Intel and Comtech Launch FoxJack X1, Fusing Robot Brain and Cerebellum on One Chip
An embodied AI robot can be understood as a four-layer intelligence stack. It includes an embodied brain for world understanding, task planning and action generation; a cerebellum for execution control, dynamic correction and locomotion; a robot body for mechanical execution and environmental interaction; and dexterous hands and sensors for contact perception, fine manipulation and feedback.
Historically, the brain and cerebellum have often run on separate computing platforms—an x86 general-purpose processor on one side and embedded AI acceleration modules on the other. The two systems communicate over buses, which increases overall system cost and introduces latency in data transmission and synchronization.
This division reflects the different requirements of high-level cognition and real-time control. However, it also creates integration challenges for compact robots that need both capabilities.
To address this bottleneck, Intel and Comtech Technology jointly released FoxJack X1, an embodied AI development kit. Based on Intel Core Ultra 300 series processors, it combines brain-level intelligent decision-making with cerebellum-level real-time control. The kit targets humanoid robots, mobile robots and industrial autonomous equipment with a high-energy-efficiency development platform.
The kit measures 125 × 125 × 57.5 mm and consumes 65 W, packaging a complete heterogeneous computing platform for embodied AI that is ready to use out of the box. The compute foundation is the third-generation Intel Core Ultra processor family, built on the Intel 18A process and officially announced at CES in January 2026. It is the first such platform to receive embedded and industrial-grade certification for edge scenarios such as robotics, supporting wide-temperature operation and 24×7 reliability.
Take the optional Core Ultra X7 358H as an example. The processor integrates four performance cores, eight efficient cores and four low-power efficient cores, with a maximum turbo frequency of 4.8 GHz. Together with integrated Arc graphics and an NPU, it forms a CPU + GPU + NPU heterogeneous AI compute system with 180 TOPS of INT8 dense compute.
Unlike the sparse compute figures commonly cited in the industry, this number is fully delivered by dense computation. Combined with LPDDR5X dual-channel memory, it is sufficient to run VLA end-to-end models, SLAM mapping and high-frequency motion control concurrently on a single chip. Official data shows the platform achieves 1.9× large language model inference performance and 4.5× VLA end-to-end model throughput.
On FoxJack X1, brain-cerebellum fusion is a deployment system that can be tailored as needed. For scenario-specific humanoid solutions, FoxJack X1 handles perception, decision-making and real-time control simultaneously on a single chip, replacing the dual-machine combination of an x86 host plus an NX or AGX acceleration card. This simplifies the compute architecture while reducing synchronization overhead.
For general high-end humanoid solutions equipped with Thor or RTX discrete graphics, FoxJack X1 can serve as an AI cerebellum, sharing part of the brain computation while allowing the expensive GPU to focus on large-model inference, reducing overall compute cost allocation. Its industrial real-time and virtualization capabilities can further replace real-time MCUs or multiple CPUs, continuously simplifying the robot architecture. In this way, the kit scales from a primary controller to a complementary real-time unit.
On the software side, FoxJack X1 comes preinstalled with Intel's embodied AI robot software development kit. It integrates the Preempt-RT real-time kernel and includes typical workflows such as ACT manipulation policies and ORB-SLAM3, along with LLM, VLM and VLA reference demos. It supports OpenVINO model optimization and both Windows and Linux.
When perception, decision-making and control can be closed-loop within a 125 mm square system, robot development efficiency, overall cost and reliability can enter a new range. The integrated approach also points to a future in which embodied intelligence platforms become more compact and easier to deploy.