The brief

Analyst view

• CL1 uses 850–1000 W for 30 units, versus multi-megawatt traditional AI data centres.
• Melbourne data centre hosts 120 CL1 units and attracts global researchers.
• Python API lets researchers program living neurons for research.

Technical approach

As reported

Biological substrate

Living neurons on silicon

Human neurons from stem cells are cultured on silicon with 59 electrodes for input.

Life support

Bioreactors maintain the neurons

Neurons are kept alive with continuous nutrient flow and waste removal. A 30-unit rack consumes 850–1000 watts, far less than traditional GPU data centres.

Programming interface

Python API for neurons

Researchers deploy code in Python to train neurons for problem solving.

Energy efficiency

Sub-millisecond latency and low power

The CL1 achieves latency below 1 millisecond, compared to 5 milliseconds for earlier DishBrain systems, while consuming a fraction of GPU power.

Show all 4