Cobalt Neural raises $160M to commercialize on-device LLMs

AI · 6 min read

Cobalt Neural announced a $160 million round led by SoftBank and Lightspeed to speed development of on-device LLMs. The startup released Cobalt Edge, a compilation and runtime toolchain that targets mobile NPUs, dedicated inference chips, and emerging browser-based WebNN backends.

Cobalt Edge includes model quantization, weight pruning, and token scheduling tuned for battery-sensitive devices. The company demonstrated a local assistant prototype running efficient retrieval-augmented generation for search and note-taking with sub-400ms response times on flagship phones.

Cobalt Neural plans to partner with OEMs and chipset vendors to bring certified models to consumer devices and to offer enterprise SDKs for private on-device processing. The funding will expand the engineering team and build a certification program for performance and privacy.