The Counterpoint Report's Forecasts
According to the latest global humanoid robot tracking report from Counterpoint Research, cited by IT之家, cumulative shipments of AI accelerators for humanoid robots will exceed one million units by 2030. The firm also estimates a compound annual growth rate of 144% over the 2026–2030 period. This is a forecast by the research institute, not actual data: the material does not indicate current volumes or margins of error for the estimate.
GPUs Lead in the Near Term, ASICs on the Rise
On the computing architecture front, Counterpoint notes that the GPU will remain the dominant solution for humanoid robots in the near term. The reason given is the progressive convergence between vision-language-action (VLA) models and world models. In parallel, AI ASICs are growing: vertically integrated manufacturers such as Tesla and Xiaopeng are transferring chips developed in-house for the automotive sector to humanoid robots.
According to the report, Nvidia currently dominates the GPU market for humanoid robots thanks to a complete ecosystem that includes computing chips, models and simulation tools. The company has collaboration relationships with several robot body manufacturers, including AgiBot, Unitree, UBTech, Galbot, Figure AI, Neura Robotics and Noetix Robotics.
Qualcomm, by contrast, enters with a heterogeneous computing architecture, focusing on energy efficiency and lower material costs. It has design agreements with Neura Robotics, VinMotion, Booster Robotics and Figure AI, but according to Counterpoint its shipment volumes remain significantly lower than Nvidia's.
Analyst Statements and the Chinese Front
Soumen Mandal, principal analyst at Counterpoint, says that the computing ecosystem for humanoid robots is moving from exclusive GPU dominance to coexistence among multiple AI architectures. As robotic models evolve from VLAs toward action-world models (WAM), according to Mandal, specialized computing will gain importance thanks to lower power consumption, fewer thermal dissipation constraints and greater energy efficiency.
Ethan Qi, deputy director at Counterpoint, argues that the GPU remains the preferred choice for on-device execution of the most advanced embodied models and that automakers, led by Tesla, will be the main promoters of ASIC architecture. Qi expects competition to intensify when, after 2027, next-generation accelerators capable of supporting artificial general intelligence (AGI) workloads enter production.
In China, D-Robotics, spun off from Horizon Robotics, collaborates with manufacturers such as Leju Robotics and LIMX Dynamics. Its RDK S100 integrated development kit adopts a CPU+BPU+MCU architecture with a BPU neural processing unit based on the Nash architecture and a stated computing power of 80 TOPS. The stated goal is to strengthen the autonomy of China's computing ecosystem for humanoid robots.
