正文 · AI 翻译
IT Home reported on October 26 that, according to Counterpoint Research’s latest global humanoid robot tracker, cumulative global shipments of AI accelerators for humanoid robots are expected to exceed 1 million units by 2030, with a compound annual growth rate (CAGR) of 144% from 2026 to 2030.
In terms of computing architecture, GPUs will remain the dominant architecture for humanoid robot computing in the near term as vision-language-action (VLA) models and world models gradually converge. Meanwhile, AI ASICs (application-specific integrated circuits) are also gaining ground rapidly, with vertically integrated manufacturers such as Tesla and XPeng transferring chip technologies developed in-house for automobiles to their humanoid robot product lines.
Nvidia currently dominates the humanoid robot GPU market with its comprehensive, full-stack ecosystem spanning computing chips, models and simulation tools. It has established close partnerships with leading robot manufacturers including AgiBot, Unitree Robotics, UBTECH, Galbot, Figure AI, Neura Robotics and Noetix Robotics.
Qualcomm has entered the market with its heterogeneous computing architecture, emphasizing energy efficiency and a lower bill of materials (BoM). It has secured design partnerships with Neura Robotics, VinMotion, Booster Robotics and Figure AI, but its shipment volume still lags significantly behind Nvidia’s.
Counterpoint Chief Analyst Soumen Mandal said the computing ecosystem for humanoid robots is shifting from GPU dominance toward the coexistence of multiple AI computing architectures. As robot models evolve from VLA to world action models (WAM), the value of dedicated computing hardware in delivering lower power consumption, reduced cooling requirements and greater energy efficiency will become increasingly apparent.
Counterpoint Associate Director Ethan Qi noted that GPUs remain the preferred choice for deploying today’s most advanced embodied models on devices, while automakers led by Tesla will be key drivers of ASIC architectures. Market competition is expected to intensify further as next-generation AI accelerators capable of supporting artificial general intelligence (AGI) workloads enter mass production after 2027.
Among vertically integrated companies, Tesla and XPeng Motors could reshape the humanoid robot AI accelerator landscape by drawing on their automotive manufacturing experience, chips developed in-house, proprietary AI models and full-stack hardware and software operations.
In China, D-Robotics (Digua Robotics), spun out of Horizon Robotics, has partnered with manufacturers including Leju Robotics and LIMX Dynamics. Its RDK S100 integrated computing and control development kit uses a “CPU+BPU+MCU” architecture and includes a BPU neural network processing unit based on the Nash architecture that delivers 80 TOPS of computing power, further strengthening China’s capacity to develop its own computing ecosystem for humanoid robots.