HANGZHOU, China, Jan. 29, 2026 — Amid a global technology shift towards open architectures and on-device AI, Chinese RISC-V firm SpacemiT unveiled its new K3 AI CPU on January 29. The chip is designed to merge the open RISC-V instruction set with both general-purpose and AI computing, providing a more adaptable, energy-efficient, and economical platform for intelligent terminals and edge computing applications.
The processor market has long been controlled by x86 and Arm architectures. However, the migration of AI workloads from the cloud to endpoint devices is fueling demand for more open and customizable computing solutions. As a completely open-source instruction set architecture, RISC-V is now widely considered a crucial building block for future open computing ecosystems.
Established in 2021 and based in Hangzhou, SpacemiT is a rare example of a Chinese semiconductor company pursuing an exclusive “pure RISC-V” strategy. Its focus is on combining high-performance general computing with AI acceleration in a single chip, a concept it terms an “AI CPU.” This design targets intelligent hardware applications demanding high computational density, low power usage, and robust system integration.
The K3 chip is the outcome of over 1,200 days of research and development. The company states it is one of the first RISC-V AI CPUs ready for mass production that adheres to the RVA23 specification. It also features support for 1024-bit RISC-V Vector extensions (RVV) and native FP8 precision tailored for AI inference tasks.
Regarding hardware specifications, the K3 incorporates eight high-performance X100 RISC-V CPU cores that can reach a maximum clock speed of 2.4GHz. SpacemiT claims its single-core performance is roughly equivalent to that of Arm’s Cortex-A76.
The chip offers up to 60 TOPS of AI computing power and supports a maximum of 32GB of LPDDR5 memory. SpacemiT clarified that the K3 is not intended to rival high-end server CPUs or GPUs directly, but rather to facilitate the local processing of medium-scale AI models and multimodal applications. According to the company, the K3 can handle models with 30 to 80 billion parameters on a single chip, with typical system power draw ranging from 15 to 25 watts.
On the software front, the K3 utilizes a hardware-software co-design methodology. It is compatible with leading AI frameworks and compilers like Triton and TileLang, and works with major open-source AI ecosystems and Linux distributions. This approach is meant to lower the hurdles for deploying AI models on RISC-V platforms, making the developer experience more similar to that on x86 and Arm systems.
“We are convinced that the long-term trajectory for computing architectures is a move from closed to open systems,” stated Chen Zhijian, Founder and CEO of SpacemiT, during the launch. “x86 is very closed, Arm is semi-open, and RISC-V is completely open. Over time, open instruction sets have a greater potential to form the basis of worldwide computing.”
Chen further emphasized the particular importance of RISC-V for China’s semiconductor sector. “Historically, Chinese computing chips were mostly confined to the domestic market. Open architectures offer a fresh avenue for Chinese chips to integrate more seamlessly into the global technology landscape.”
The K3 highlights a blend of general computing, AI computing, and data-coherent interconnects. Chen characterized this progression as an evolution towards a “next-generation AI CPU,” where conventional CPUs gain greater intelligence.
“In the age of AI, CPUs cannot remain merely control processors,” he explained. “They also need to deliver native AI computing power. This shift is comparable to the transition from feature phones to smartphones—a fundamental transformation in computing’s nature.”
SpacemiT also revealed that its earlier K1 chip has surpassed 150,000 units shipped and is being used in industrial control systems, robotics, edge computing platforms, and open-source intelligent hardware. This market experience lays the groundwork for the K3’s introduction. The company reported that the K3 has already secured orders, with the first shipments scheduled to begin by the end of April 2026.
The firm is advancing its “full-stack RISC-V” strategy, which encompasses CPU IP, chip design, operating systems, compilers, AI software stacks, and developer platforms. The K3 is compatible with various operating systems, including Ubuntu, OpenHarmony, and OpenKylin. SpacemiT is also introducing companion products like PICO-ITX single-board computers, robot core boards, and array server platforms, while making its hardware reference designs available to developers and system integrators.
Hangzhou has recently become a significant center for AI, semiconductor, and open-source technology innovation in China. Industry analysts observe that a new wave of Chinese tech firms is emerging in the area, bolstering China’s position in advanced “hard-tech” industries. SpacemiT is viewed as a key representative within this ecosystem.
Concurrently, the industry widely accepts that RISC-V still trails x86 and Arm in high-performance computing, software ecosystem development, and overall industrial size. SpacemiT also acknowledges that its present products are more appropriate for intelligent terminals and edge AI applications than for direct competition with leading server processors or GPUs.
“Our objective is not to challenge global titans directly,” Chen remarked. “Rather, we seek to carve out distinct advantages in the mid-range computing market by leveraging lower power consumption, greater integration, and superior cost-effectiveness to bring AI computing to a wider array of devices.”
As AI increasingly transitions from centralized cloud platforms to local deployment, energy efficiency, system integration, and open ecosystems are emerging as critical competitive differentiators. The introduction of the K3 marks a tangible advancement by China’s RISC-V community in uniting open-source architectures with AI computing, illustrating China’s wider initiative to investigate new technological routes in the future of global computing.
CONTACT: Taliyah yi.tang@spacemit.com
