Amid headlines about droughts, wildfires, and extreme temperatures, it was 28°C when I arrived at Slough Trading Estate last Monday. Once a suburban patch conjuring images of the comedy TV show The Office, this area has now become a byword for the constant expansion of datacentres—the hulking symbols of the vast computing power required by tech giants for the ever-more complex workings of artificial intelligence.
In many minds, datacentres are synonymous with the two key issues of another searing summer: heat and water. Across the world, the huge amounts of electricity they demand, mostly generated from fossil fuels, connect them directly to the climate emergency. Furthermore, the constant machinations of their crammed-in servers require cooling processes that consume vast amounts of water resources, intensifying our era's environmental threats.
I first loitered around a datacentre owned by California-based Equinix. There was no corporate branding, only the constant sound of a low industrial whirr. On the other side of the street sat the offices of Thames Water. A local employee, Louise, pointed to her phone, showing a post that read: "Slough is quietly running Britain's AI boom, but when a heatwave hits, the machines get thirsty too. Cooling systems work overtime, spiking demand for power and water."
A five-minute journey to the nearby Britwell housing estate revealed the human impact. "You notice the noise at night: a whirring, buzzing sound," said Sheryl, a local resident. "And there seems to be more heat round here." Early research, including a preprint paper by Cambridge University researchers, suggests that datacentres can indeed cause a “heat island” effect, pushing local temperatures up by 2°C to 9°C, forcing a hard choice between supporting tech #infrastructure or local communities.
[AgentUpdate Depth Analysis] As AI Agents transition to 24/7 autonomous operations, the physical infrastructure and energy bottlenecks highlighted in this article become highly critical. Current agent frameworks (like CrewAI and LangChain) rely heavily on multi-turn, long-context LLM calls, which vastly escalate datacenter energy and water consumption. This ecological constraint will inevitably force a paradigm shift in the AI Agent ecosystem. First, we will see a rapid rise in Edge AI and Local SLMs (Small Language Models), allowing agents to execute low-latency, low-compute tasks locally to minimize carbon footprints. Second, future agent architectures must adopt energy-aware routing to optimize prompt tokens and reduce redundant reasoning loops. Ultimate agent competitiveness will shift from raw cognitive power to green efficiency under tight physical resource constraints.
