AI is no longer a distant dream. It’s embedded in daily life – from smart devices to decision-making networks – and it’s only getting bigger. But with that exponential growth comes a serious challenge: how do we power it all without blowing our energy goals?
That was the crux my chat with Lieven Levrau’s chat on this week’s podcast. Representing Nokia and the IOWN Global Forum, Lieven broke down how AI’s surging energy needs could clash with the push for net zero unless smarter infrastructure is built – fast.
Lieven laid it out simply: AI growth isn’t just about your phone chatting back with Gemini. Behind the scenes, it’s reshaping everything from traffic control and healthcare to how we run energy grids. But it all needs serious computing power, mostly from GPUs – the brainpower behind AI models. These GPUs are expensive, power-hungry and need careful placement, ideally somewhere with cheap, green electricity and good fibre connections.
Enter photonics. Unlike traditional copper networks, optical systems using glass and light (not electrons) offer higher capacity, lower latency and far better energy efficiency. “Optical transmission is simply less power-hungry,” said Lieven. “And photonic networks allow you to offload traffic from traditional systems, cutting power demand.”
These hollow-core fibre networks – think of them like express trains bypassing local stops – are already being built globally. They’re faster, more efficient and play a key role in streamlining data traffic between GPUs and the AI users at the edge of the network.
But Lieven didn’t downplay the challenge. “We are going to have a ramp-up of higher power-consuming AI,” he admitted. “But as it matures, optimisation will come.” Smarter routing, photonic hardware and AI-powered system design itself could all rein in energy use – if done properly.
And the final twist? AI might even help solve its own energy problem. As Lieven put it, “We should use AI to figure out where AI should be deployed.”
Listen in to the chat now and explore how our future might look! Please subscribe too.
