After the era of PCs and smartphones, what computing product will dominate the mass market next?

From model training to machine control, AI is making its way into the physical world, unlocking new markets for RISC-V.

Recently, the 6th RISC-V China Summit took place in Shenzhen. Alibaba's DAMO Academy introduced the automotive-grade enhanced processor XuanTie E908A, Spacetime Technology unveiled the server-grade chip V100, and Leading Wisdom presented the 'Everest 200' designed for intelligent computing centers. To meet the evolving demands of AI, RISC-V vendors are expanding their reach from embedded systems and IoT to automotive, robotics, and data centers.

According to SHD Group, by 2031, RISC-V is projected to capture 33.7% of the overall hardware market, with its edge computing and data center markets surpassing $45 billion and $70 billion, respectively.

From edge devices to cloud infrastructure, physical AI and AI data centers are emerging as two critical areas for RISC-V companies. The former targets scenarios like robotics and autonomous driving, necessitating closed-loop perception, decision-making, and action in real-world environments, with stringent requirements for real-time response, energy efficiency, and security. The latter focuses on handling workloads such as large model inference, emphasizing data throughput, memory bandwidth, and the synergy between CPUs and AI accelerators. These diverse computational needs are propelling RISC-V to broaden its application scope.

Robotics Drives New Computational Demands

After the PC and smartphone eras, what will be the next computing product to capture the mass market?

Sun Yanbang, co-founder of Spacetime Technology, stated in an interview that there is currently no definitive answer. From Spacetime Technology's viewpoint, robotics is the most promising candidate, encompassing not only humanoid robots but also various intelligent devices in household, industrial, and public settings. He provided an example: traditional home appliance voice control relied on fixed command words, often failing to recognize alternative expressions. With enhanced model capabilities, devices can now better comprehend human intentions.

Future products will require local computing power and models to autonomously complete tasks, forming closed loops. This transformation alters chip design requirements. Sun explained that robotics necessitate CPUs to run robot operating systems and manage motion control tasks, AI computing power to execute models, and real-time responsiveness—all three computational capabilities must be seamlessly integrated.

RISC-V's modular and scalable attributes offer design flexibility for such integration. Companies can tailor computing cores and acceleration units around specific tasks, but competitiveness still hinges on validation through complete system applications. According to Sun, Spacetime Technology has customized its K3 chip for the Unitree robotics platform, replacing relevant software and computing platforms. Currently, the focus is shifting from basic software adaptation to application development, aiming to foster the creation of scalable products.

Simultaneous Progress in Automotive, Distributed Computing, and Cloud

The automotive sector is another domain with stringent demands for computing power and real-time performance.

To date, XuanTie chips have shipped over 6 billion units. Leveraging existing applications, XuanTie is further venturing into automotive and embodied AI fields. At the summit, DAMO Academy announced the automotive-grade enhanced processor XuanTie E908A for safety islands, domain controllers, and in-vehicle gateways, and outlined its roadmap for autonomous driving and embodied AI processors. Subsequent releases will include automotive-grade versions of C925, C930, and C950, with the C950 automotive version primarily targeting autonomous driving and expected by the end of 2027.

Meng Jianyi, chief scientist at DAMO Academy, informed Semiconductor Industry Insights that physical AI demands heightened computing power and real-time performance, necessitating computational architectures to evolve around algorithms and models—a forte of RISC-V's customizability.

In the automotive MCU sector, Zhao Liping, software technology director at ChipInfinity, introduced the XC32LS (XVA170) chip, which features a dual-core lockstep architecture based on in-house RISC-V IP. Designed for high-reliability applications like automotive powertrains and advanced electronic control systems, it adheres to ISO 26262 ASIL-D functional safety standards. The XC32MC body motor control MCU commenced mass production in 2026, with cumulative orders surpassing 5 million units, marking a new commercialization milestone for RISC-V in automotive electronics.

ESWIN Computing showcased RISC-V products and solutions encompassing 'perception-interaction, understanding-decision-making, and control-execution.' He Ning, senior vice president and CTO of ESWIN Computing, believes that as large models enhance understanding capabilities, the bottleneck for intelligent systems is shifting towards reliable execution. Embodied AI requires a blend of central coordination and local autonomy, with each node necessitating local perception, decision-making, and real-time response capabilities.

The competition for AI computing demand has also extended to data centers. Spacetime Technology's V100 integrates 46 RISC-V cores, supporting 8-channel DDR5 and 52 PCIe 5.0 lanes. According to Spacetime Technology, the chip fully complies with RISC-V server-grade specifications, enabling direct installation, boot, and operation of general-purpose distributions.

Leading Wisdom's 'Everest 200' is optimized for KV Cache and long-context large model inference, integrating matrix and vector acceleration engines and supporting DDR5, PCIe, and CXL interfaces, with detailed specifications slated for release by the end of this year. Despite their differing application focuses, these products are redefining CPU capabilities around AI workloads.

From Chip Availability to Mass Production

The market space is expanding, but commercialization still encounters software and application obstacles.

Frost & Sullivan data indicates that China's RISC-V main control chip market surged from RMB 1.1 billion in 2021 to RMB 37.7 billion in 2025, with an anticipated reach of RMB 219.4 billion by 2030. This forecast spans multiple application areas, but the contribution of physical AI hinges on the implementation of specific products.

'Ultimately, everything revolves around applications,' Sun remarked. Compared to PCs and smartphones, which require compatibility with extensive existing software, robotics and specific intelligent agents have more focused software needs, allowing adaptation around specific scenarios to create functional systems.

However, operation is merely the first step. After robots enter the physical world, motion control must remain reliable despite model or system anomalies. Sun noted that customers can continue using a combination of computing chips and standalone MCUs, while Spacetime Technology will also persist in refining on-chip real-time computing and safety capabilities.

The software ecosystem similarly demands industry-wide investment. David Patterson, founder of the RISC-V International Open Source Laboratory, highlighted at the summit that AI inference faces memory bandwidth and cost bottlenecks, necessitating enhancements to the AI software stack and leveraging AI to reduce hardware development costs.

From chips and basic software to applications and mass production, for RISC-V companies venturing into physical AI, the next phase of competition will hinge on the success of this path.

*Disclaimer: This article represents the original creation of the author. The views expressed are personal opinions. Our republication is solely for sharing and discussion purposes and does not imply endorsement or agreement. For objections, please contact our backend.