QBit Semiconductor Announces Physical AI and Drone Road Map With QB88XX Series

QBit Semiconductor (TWO: 7913) announced a road map for chipsets for robots and drones targeting physical AI control opportunities at SEMICON Taiwan 2026 on September 17, 2026. At its core is the QB88XX series, which integrates a robot’s “cerebellum” control functions and is intended to address space and power constraints in dexterous robotic hands with more than 20 degrees of freedom. For its drone business, the company outlined a two-stage strategy: QB77XX from 2026 to 2028, followed by QB88XX from 2029 onward.

QBit Semiconductor’s road map addresses robot joint control and drone flight control together, signaling a strategy of moving control functions onto the SoC. According to the company, QB88XX supports control of up to 32 motors, while QB77XX supports up to 32GB of LPDDR5/5X. This shifts the design focus beyond computing performance to how control, communications and memory are integrated.

What is new in this announcement is that the company presents robots and drones not as separate control-device categories but as connected targets within its chipset road map. QB88XX brings together a CMAC architecture, a quad-core Arm Cortex-A78, an NPU, an Arm Cortex-M7, TGEN and a secondary NPU in one design, and also includes control for up to 32 motors. This suggests that the main focus is not simply a compute SoC, but the degree of integration in bringing together joint and servo control. The drone concept, meanwhile, is phased. The QB77XX, planned for 2026 to 2028, supports up to 32GB of LPDDR5/5X, PCIe, USB 3.2 and GbE, and includes a Cortex-M33 and TGEN. It aims first to streamline and separate foundational flight control and mission computing. The company then plans to move to QB88XX from 2029 onward, consolidating optical-flow control and target tracking on a single chip. This points to an extension of the integration approach used for robots to drones. A notable feature is the sequence of capabilities, moving from integrated control to integrated perception and tracking. In terms of industry structure, the competitive focus may shift away from general-purpose, GPU-like computing toward how extensively motor control, real-time flight control, camera input, bus connectivity and secure boot can be integrated. For robots, the company cited dexterous hands with more than 20 degrees of freedom; for drones, it cited tests of single-object tracking at 60km/h for 10–15 minutes. These represent only part of real-world performance, however, and power consumption, heat, yield and software-stack portability in mass production are likely to be the next points to assess.

The QB88XX series is based on a cerebellum-model articulation controller (CMAC) architecture and combines a quad-core Arm Cortex-A78 processor with an NPU to execute commands from the brain, according to the company. It also combines an Arm Cortex-M7, TGEN and a secondary NPU for motor servo control, and supports up to 32 motors and AI servo tuning. It integrates PCIe, USB, CAN bus and camera interfaces for robotic arms and industrial automation. In phase one for drones, QB77XX is planned to support up to 32GB of LPDDR5/5X, PCIe, USB 3.2 and GbE, and to provide real-time flight control with TGEN and a Cortex-M33, according to the company. In phase two, the company plans to introduce the QB88XX SoC, integrating flight control and mission computing on a single chip and featuring optical-flow control and target tracking.

In Japan, the adoption requirements for integrated SoCs of this kind could affect components, implementation and embedded design related to robot joint control and drone flight control. In particular, if designs integrating PCIe, USB, CAN bus and camera interfaces become more widespread, assumptions about peripheral configurations and board design could change. Domestic control-equipment makers and embedded developers may therefore need to keep pace with the specifications.