EchoML unveils 'Replicate X': on-device multimodal inference chip for edge AI

Tech · 6 min read

EchoML unveiled Replicate X, an inference accelerator designed to run compact multimodal transformer models with optimized power efficiency. The chip supports quantized attention and sparse retrieval, allowing vision-plus-language tasks to run locally on devices like AR glasses and mobile robots.

Alongside the product launch, EchoML raised $60 million in a Series B led by Apex Capital, which the company plans to use for wafer contracts and partnerships with OS vendors. The firm also announced pilot programs with two AR hardware manufacturers planning to ship developer kits this fall.

EchoML positions Replicate X as a privacy-first solution: by keeping inference on-device, it removes the need to stream raw sensor data to the cloud. Developers get an SDK that exposes a high-level multimodal API and model conversion tools to adapt popular open models to the chip.

Analysts say the move addresses growing demand for local inference in safety-critical and latency-sensitive use cases, but success depends on software ecosystem adoption and competitive pricing against incumbents.