Looking at the "2.5 billion cumulative units shipped" figure that T-Head announced in 2021, one might conclude that China has already completed its shift away from RISC-V dependence on Arm and x86. This figure has been repeatedly cited without ever specifying its timeframe, spreading a vague image of self-sufficiency free from Arm or x86. But the actual Chinese server market looks entirely different. According to IDC's full-year 2025 tally, x86-based systems derived from Intel and AMD still account for a combined 74.3% of the market, Arm-based systems (mostly Kunpeng), often held up as the symbol of domestic substitution, remain at just 25.5%, and RISC-V doesn't even appear in this breakdown. What China is pursuing isn't market share capture — it's a long-term institutional strategy: seizing control of a neutral governance body that no one can legally shut down, and building influence from within it.
The Myth of "2.5 Billion Units" and the Persistent Reality of x86 Dependence
On October 19, 2021, at Alibaba Cloud's annual event, Zhang Jianfeng, who then led the company's cloud business, announced that cumulative shipments of XuanTie CPU cores — developed by T-Head, the group's chip design unit, and spanning both proprietary architecture designs and RISC-V-based designs — had surpassed 2.5 billion units. The breakdown of applications centered on IoT-related embedded devices such as machine vision, industrial control, and automotive terminals; server-related figures were not included. Nevertheless, this number has continued to be cited in subsequent reporting without specifying when it was announced, spreading the impression that RISC-V is currently sweeping through China's computing infrastructure in real time. The number itself is factual, but each time the context is stripped away — whether it represents RISC-V alone, and at what point in time, in which product category — the gap between impression and reality widens further.
Looking at China's overall server market, the picture is entirely different. According to IDC's survey of the full-year 2025 Chinese general-purpose server market, Intel-derived x86 accounts for 47.8%, AMD-derived x86 for 15.1%, and other domestic x86-compatible chips for 11.4% — x86-based systems combined account for 74.3%. Arm-based systems (mostly Kunpeng) remain at just 25.5%, and RISC-V doesn't even appear in this breakdown. At the foundation of a national strategy touting "self-sufficiency," the majority of servers actually running today are still imported or based on imported x86 architecture.
Even narrowing the focus to procurement limited to domestic substitutes, the choice comes down to just two options: Arm-based and x86-compatible chips. In a roughly 11.5 billion yuan procurement conducted by China Telecom for 2026-2027, bids were effectively limited to just two options — the Arm-based Kunpeng and the x86-compatible Hygon C86 — with Kunpeng accounting for 70.0% (28,000 units) and Hygon C86 for 30.0% (12,000 units) by unit count (by value, Kunpeng's winning bids accounted for roughly 70.7% of the total). At China Mobile, one concentrated procurement round reportedly saw Arm-based models reach a 65.01% share — said to be the highest ratio ever recorded for a telecom carrier's domestic Arm server procurement. According to Tom's Hardware's tally, the share of domestic CPUs across China Mobile's overall procurement has risen from roughly 21% in 2020 to approximately 43.5% most recently, but this rise has been driven primarily by Arm-based players like Kunpeng and x86-compatible players like Hygon C86 — RISC-V has essentially no place in this procurement mix. Between the "2.5 billion cumulative units shipped" achievement of the IoT era and the architectures actually running servers today, there's a gap of nearly five years and a disconnect in application.
In March 2025, T-Head launched the "XuanTie C930," marketed for servers, PCs, and autonomous vehicles, achieving 15 points per GHz on SPECint2006, an integer performance benchmark, and compliance with the RISC-V extension specification "RVA23." At this point, server-market entry on the performance front had only just begun.
A year later, on March 24, 2026, the company unveiled the successor "XuanTie C950." Reportedly manufactured on a 5nm process (the manufacturing partner has not been officially disclosed, though some reports name TSMC), it reached a maximum clock speed of 3.2GHz and reportedly exceeded a single-core SPECint2006 aggregate score of 70. Its primary intended use is inference processing for AI models with parameter counts in the hundreds of billions — a design clearly aimed at the server and AI market. Normalizing this score by clock speed yields roughly 22 points per GHz, only about a 1.5x improvement over the C930's 15 points per GHz. The C930's manufacturing process was never disclosed, and it's unconfirmed whether the two scores were measured under identical conditions — there remains a distance to close before matching the level of mature Arm- or x86-based server CPUs.
In August 2024, a vulnerability dubbed "GhostWrite" (CVE-2024-44067), stemming from a flawed vector extension implementation, was discovered in the older-generation C910 and C920. Because it stems from hardware, it could not be fixed via software update, and the only mitigation was to fully disable the vector extension feature itself — a response that meant losing roughly half of the instruction set's overall scope. This tradeoff meant sacrificing part of the performance, demonstrating that designs fresh off mass production still had room for technical maturity.
Separately, the AI accelerator "Zhenwu," which T-Head reported had shipped 470,000 units as of February 2026, is a GPU-positioned product and a separate line from RISC-V-based XuanTie. For T-Head itself, technology investment in XuanTie is seen as also nurturing IP licensing revenue, viewed by some as part of preparation for an independent listing.
"RVA23," the specification that both C930 and C950 comply with, is a standard profile for application processors set by RISC-V International; compliance improves compatibility with major operating systems and makes it easier to secure software portability. Even as individual vendors add their own proprietary extension instructions, as long as they meet this common baseline profile, ecosystem-wide compatibility is maintained. The final decision on which extensions get incorporated into this standard profile is made by the board — the subject of a later section.
