Tensordyne targets AI inference market with logarithmic math and Juniper-derived rack architecture

Tensordyne emerged from stealth with a chip in production at TSMC using logarithmic math to replace power-hungry multipliers with smaller adders. A 72-chip inference pod fits in 13 rack units, draws 30 kilowatts versus 150 kilowatts for comparable Nvidia systems, and delivers under one microsecond chip-to-chip latency.