter the system? A data centre could therefore reduce its carbon intensity substantially while still depending indirectly on fossil carbon during particular hours. Conversely, a future energy system could potentially use carbon-containing molecules without continually extracting new geological carbon, provided the carbon itself were recovered and managed within a sufficiently controlled cycle. The distinction is not semantic. It changes what we measure. The Data Centre Could Become Part of the Energy System There is another opportunity. Data centres do not necessarily have to remain passive loads. Integrated correctly, they could combine: Firm generation + renewable electricity + storage + flexible computing loads + thermal storage + heat recovery + cooling + water management Some workloads can potentially respond to electricity-system conditions. Cooling systems can incorporate thermal inertia or storage. Waste heat may have productive uses where suitable heat sinks exist. On-site generation can improve resilience and reduce dependence on constrained networks, although its complete fuel and carbon pathway must still be accounted for. The design question therefore becomes larger than: How do we supply enough electricity to the data centre? It becomes: How do we design the data centre and its energy infrastructure as one integrated system? A Better Measure of Sustainable Compute PUE remains useful for measuring data-centre energy efficiency, but the transition may eventually require broader system metrics. Alongside PUE, we may need to understand: • primary energy consumed per unit of useful computation; • carbon introduced and released; • water consumed and recovered; • heat rejected and productively reused; • firm capacity required from the wider electricity system; • and flexibility supplied back to that system. This is particularly important as AI infrastructure grows from individual facilities into energy-intensive industrial clusters. The digital economy cannot escape thermodynamics. Every computation requires physical energy. Every energy conversion has consequences. And every carbon atom has an origin and a destination. So when assessing whether the next generation of data centres is truly sustainable, perhaps we should look beyond the server rack and ask two simple questions: Where did the energy come from? Where did the carbon go? Follow the Energy. Follow the Carbon. #FollowTheCarbon #DataCentres #ArtificialIntelligence #EnergyTransition #Decarbonisation #Defossilisation #EnergyEfficiency #SustainableCompute #NetZero