Houmo.AI's Series B Funds the Work Behind Edge AI Chips

The money behind Houmo.AI has two engineering calendars to support. Its M50 chip is already being designed into customers' devices, while the Chinese semiconductor company develops the architectures intended to follow it. A reported Series B worth several billion yuan puts current delivery and future research inside the same financing story.

Securities Times reported on November 4, 2026 that Houmo.AI had recently completed the round, with industrial capital, state-backed capital and international investors participating alongside returning shareholders. The report gives neither an exact amount nor named institutions. For device manufacturers considering local AI, the consequential part is the work those investors are financing between a processor specification and a product somebody can use.

What Houmo.AI's Series B will fund

The proceeds are intended to expand M50 delivery, advance M6X and next-generation 3D compute-in-memory development, and strengthen the software toolchain and ecosystem, according to November 5 reporting from STAR Market Daily. The amount remains an imprecise several billion RMB; converting it into a specific dollar figure would give the disclosure an accuracy it never had. The reporting date also differs from the undisclosed day on which the financing actually closed.

Founded in 2020, Houmo.AI builds chips and supporting hardware for large-model inference on devices and at the network edge. Its current company information identifies a Beijing headquarters and offices in Nanjing, Shanghai, Shenzhen and Hefei. The business addresses agent computers, mobile terminals, robots and edge inference systems, giving its engineering team several kinds of device constraints to work within.

Founder and CEO Qiang Wu remains the company's central public technical and commercial voice. Its October partner-conference account also identifies Xiaoxu Xin, or 信晓旭, as co-founder and VP of Product. The Series B's institutional participants have not been individually disclosed, so earlier investors cannot responsibly be promoted into this round's syndicate. The financing supports a company with an existing product family, while leaving valuation, cumulative capital raised and the precise ownership changes outside the public account reviewed here.

Why local AI changes the chip purchase

Houmo.AI's compute-in-memory approach places computation within or closer to memory to reduce data movement. Its SRAM-based architecture and developing 3D approach pursue the relationship between processing capability and the information that has to reach it. For a manufacturer, that relationship becomes practical through power consumption, memory capacity, heat and the model the finished machine is expected to run.

The company's M50 product specifications list 160 TOPS of INT8 compute, typical power consumption of 10W and configurations with 12GB to 48GB of LPDDR5/5X memory. Those are vendor specifications rather than independently reproduced benchmarks. A device built around the chip still has a complete system to power and cool, making the surrounding engineering part of the purchase decision.

Houmo.AI's account of the Great Wall N90 Pro collaboration describes joint optimization involving chip and operating-system suppliers. That is a revealing commercial detail: choosing an accelerator creates work across organizational boundaries. The laptop maker needs its local AI functions to coexist with the ordinary expectations attached to a portable computer, and the supplier needs to help make that combination deliverable.

The relevant competition therefore extends into the customer's engineering schedule. A processor supplier can make adoption easier through available modules, supported models and usable tools. Those choices influence how much specialist work a manufacturer must absorb before it can put an AI feature in front of a buyer, even when the underlying architecture looks compelling on paper.

Design wins bring a software obligation

In its October 20 partner-conference account, Houmo.AI said M50 had secured more than 150 product design wins and served over 100 customers. These are company-reported indicators of adoption. They describe engagement with manufacturers, while leaving shipment volumes, recognized revenue and the economics of individual deployments undisclosed.

The company's Lenovo AI Host P7 announcement supplies another named product using M50. Together with the Great Wall notebook, it shows the chip entering different device formats. That variety gives Houmo.AI opportunities to sell its technology, and also places demands on the software and hardware support surrounding each implementation.

Its software platform includes a compiler, runtime, model tools, quantization and profiling capabilities. The company also describes support for mainstream inference frameworks. Those components address different stages of deploying a model, from preparing it for the chip to examining execution and managing the device while it runs.

This is where the funding's ecosystem allocation becomes commercially legible. Reusable development support can reduce the amount of bespoke adaptation attached to the next project. The benefit depends on what manufacturers can actually deploy with it, but the work itself is concrete: engineers need supported paths through model preparation, execution and diagnosis. A growing design pipeline gives that software effort an audience already making product decisions.

Financing today's devices and tomorrow's architecture

Houmo.AI's roadmap joins continued M50 delivery with M6X and 3D compute-in-memory development. The two demands overlap inside one organization. Existing partners need an implementation they can use, while new architectures need research, engineering attention and a route into the products that will eventually depend on them.

The company has also described collaboration with China Mobile around edge AI terminals, providing an earlier example of its chip-to-device approach. Those relationships give the business context beyond a laboratory architecture. Their commercial significance will depend on the devices, applications and support arrangements each partner builds, rather than on treating every collaboration as a disclosed revenue stream.

For operators evaluating local inference, Houmo.AI's story puts supplier support alongside power and compute specifications. The buyer inherits a model lifecycle, a software environment and a relationship with the chipmaker. Capital allocated to that environment can affect the practical effort required to keep a device's AI capability working as its applications evolve.

Qiang Wu's team now has a reported financing round supporting both sides of that relationship. M50 partners are working with silicon that exists today, while the roadmap reaches toward different memory and compute arrangements. Each additional device brings another set of engineering decisions back to the company, giving its next generation of chips a growing body of customer work to learn from.