TL;DR:

• The Commons Science, Innovation and Technology Committee has opened a short inquiry into whether low-energy computing techniques can offset AI’s rising electricity draw.
• Committee chair Dame Chi Onwurah cited a forecast four-fold rise in UK data-centre electricity consumption by 2030, against current use of about 2.5% of national electricity.
• The inquiry focuses on neuromorphic photonics and silicon photonics — emerging techniques that combine photonic components with brain-inspired computing principles.

The deadline for written evidence is 14 May. Onwurah framed the inquiry as a clean-energy-targets question as much as a compute-infrastructure question: “This challenge is particularly pressing as we approach the government’s 2030 clean energy targets.”

Why the committee picked these techniques

Neuromorphic computing architectures process information in ways closer to biological neurons than to conventional von Neumann CPUs. Silicon photonics performs computation using light instead of electrical signals, reducing resistive losses. Combining them — neuromorphic photonics — is an active research area with potential for sharp reductions in the energy per inference for some workloads. The committee first looked at these techniques during its “Under the Microscope” exercise, where researchers pitched cutting-edge science for political attention.

The Sovereign AI crossover

The inquiry will examine “whether the UK has sovereign capabilities in this area” — a deliberate echo of the Sovereign AI Fund launched 24 hours earlier, which prioritises UK-controlled AI infrastructure. The two threads could converge quickly. Callosum, the Sovereign AI Fund’s first equity stake, works on making different chip types cooperate efficiently for AI training. Low-energy computing research is exactly the kind of long-horizon strategic asset that a state-backed vehicle could credibly back.

The realistic timeline

Onwurah’s final prompt to witnesses is practical: how long would low-energy computing take to implement, and when would its benefits be felt. That question matters because the UK’s 2030 clean-energy deadline is a harder constraint than any compute target. If neuromorphic photonics is a 10-year prospect, it does not solve this decade’s grid-load problem, and the committee will have to consider shorter-horizon levers — efficiency requirements in data-centre planning, nuclear baseload siting, and demand-side management via workload routing.

Looking forward

The committee’s findings will shape DSIT’s response to data-centre growth. Expect evidence from UK photonics groups (the University of Southampton, Imperial College and the UK’s compound semiconductor catapult), hyperscaler operators, and AI infrastructure firms backed by Sovereign AI. The political test is whether the inquiry can convert a niche research area into credible industrial policy before the next spending review.