UK MPs hear neuromorphic chips: greener brain-inspired computing, not datacenter replacement

UK MPs got a reality check this week on neuromorphic computing. University of York professor Martin Trefzer told the House of Commons Science, Innovation and Technology Committee that brain-inspired systems could improve efficiency, because they borrow a key idea from biology: memory and processing can be integrated as one system rather than split between separate components. In plain English, the “data movement” problem matters, and the brain teaches a lesson about keeping information closer to where it’s used.

That point lands hard when you look at what AI is doing to electricity demand. Trefzer referenced analysis from last year showing AI as the biggest driver pushing global datacenter electricity use to more than double by 2030 to around 945 terawatt-hours (TWh), slightly more than the entire electricity consumption of Japan. The argument is not that neuromorphic computing magically fixes the whole grid overnight. It is that the architecture of today’s computing, optimized for moving data around separate memory and compute units, is part of why AI power draw keeps ballooning.

Trefzer also made a second, equally important clarification for anyone hoping for a clean replacement story. The brain, he said, “is not a rigid computer that is kind of clocked in a digital system.” That is the barrier for adoption. Neuromorphic computing is still experimental, and the model’s complexity makes it unlikely to “prove its worth as a replacement for mature computing systems” soon. In committee testimony, he framed it as an apples-to-oranges comparison too many people jump to: neuromorphic systems are often pitched against a mature technology like LLMs running in datacenters, but those datacenters also come with energy and sustainability problems.

So what is the path that actually gets results? According to Trefzer, the only likely short-term impact is through specific applications that run alongside conventional computing, not through a full swap. He gave a wearable example: a hearing aid. Today, such devices are built on a digital substrate. Models are trained offline, but you can imagine pushing more functionality into a neuromorphic substrate that is sensitive to sound, with modalities that allow it to operate more in a brain-inspired computational manner. The key claim here is about moving tasks toward the sensor and away from energy-hungry digital processing. He said this is “where there is significant potential to be much more energy efficient, by orders of magnitude.”

That is a powerful statement, but it is not an automatic green light. University of Manchester physics professor Caterina Doglioni, also speaking to the committee, argued that the efficiency benefits need to be offset against the energy and carbon cost of building more devices on the edge. Her point is basically lifecycle accounting: yes, you might reduce compute in datacenters and save energy by doing more at the device level, but manufacturing extra hardware has its own footprint. Doglioni said there could be a threshold, a break-even point, where the new approach ends up improving environmental sustainability overall, but that it needs “the studies.” She also used a blunt framing: she “hates to be the person that breaks it,” but decision-makers have to consider cost and environmental impact, not just the performance narrative.

Put together, the committee discussion sketches a competitive and governance-friendly story for executives: neuromorphic is less about replacing the current AI stack and more about selectively redesigning where and how work is done. Datacenters still sit at the center of today’s LLM deployment because they are mature, scalable, and already wired into how models are trained and served. But if global electricity growth from AI continues on the trajectory described to MPs, the pressure to find energy-efficient architectural approaches will only intensify. That makes “hybrid integration” the likely battleground: smart combinations of conventional infrastructure with specialized neuromorphic substrates for tasks where energy savings are plausibly largest.

For boards and leadership teams in adjacent sectors, this is not just a research update. It is a signal about how the efficiency conversation is shifting from software knobs to systems design, including the edge. Procurement and sustainability teams will still care about energy dashboards and reporting frameworks, but they will increasingly ask harder questions about architecture: Where is data movement happening? What portion of inference could be shifted to sensors? What are the lifecycle carbon costs of adding new device complexity? And what threshold triggers a net sustainability gain?

Neuromorphic computing may one day behave more like a “one system” model closer to biology, but the near-term reality in this testimony is more pragmatic. Expect experiments, targeted pilots, and hybrid deployments that aim to cut energy use in specific scenarios like audio sensing. For decision-makers, the strategic stake is clear: waiting for a total replacement of datacenter AI is probably the wrong timeline, but ignoring energy as a systems constraint is also risky when projected electricity demand rises toward 2030 levels like those cited.