There’s a growing narrative that AI is bad for the planet. Training large models consumes vast amounts of energy. Data centres are expanding rapidly. Demand for compute is rising faster than most infrastructure can keep up.
But the reality is more nuanced. AI itself is not the problem. The question is how we choose to run it. Right now, much of the world’s digital infrastructure still relies on a centralised model. Large hyperscale data centres sit far away from where data is created and used. Every request, every interaction, every inference travels back and forth across networks. That movement has a cost. Not just in latency, but in energy.
Every kilometre power travels through electricity grids results in energy loss due to the laws of physics. Even across more efficient systems, moving data at scale requires power. Multiply that across billions of interactions per day, and the environmental impact becomes significant.
This is where the conversation needs to shift. Not away from AI, but towards the infrastructure that supports it.
A more efficient model is emerging. One that brings compute closer to where it is needed. Distributed edge environments allow data to be processed locally, near devices, users, and operations. Instead of sending everything back to a central location, workloads can run at or near the source. The effect is simple but powerful. Data travels less distance. Networks carry less load. Systems respond faster.
And importantly, less energy is wasted moving information around. This is not about replacing hyperscale. Large data centres still play a critical role, particularly for training models and handling global workloads. But not every task needs to make that journey. Many AI use cases are inherently local.
A factory analysing sensor data in real time does not need to send every signal across the country. A hospital running diagnostic tools cannot afford latency or dependency on distant infrastructure. A transport network optimising flows in real time benefits from immediate, local processing.
In these scenarios, proximity matters. Not just for performance, but for efficiency.
The physics behind this is straightforward. Electrons moving through copper lose energy over distance. Photons travelling through fibre can carry far more data with far less loss. When we reduce the distance electrons need to travel and rely more on efficient optical networks for long haul connections, the system as a whole becomes more efficient.
Edge computing does exactly that. It keeps local workloads local, while using fibre to connect distributed sites where needed. The result is a more balanced system. One that reduces unnecessary data movement without compromising scale.
However, this is not a silver bullet. Distributed infrastructure also comes with trade-offs. More sites mean more equipment, more deployment, and more operational complexity. If not designed properly, the benefits can be offset. Hyperscale environments remain highly efficient at scale, particularly when it comes to energy usage per unit of compute.
So the future is not a choice between centralised or distributed. It is a combination of both. Training may remain centralised. Inference, especially where latency and responsiveness matter, can increasingly move closer to the edge. Data can be processed where it is generated, and only what is necessary can be sent further upstream.
This hybrid model is where meaningful efficiency gains can be realised. As AI adoption accelerates, the environmental conversation needs to mature alongside it. Focusing only on the scale of compute risks missing the bigger picture. The architecture of the system matters just as much as the workloads running on it.
Reducing unnecessary data movement. Using infrastructure more intelligently. Aligning compute with demand at a local level. These are the decisions that will shape whether AI becomes an unsustainable burden or a manageable part of the digital economy.
AI does not have to come at the expense of the planet.
But it does require a shift in how we think about infrastructure. Not as something distant and invisible, but as a critical lever in building a more efficient, more resilient digital future.