On August 19, during the 2026 World Robot Conference, Huixi Intelligence (辉羲智能) launched its Huixi Embodied product series, a full-stack computing platform aimed at embodied intelligence. At its core is the R1 PRO SoC, which packs CPU, NPU, GPU, and real-time interfaces into a single chip, closing the loop of "perception — reasoning — decision — motion control" on the same piece of silicon.
Founder Xu Ningyi put it plainly: for embodied intelligence to enter the physical world, bigger models alone aren't enough — you need a computing platform that can sustain the closed loop of sensing, reasoning, decision-making, and control.
Why "Brain + Cerebellum Fusion" Matters
Many past robot architectures were "multi-chip patchworks": one chip running a large model as the brain, another handling motion control as the cerebellum, communicating over board-level buses with high latency and system complexity.
Huixi's approach fuses both into one chip. The R1 PRO delivers 375 TOPS INT8 compute with 16 Cortex-A78AE cores, supporting 1KHz real-time control and 22 camera inputs. An on-chip closed loop cuts communication overhead and complexity.
The R1 MAX steps up further: 500 TOPS, 24 cores, targeting more complex, generalizable long-horizon tasks.
From Chip to Module
The chip is only the starting point. The real play to lower barriers is the module: RCM400 / RCM500 integrate the SoC, LPDDR5, UFS storage, and power management into mature modules with a unified 699-pin interface and the RISE/Robotics SDK. Robot companies no longer need to design high-layer PCBs or tune power and clocks from scratch.
According to Xu Ningyi, validation versions can save at least 3 months, and production-ready solutions at least 6 months.
Why This Is More Than "Another Chip Launch"
The robotics industry has spotlighted robot bodies, motion control, and large-model capabilities in recent years. Huixi pushes the competitive focus down to underlying compute infrastructure — precisely the hidden battleground for scaled deployment.
Key signals:
- Compute value is no longer just "how many TOPS" but whether intelligence can run more efficiently, reliably, and at scale across different robots. Xu predicts a burst of large-scale deployment within the next year or two, with broad industry adoption over 5 years.
- Software ecosystem is the second battlefield. Huixi has built a stack covering model frameworks, compilers, operator libraries, runtime, middleware, OS, simulation, and real-time control, plus Agent capabilities in the development workflow — targeting a reduction of the same-algorithm deployment cycle from 2 weeks to 2 days.
- Ecosystem alliance locks in scenarios. At the launch, Huixi formed an "Embodied Intelligent Computing Ecosystem Alliance" with first-batch partners covering robot bodies (Leju Robotics), sensors (Seyond/图达通), simulation (Lightwheel AI), and basic research (AIR institute). The chip company's role shifts from hardware vendor to infrastructure layer.
- The embodied AI narrative is shifting from "whose model has more parameters" to "who can work reliably, at acceptable cost, and replicate at scale." Computing platforms are the foundation of that shift.
- Domestic compute is positioning in the embodied-AI race: Beijing E-Town is supporting Huixi as a compute player in its full-chain layout spanning core components, smart chips, robot bodies, and scenarios.
- For investors: track RCM module customer adoption and real deployment counts — more informative than peak TOPS.
Notably, power consumption is a core issue for real deployment. Huixi optimizes across four layers — physical, logic, system, and software — because battery life directly determines whether robots can actually do work.