Northwestern engineers printed artificial neurons that successfully communicated with living brain cells, the McCormick School announced earlier this month. [1] The result reads as a neuroscience advance. It is also a power-consumption argument.

Mark Hersam, the materials scientist on the project, made the argument explicit: "The brain is five orders of magnitude more energy efficient than a digital computer." [1] An order of magnitude is a factor of ten. Five of them is a factor of one hundred thousand. AI training runs and inference clusters are now negotiating gigawatts and rural water rights; the biological substrate they imitate runs on roughly twenty watts.

News-Medical's writeup adds the technical detail: the artificial neurons are organic, not silicon, and operate at low voltages compatible with cellular signaling. [2] That is what allows the activation of living cells. It is also what gestures, faintly, at a different compute architecture.

This is a foundational paper, not a product. No data center is being replaced by a petri dish. But the question that AI-infrastructure investors keep asking — whether the energy curve bends — has had an under-quoted answer in biology for the entire history of the field. Northwestern's contribution is to make that answer testable in materials. [3]

-- KENJI NAKAMURA, Tokyo