CoreMLab Raises $95M Series C to Commercialize On-Device LLM Acceleration

Tech · 5 min read

CoreMLab, a developer of neural acceleration silicon and runtime, secured $95 million from a consortium led by Horizon Foundry. The round will support production runs of their second-generation NPU and expansion of their on-device model marketplace.

The company pairs custom hardware with an optimized compiler and quantization toolchain to run trimmed-down language models with high responsiveness and minimal battery drain. CoreMLab's CEO says the goal is to enable local inference for personal assistants, offline summarization, and private search.

CoreMLab also introduced StudioEdge, a developer suite that helps teams compress and adapt models for specific device constraints without significant accuracy loss. The startup is working with several Android OEMs for pilot programs targeting privacy-focused assistants and translator features.

Investors view on-device LLM acceleration as a strategic counterpoint to cloud-first models, reducing latency and data exposure. With the new capital, CoreMLab will expand hardware partnerships and certify models for its secure runtime.