Huixi Embodied Full-Stack Compute Platform: Welding the Robot's Brain and Cerebellum into One Chip
On August 19, during the 2026 World Robot Conference, Huixi Intelligent launched the Huixi Embodied product series for embodied intelligence. The core is a SoC called R1 PRO, packing CPU, NPU, GPU and real-time interfaces into a single chip that completes the full "perception–reasoning–decision–motion control" loop on the same silicon.
Founder Xu Ningyi put it plainly: for embodied intelligence to enter the physical world, bigger models alone are not enough—you need a compute platform that can sustain the closed loop of perception, reasoning, decision-making and control.
Why "brain + cerebellum fusion" matters
Many past robot designs used a multi-chip approach: one chip running large models as the "brain," another handling motion control as the "cerebellum," communicating over board-level buses with high latency and system complexity.
Huixi's approach merges both into one chip:
- R1 PRO: 375 TOPS INT8, 16-core Cortex-A78AE, 1KHz real-time control, 22 camera inputs. On-chip closed loop cuts communication overhead and complexity.
- R1 MAX: 500 TOPS, 24 cores, targeting more complex, generalizable long-horizon tasks.
- Compute value is no longer just "how much it computes" but "whether it lets intelligence run more efficiently, reliably and at scale across different robots." Xu expects mass-deployment volume to explode within the next year or two, and penetration into every industry within 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, and introduces Agent capabilities into the dev workflow—targeting a compression of deployment cycles for the same algorithm from 2 weeks to 2 days.
- Ecosystem alliance locks in scenarios. At the launch, Huixi formed an "Embodied Computing Ecosystem Alliance" with first-batch partners covering robot bodies (Leju Robotics), sensors (Seyond), simulation (Lightwheel AI) and fundamental research (AIR Institute). The chip company's role shifts from hardware supplier to infrastructure layer.
- The embodied AI narrative is shifting from "who has the most parameters" to "who can work reliably, at acceptable cost, and replicate at scale." Compute platforms are the foundational pivot of that shift.
- Domestic compute is staking its claim in embodied intelligence: Beijing E-Town backs Huixi as a compute player in its industry chain, with a full-stack layout spanning core components, smart chips, robot bodies and application scenarios.
- For investors: watch RCM module customer adoption and real deployed unit counts—more informative than peak TOPS numbers.
But the chip is only the starting point. The real barrier-lower is the module: RCM400 / RCM500 integrate the SoC, LPDDR5, UFS 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, validation builds save at least 3 months, and production-ready solutions save at least 6 months.
Why this is bigger than "another chip launch"
The robotics industry has spotlighted hardware bodies, motion control and large-model capability. Huixi pushes the competitive focus down to the underlying compute infrastructure—exactly where the quiet battle for scaled deployment is fought.
Signals to note:
Also notable: power consumption is the core challenge on the path to deployment. Huixi optimizes across four layers—physical, logic, system and software—since battery life directly determines whether a robot can actually get work done.