The brief

Analyst view

• Raised $14.1M seed (Dec 2023) led by Kindred Ventures, with Aidan Gomez of Cohere and others.
• Signed $75M letter of intent with U.S. Department of Commerce for CHIPS R&D funding.
• Thermodynamic computing targets 10,000x energy efficiency for AI sampling.

Technical approach

As reported

Pbits

Probabilistic bits from thermal noise

CMOS transistors generate thermal noise. Extropic harnesses this into programmable bits that sample distributions in silicon.

Energy

Sub-watt operation

Z1 consumes <1 watt at 50M samples/sec. Local processing on-chip avoids Von Neumann bottleneck and memory shuttle costs versus conventional compute.

Scaling

Billion-pbit clusters

Z1 cards host 4M+ pbits each. Multi-card clusters reach 1B pbits, the scale where sampling-heavy AI (images, video, robotics) sees practical speedups.