GuShiio.com 
(Photo source: Photo taken by Zhijia Lin, editor of GuShiio.comAGI)
With the explosion of DeepSeek in February this year, the AI chip industry has ushered in new opportunities, and Ali is also accelerating the improvement of AI computing technology capabilities.
GuShiio.comAGI learned that on February 28, XuanTie, a brand owned by Alibaba Dharma Institute, announced its first server-level central processing unit (CPU) IP core, XuanTie C930, based on the open source RISC-V architecture. Delivery is expected to begin this month (March).
At the same time, Dharma Institute disclosed that C908X, R908A, XL200 and other Xuan Iron processor families, and their finished chips will be used in AI acceleration, vehicle-mounted, high-speed interconnection and other directions. Dharma Institute is based on three mainstream operating systems (Linux, Android, RTOS) launched three sets of Xuan Iron SDK software development kits.
in fact, in February this year,Alibaba Group CEO Wu Yongming announced that in the next three years, Alibaba will invest more than 380 billion yuan in building cloud and AI hardware infrastructure, totaling more than the total of the past ten years.“The primary goal of Alibaba Group’s AI strategy is to achieve AGI (General Artificial Intelligence), which is defined as AI can complete more than 80% of human capabilities. So if AGI comes true, the AI industry may become the largest industry in the world.& rdquo;
It is reported that Xuan Tie willAnchoring the two major directions of high performance and AI that are consistent with Ali AGI’s goals,At present, the Xuantie team has promoted the implementation of more than 30% of RISC-V high-performance processors, accelerating the penetration of RISC-V in various fields.
After the meeting, Zhang Xiantao, president of Alibaba Cloud Shadowless Business Unit, told.comAGI thatIt took ten years for the ARM architecture to build the server software ecosystem step by step, and Alibaba’s Yitian 710 built based on the ARM architecture has the industry-leading level of computing power. Looking at the RISC-V industry,
“I think after 5-8 years of development, large-scale applications in servers should not be a problem in the future. Many companies have high expectations for it and will definitely accelerate the process forward.& rdquo; Zhang Xiantao said that the RISC-V architecture can be implemented in about 5-8 years from large-scale application of low-power IoT terminals to data centers.
Ni Guangnan, academician of China Academy of Engineering, said at the 2025 Xuantie RISC-V Ecological Conference that RISC-V, as an open source and open hardware architecture, is becoming a new engine for transformation in the global chip industry, and China’s contribution in this field is particularly prominent. Open source is booming in China and has become a powerful driving force for technological innovation. The open source model represented by DeepSeek has changed the competitive landscape of the global AI industry. Practice has shown that open source not only realizes the sharing and optimization of global resources, but also accelerates the continuous evolution of technology.
Ni Guangnan, academician of China Academy of Engineering, said at the 2025 Xuantie RISC-V Ecological Conference that RISC-V, as an open source and open hardware architecture, is becoming a new engine for transformation in the global chip industry, and China’s contribution in this field is particularly prominent. Open source is booming in China and has become a powerful driving force for technological innovation. The open source model represented by DeepSeek has changed the competitive landscape of the global AI industry. Practice has shown that open source not only realizes the sharing and optimization of global resources, but also accelerates the continuous evolution of technology.
A new open source track with a scale of 670 billion will break out
RISC-V is an open instruction set architecture (ISA) first released in 2010 by a research team at the University of California, Berkeley. Currently, the RISC-V Foundation has more than 4120 members from 52 countries around the world, including Google, Nvidia, Intel, Ali and other companies as important members, and has more than 80 technical working groups promoting RISC-V standards, software, tools, etc.
In fact, the currently widely used instruction set architecture ARM and the latest RISC-V both originated from the reduced instruction computer RISC of the 1980s. The difference is that ARM is a closed instruction set architecture. Manufacturers applying ARM architecture can only adjust product frequency and power consumption according to their own needs, and are not allowed to change the original design. At the same time, ARM CPU instructions are complex in number and numerous in versions, are neither compatible with each other nor support modularization, and have expensive patent and architecture licensing issues.
Unlike 6 and ARM architectures, which have authorization restrictions,RISC-V has features such as streamlining and open source, which can develop unique chips that are more suitable for specific needs, and breaks the conventions of high licensing fees and difficulty in customization for x86 and ARM architectures.
As an open source standard, RISC-V allows any individual or organization to use, modify and extend freely without paying patent fees. Its design goal is to provide a simple, extensible and flexible instruction set, while being modularized so that users can freely customize different instruction subsets according to their needs, suitable for a wide range of application areas, from embedded microcontrollers to high-performance computing.
The scale of the RISC-V market is growing year by year.As early as 2022, global RISC-V chips have shipped more than 10 billion chips, half of which come from China.
Bao Yungang, deputy director of the Institute of Computing Technology, China Academy of Sciences, revealed thatBy 2030, the global RISC-V chip market will reach US$92.7 billion (approximately RMB 675.208 billion), with a compound average annual growth rate of 47.4%.
