China and Europe Bet on RISC-V: Is It Enough to Shake up U.S. Semiconductor Supremacy?

The open-source RISC-V chipset architecture has been gaining ground across many areas, from server applications to the far edge, including the Internet of Things (IoT). More recently, with the advances of Artificial Intelligence (AI), the technology has attracted many chipset suppliers addressing the edge AI/Machine Learning (ML) acceleration space. New entrants (and established incumbents alike) are now developing chipset architectures and Intellectual Property (IP) around the RISC-V license-free Instruction Set Architecture (ISA). Modularity and scalability of the technology make RISC-V a viable alternative to established architectures such as Arm and x86 chips. In the AI space, RISC-V chips can be used to handle the workloads of many applications, ranging from low-powered inferences in battery-powered always-on IoT devices, all the way up to Large Language Model (LLM) training in data centers. This flexibility and applicability to AI/ML use cases—coupled with the savings on licensing fees for alternative proprietary architectures—have contributed to the groundswell of interest in the technology.

U.S. supremacy has shaped the semiconductor market, to date, particularly in advanced computing chips. This continues to apply to today’s emerging AI processors, currently dominated by NVIDIA, Intel, AMD, Qualcomm, and Google. RISC-V is seen by chipset players in China and Europe as a foundational platform that could enable them to compete with their U.S. counterparts. This is particularly the case with AI and AI inference workloads at the edge.

The growing interest in RISC-V is noticeable in Europe and China, in particular. Europe’s first RISC-V summit in Barcelona was attended by Intel and other established chipset ecosystem players. The European Union (EU) is promoting RISC-V development in both edge and data center settings, and views it as a vital technology in the quest for semiconductor sovereignty. In December 2022, EuroHPC dedicated €270 million to developing High-Performance Computing (HPC) RISC-V chipsets for the wider ecosystem. In China, U.S. sanctions restricting access to chipsets vital for handling advanced AI workloads are encouraging domestic public and private investment, with tech giants like Alibaba betting on RISC-V; last month, it joined eight other firms in a patent alliance to promote the architecture’s development. Geopolitical risks also likely prompted the industry body, the RISC-V Foundation, to move its headquarters from the United States to Switzerland, a notably more neutral location, ensuring Chinese players are less exposed to sanctions. The foundation’s 22-strong board is made up of six Chinese directors.

The aspirations of China and Europe to use RISC-V to compete with U.S. AI chipsets, notably in HPC and advanced AI processing, are noteworthy because these technologies are essential for shaping the industry’s digital transformation. ABI Research’s forecasts AI/ML hardware revenue for inference and training workloads (edge and cloud) to grow to US$68 billion by 2028, while software revenue is set to increase more than sixfold between 2023 and 2030. This trend is reinforced by the ascendency of generative AI, the need for low-latency inference, and concerns around data security (as well as sovereignty), which will also see processing continue its move from cloud to edge (e.g., smartphones, Personal Computers (PCs), home appliances, and enterprise gateways). Amid these conditions, a low-power, highly-customizable AI acceleration-optimized ISA that is open source provides fertile ground for gaining market share from proprietary architectures—especially because its performance rivals alternative architectures such as Arm and x86.

So, who are these ascendant Chinese and European players carving out market share from their more established U.S. rivals? In China, companies like Nuclei System Technology provide a range of Central Processing Units (CPUs) for AI-intense workloads in automotive, Augmented Reality (AR)/Virtual Reality (VR) and Artificial Intelligence of Things (AIoT) applications. RiVAI Technologies—explicitly touting its domestic production capabilities and with management that includes engineers from the original RISC-V project at UC Berkeley—supplies a range of RISC-V microprocessors for AI/ML use cases ranging from edge servers to noise cancellation in consumer wearables. In Europe, Axelera AI is focusing its IP on edge AI applications like sensor data evaluation, gesture control, and vibration analysis.

Concurrently, U.S. activity around RISC-V is by no means falling behind. This year, Meta announced its own RISC-V silicon for running generative AI workloads for its recommendation engine. Startups like Tenstorrent (which recently closed a US$100 million financing round led by Hyundai and Samsung) will spur chipset development for AI inference on household consumer goods like TVs and automotive applications. Ventana Micro Systems is also making headway in low-latency network edge use cases such as Vehicle-to-Infrastructure (V2I) communication networks, alongside more established players like SiFive (founded by the inventors of RISC-V), which are leaders in the space, and have a suite of automotive chipsets. Add to that the recent announcement by chipset industry giant Qualcomm, and leading automotive suppliers including Bosch, of a joint venture to boost RISC-V hardware development in the automotive and eventually mobile and IoT spaces. Finally, Esperanto, another established AI/ML RISC-V chipset supplier, announced a pre-loaded, self-contained hardware and software solution for deploying generative AI at the enterprise edge.

The RISC-V suppliers are unsurprisingly bullish on their technology—and no wonder given the investment needed for Research and Development (R&D) (notwithstanding the savings on licensing fees). However, it is a relatively new architecture, having only been invented in 2010, with commercial solutions coming to market toward the end of the last decade. The ecosystem is still developing and areas like software development are playing catch-up.

RISC-V processors will continue their ascendance as a viable alternative to existing ISAs, including x86 and Arm.  The race for AI-optimized chipsets will encourage further investment in the technology from both private sector players and public funds, mainly in regions concerned about their vulnerability to complex international supply chains for semiconductors.

Finally, with interest in RISC-V increasing in all parts of the world (China, Europe, Japan, and the United States, in particular) it is unlikely to see any particular company controlling the RISC-V agenda or steering its roadmap. However, the technology may enable companies and regions to break free from political constraints and form technology control and path-dependency of industry forces. This will enable new entrants and challengers, wherever they are in the world, to build on innovations and compete effectively with legacy players. Not only could this spur healthy competition, but it could also engender significant economic growth across nations and industries.