Polaris Systems unveils EdgeLLM M1: a compact AI module for on-device language models

Tech · 5 min read

Polaris Systems’ EdgeLLM M1 promises to bring large language model capabilities to edge devices without continuous cloud connectivity. The module combines a custom NPU, 32GB LPDDR memory, and optimized quantization stacks to run mid-sized LLMs with sub-100ms latency for short queries.

The product launch coincided with a $60 million Series C led by Harborstone Partners and a strategic investment from a major consumer electronics OEM. Polaris said the capital will expand production lines and accelerate partnerships with chip foundries to bring modules to consumer devices, industrial controllers, and automotive infotainment systems.

Developers get a turnkey SDK for on-device fine-tuning with privacy-preserving federated learning options, plus integrations for ROS and popular mobile frameworks. Polaris claims a target of 10x energy savings compared with cloud inference for certain on-device customer-service and voice assistant workloads.

While EdgeLLM M1 is aimed at enterprises initially, Polaris envisions a broader consumer rollout via device makers. The key proof will be in real-world deployments where connectivity is sparse or privacy is paramount, such as in healthcare, defense, and in-vehicle assistants.