Calista Redmond, CEO of the RISC-V Foundation, said at the end of last year that more than 2 billion SoCs are currently using RISC-V architecture cores.By 2031, the number of SoCs using RISC-V cores will surge to 20 billion.
At present, the RISC-V industry is mainly divided into two groups: one is to use the RISC-V architecture to sell IP cores; the other is to design chip products through the RISC-V architecture, and then stream them, mass produce and sell them, or use other RISC-V IP cores to mass-produce chips or end products.
In fact, NVIDIA is also using RISC-V technology. At last year’s RISC-V North America Summit, Frans Sijstermans, vice president of multimedia architecture at NVIDIA, revealed that ten years ago, NVIDIA changed the Falcon microprocessor (used as a logic controller for GPU products) to the RISC-V architecture. Currently, each NVIDIA chipset contains 10-40 RISC-V core IPs, depending on the configuration.In 2024, NVIDIA shipped more than 1 billion RISC-V processors.
As an important domestic technology Internet company and an early company in deploying open source RISC-V architecture and chip semiconductors, Alibaba launched its self-developed chip project more than six years ago.
At the 2018 Yunqi Conference, Zhang Jianfeng, partner of Alibaba Group, then CTO of Alibaba and president of Alidamo Institute, announced the merger of Zhongtianwei and the establishment of Xuantie team.
In July 2019, Xuantie released its first RISC-V processor C910. Later, in 2021, Xuantie E902, E906, C906 and C910 and full-stack software tools will be open source, and the sword without 100 will be open source.
In August 2022, Ali released the first high-performance RISC-V chip design platform, the Sword Without 600 and SoC chip Shadow 1520. Later, it also announced the C920 and more advanced RISC-V IP and software.Today, Xuantie RISC-V belongs to the brand of Aridamo Institute, and the team is led by Zhang Jianfeng.
Up to now, Xuantie has successively launched a series of processor IPs and chips, which can meet the full range of high, medium and low performance requirements, covering multiple scenarios including smart terminals, network communications, AI smart computing, servers and peripherals. Among them, Xuantie provides RISC-V IP core sales and service solutions for C series (computing), R series (reliable real-time), E series (embedded), and XT-Link network series.
It is reported that the general computing power performance of the new C930 reaches the SPECint2006 benchmark of 15/GHz, and is aimed at server-level high-performance application scenarios. At the same time, the C930 is equipped with 512 bits RV1.0 and 8 TOPS Matrix dual engines, which natively combines general high-performance computing power with AI computing power, and opens a DSA extension interface to support more feature requirements.
Among the new members of the Dharma Institute Xuan Tie processor family, the C908X is positioned as Xuan Tie’s first AI dedicated processor, supporting 4096 bits long data bit width RV1.0 vector expansion; the R908A is targeted for the high reliability needs of vehicle-scale chips; the XL200 will provide larger-scale, higher-performance multi-cluster consistent interconnection.
In addition, at the meeting, Dharma Institute also officially announced that a group of new members of the Swordless Alliance would join, including EDA giants Cadence and Siemens EDA. In addition, the previous addition of Synopsys brought Ali together the three EDA giants, which is worthy of attention.
Last year, Zhang Jianfeng said that with the surge in demand for new computing power, RISC-V is about to enter an application explosion period. rdquo;
“RISC-V is the open source architecture needed in the AI era, leading us on an open AI revolution path. rdquo; Luca Benini, a professor at the Federal Institute of Technology in Zurich in Switzerland, believes that the market should accelerate its embrace of the RISC-V architecture, and he led the team to build an open source AI software and hardware platform based on processors such as Xuantie C910 and improve AI computing efficiency.
It is predicted that the future market share of RISC-V chips in consumer PCs, autonomous driving, network communications, industrial control, smart devices and high-performance servers will exceed 25%, becoming a force that cannot be ignored in the global semiconductor industry.
The market is still in its early stages, and RISC-V high-performance AI chips lack a benchmark”
On February 28, Li Chunqiang, a senior technical expert at Alibaba Dharma Institute, told GuShiio.comAGI:
“We haven’t seen any more features on the RISC-V architecture (AI performance release). In fact, it lacks a benchmark, including at the chip level or the whole machine level. As long as there is a benchmark, it can be applied immediately. High performance and AI, so that the performance of the RISC-V chip can compete with x86 or ARM, at least at the same level, otherwise how can people use your product?& rdquo;
Later, GuShiio.comAGI continued to ask: Does Xuan Tie want to be a benchmark?& rdquo;
Li Chunqiang bluntly said that the Xuan Tie C930 must be going in this direction. rdquo;
This conversation highlighted that high performance is a challenge that the RISC-V architecture must currently break through.
From the embedded field to high-performance (HPC) AI development, although the RISC-V architecture has open source characteristics and its participants are also striving to solve the issues of standardization and tool unification, as the RISC-V ecosystem is still in its early stages of development, the fragmentation problem has become increasingly prominent. At the same time, RISC-V requires advanced processes, so the construction of technology and ecosystem systems is slow, triggering market competition. Currently, the entire industry still lacks iconic technologies such as DeepSeek and ChatGPT.
