后摩智能

Houmo AI (后摩智能) is a Chinese Semiconductor company focused on AI chips built around an in-memory computing architecture, a design APProach that blends stoRAGe and computation to reduce data movement. Rather than offering a consumer app, the company supplies Silicon and hardware modules that let device makers and automotive manufACTurers run AI workloads locally, including large language models at the edge.

What it does

The company develops AI Chips, system-on-modules (SoM) and AI accelerator cards aimed at multi-scenario intelligence. Its flagship M-series parts are positioned for edge and endpoint deployment, while a separate automotive line targets advanced driver-assistance and Intelligent driving platforms. The core proposition is combining high compute density with modest power draw so that capable models can run on devices rather than in distant data centers.

Key capabilities

• In-memory computing Architecture: the company states this design improves data processing efficiency by cutting the separation between mEMOry and compute.
• Compute and power balance: the M30 chip is described as reaching up to 100 TOPS with a typical power consumption of about 12 W, though configurations and measured figures may vary.
• Edge LLM support: the hardware is designed to host Large Language Models locally in real time; the company has demonstrated a model with over 7 billion parameters running on the edge in collaboration with China Mobile.
• Automotive silicon: the Hongtu H30 driving chip is cited at 256 TOPS of physical compute for intelligent driving workloads.
• Deployment options: compute modules and accelerator cards are offered to simplify integration into products.
• Functional safety: the company reports certification under the ISO 26262:2018 ASIL D process, the highest automotive safety integrity level.

Who it is for

Houmo's products are relevant to hardware engineers and product teams building intelligent vehicles, robotics, smart cameras, industrial equipment and other edge devices that need on-device inference. Automotive tier-one suppliers and OEMs evaluating driving compute platforms are a core audience, as are System Integrators assembling edge servers or gateway devices with accelerator cards.

How to get started

There is no self-serve signup in the usual software sense. Access typically begins through a business inquiry: interested teams contact the company to discuss chip or module selection, request technical documentation, and obtain development boards or evaluation kits. Software support, model conveRSIon tooling and pricing are Generally ARRanged through direct engagement, and availability outsIDE China may be liMITed.

Pricing and access

Chip, module and accelerator pricing is not publicly listed and is likely quoted per project, volume and configuration; treat any figures as subject to negotiation and change. Because the company is China-based, some visitors outside the region may need a VPN to reach its website, and export, compliance or logistics considerations could apply. Details such as supported Frameworks, memory configurations and roadmap dates may vary, so confirm specifications directly with the vendor before designing a product around them.