QuantumLeap raises $300M to develop edge AI ASIC for mixed-precision inference

Tech · 5 min read

QuantumLeap’s architecture focuses on dynamic precision scaling and temporal sparsity exploitation to reduce energy per inference. The company targets edge devices that need continuous multimodal inference, such as smart cameras and robots.

Alongside the funding, QuantumLeap revealed product details: a 16 TOPS peak mixed-precision engine with programmable microkernels and a runtime that adapts precision per layer. Early silicon samples will be available to partners later this year.

Investors include a major foundry and cloud vendor strategic funding. Proceeds will finance tape-out, software toolchain development, and a developer program to port popular models and optimize latency and power profiles.