Researchers at the University of Cambridge developed a hafnium oxide-based memristor that mimics brain neurons, storing and processing information in the same place. The device demonstrated exceptional stability across millions of cycles. The neuromorphic computing approach could reduce AI hardware energy consumption by up to 70% compared to conventional chips that constantly shuttle data between memory and processing units.
Why it matters
Energy costs are the fastest-growing line item in AI infrastructure budgets. A 70% reduction in hardware energy use, if commercialized, would fundamentally change the economics of running AI at enterprise scale and ease data center power constraints.
What to do
This is still research-stage technology, but CIOs should track neuromorphic computing developments as part of long-term infrastructure planning. The energy savings potential could reshape AI datacenter economics within 5-7 years.