Bao Yungang pointed out that the software and hardware toolbox developed by RISC-V is not rich enough, and there is still a big gap compared with x86 and ARM architectures. In addition, talents at all levels such as RISC-V chip design, verification, solutions, and technical support are also facing a shortage. In order to build industry confidence, creating a RISC-V benchmark product case will be the key to the growth of the RISC-V ecosystem.
However, what RISC-V is very certain today is that there will be more and more RISC-V participants, and the entire technology direction needs to move towards RISC-V-based high-performance computing.
Bao Yungang said on the evening of February 28 that the fields of AI and intelligent driving vehicles will become emerging application scenarios for RISC-V and have huge potential.
“With the rise of large models, the computing power requirements generated by AI reasoning will increase by orders of magnitude. Especially recently, all industries have been deploying DeepSeek locally, creating huge demand for computing power across the country. The computing power requirements of AI reasoning show two characteristics: 1. Close collaboration with the CPU. AI inference will become an indispensable part of various businesses in the future, but the business main program still runs on the CPU to offload AI inference requests to the AI accelerator through API calls, and then return the inference results to the user by the CPU. 2. Present diversified needs. Different scenarios generate different computing power requirements, and the corresponding resource constraints are also different. For example, cloud inference computing power must consider the efficient deployment of full-blooded versions of large models, while end-side application scenarios often deploy tailored versions of different capacities. Therefore, AI accelerators need to consider co-design with the CPU and need to be able to achieve efficient customization according to different needs. If RISC-V’s flexibility advantages can be fully utilized, it is expected to become the best partner for AI reasoning and computing power, and RISC-V+AI will become a new combination in the future. rdquo; Bao Yungang wrote.
You know, Arm currently sells the most CPU IP cores. The company hopes to be a competitor to Nvidia’s 3 trillion market value. Although licensing designs from Nvidia accounts for only a small part of Arm’s business, the company may not be neutral towards Nvidia, which is reportedly developing its own AI chips.
On February 27, Arm CEO Rene Haas said that Nvidia still has many competitors, such as making general-purpose computing, reasoning and edge AI chips, and it is only a matter of time before sales of other AI chips will also increase significantly.
Haas emphasized that in the next few years, as models become larger and larger, the field of AI computing power will be divided. Among them, more and more funds will be invested in specialized GPUs or training solutions, but at the same time, CPU general-purpose computing chips based on inference AI still have great development prospects.
Now, with the DeepSeek craze coming, NVIDIA CEO Huang Renxun has said that the demand for AI computing power has not decreased. Therefore, RISC-V targeting AI is an important development prospect.
Li Chunqiang told GuShiio.comAGI that DeepSeek’s innovation has brought great inspiration and promotion to the industry. Among its innovations, DeepSeek has significantly reduced the activation parameter ratio through MoE-like technology, which significantly reduces the computing power required to achieve a large model with the same effect. This decline in computing power, from the perspective of chip design, will promote the formation of a new balance point in terms of computing power, storage capacity, inter-chip interconnection and storage bandwidth. This new balance point is a very good opportunity for RISC-V. For example, due to the significant decline in DeepSeek’s computing power, with the help of MOE technology, a lot of computing power can be completed in the CPU without running on a computing card. This is a great opportunity for the development of RISC-V or the CPU.
Li Chunqiang emphasized that RISC-V itself has the characteristics of open source, openness, and scalability, especially in the AI field. Recently, there have been many innovations in Vector, Matrix and other aspects, and even multi-core, heavy-core and other methods. By adopting RISC-V to carry out relevant work, good convergence points can be found in these aspects.
As the global AI industry has entered the stage of an arms race for computing power, companies such as Meta, Amazon, and Apple have increased their investment in AI computing infrastructure. In addition, SoftBank, OpenAI, and Oracle have announced the joint venture Stargate project, which aims to develop AI data centers for the United States and has committed to investing US$500 billion over the next four years.
Based on the RISC-V open source architecture, the independently developed Xuantie series processors are expected to become key engines in the industry’s AI computing power ecosystem.
At present, there are more than 100,000 derivative models of Alibaba’s open source model Qwen series, ranking first in the world. As Asia’s largest cloud service vendor and a force that cannot be ignored in China’s AI in the future, Ali will continue to invest in computing facilities and will inevitably restructure the AI landscape.
Ni Guangnan said that the Black Iron Team of Aridamo Institute has made outstanding contributions to promoting the development of global RISC-V ecology. From the early C910 high-performance cores to the C920 and more advanced RISC-V IP, the efforts of the Xuantie team have enabled more developers to innovate on high-performance RISC-V processors and promote the progress of the entire ecosystem.
Ni Guangnan emphasized that we strongly support the development of open source RISC-V because it is not only a technological innovation, but also a global change that affects future computing architecture. We must strengthen international cooperation, extensively absorb outstanding technical talents and resources from around the world, and jointly promote the prosperity and development of the RISC-V industry and ecology. rdquo;