Even Kunpeng, which dominates within the scope of domestic substitution, is itself Arm-based, and x86 still holds a majority share across the market as a whole. As the next chapter shows, Arm is not a politically ironclad option for China, and x86 carries similar vulnerabilities. Yet the fact that one or the other continues to dominate servers today confirms that China's investment in RISC-V is more a hedge for the future than an immediate market grab.
Why Arm and x86 Became "Dangerous Choices"
The moment China's move away from Arm and x86 became decisive was May 2019. When the United States added Huawei and its chip subsidiary HiSilicon to its export blacklist, Cambridge, UK-based Arm temporarily suspended all licensing agreements with Huawei on the grounds that its designs contained "US-origin technology" (partially restored later). Arm is a British company, not under direct US government jurisdiction, but this episode strongly impressed upon China's semiconductor industry the reality that even UK-originated technology can have its supply halted once it passes through the US export control network. Huawei continued development through its in-house HiSilicon designs, but not without experiencing disruption.
Arm went public in the US market in September 2023, but its parent company, Japan's SoftBank Group, has continued holding roughly 87-90.6% of the shares since. The value of this holding was estimated at roughly $130 billion (about ¥20 trillion at an exchange rate of ¥158-159 to the dollar) as of December 2025, though stock price volatility means this estimate carries considerable uncertainty. Judging from this ownership structure, Arm's own governance is far from monolithic.
In other words, the entity that effectively controls the British-registered Arm is a Japanese company — meaning that, from China's perspective, this architecture carries not only US risk but also the jurisdiction of Japan. Moreover, Arm's own corporate scale is small relative to the size of the ecosystem it controls. Public data as of 2025 puts its employee count at roughly 8,330 — meaning the management decisions and internal conflicts of a single company of this size can affect nearly all processor designs for smartphones worldwide. Furthermore, even after Arm China's CEO Allen Wu was dismissed in June 2020, he retained control of the company seal and registration documents and refused to step down; it wasn't until two years later, in 2022, when the co-CEO structure of Liu Renchen and Eric Chen was established, that SoftBank and Arm regained actual authority. A single licensor company's internal conflict alone effectively left China's entire Arm ecosystem in limbo for two years.
Intel and AMD have maintained a relationship allowing mutual use of each other's architectures through patent cross-licensing agreements since 1976, renewing the agreement in 2009 following a settlement of antitrust litigation. This relationship is limited strictly to the two companies and is not open to any third party at all. Intel previously settled a patent dispute with Taiwan's VIA Technologies in 2003, granting an exception through licensing, but has otherwise historically excluded x86-compatible chipmakers from the market through litigation. VIA's rights were passed on after 2013 to Zhaoxin, a joint venture with the Shanghai municipal government, keeping the lineage alive, though its market share remains limited.
Arm could see its entire supply chain halted if a single licensor company becomes dysfunctional due to political pressure or internal conflict, while x86 remains dominated largely by the existing lineage of Intel, AMD, and Zhaoxin. In July 2026, reports emerged that Intel would supply design information for Atom-series cores to the startup RosaicLabs, suggesting some new entrants are emerging — but in terms of the sheer scarcity of practical alternatives, the choices remain similarly closed off. For China, neither option changes the fact that it remains dependent on an entity it cannot fully control through its own will alone.
From the 2019 Huawei sanctions to the Arm China internal conflict, the lesson these events left with China is consistent: the very structure in which an entire supply chain can be swayed by the licensing decisions of a single company or country is itself the risk — and this recognition fueled interest in RISC-V as the next option.
The Mechanism Behind RISC-V's "Unstoppable" Design
The origin of RISC-V lay in a very practical licensing barrier. In 2010, in Krste Asanović's lab at UC Berkeley, graduate students Yunsup Lee and Andrew Waterman reportedly attempted to adopt Arm's ISA for the instruction set of a new research processor. However, Arm's licensing terms did not permit free modification or redistribution, and even research and educational use was constrained. Running into this wall prompted the lab to pivot toward designing its own instruction set architecture, which became the prototype for RISC-V. The Asanović lab was also part of Par Lab, a parallel computing research group led by Professor David Patterson, known as a pioneer of RISC design — making the pivot to a proprietary ISA a natural continuation of this academic lineage.
RISC-V is designed so standard and custom extensions can be added on top of a base instruction set, letting vendors freely add specialized instructions — such as matrix operation instructions for AI. One research example reported that RISC-V E203 (0.145mW) reduced power consumption by roughly 15% compared to Arm Cortex-M0 (0.17mW), though this is an isolated experimental result that cannot be generalized.
This design was formally organized in 2015 as the nonprofit "RISC-V Foundation," incorporated in Delaware, with 36 founding member companies. At the time, its base was in the United States, and debate over the location of its legal incorporation had not yet surfaced.
RISC-V International went through a three-stage transition: presenting a relocation-to-Switzerland concept at a summit in December 2018, formally announcing it on November 26, 2019, and completing re-incorporation as the nonprofit "RISC-V International Association" in Switzerland in March 2020. It was a decision made to institutionally guarantee its neutrality as a nonprofit organization. Calista Redmond, then CEO, explained, "We heard from members around the world that they couldn't feel secure with the legal entity remaining in the US," stating that no single country or company was the reason — but this period came right after the May 2019 Huawei export ban. Switzerland's long-standing selection as a base for international organizations reflects the country's tradition of political neutrality, and RISC-V International's choice of location can be seen as reflecting this same premise.
According to an analysis by Georgetown's CSET, the current US export control regime is built primarily around hardware and is ill-suited to targeting an architecture itself, and RISC-V International's incorporation in Switzerland further narrowed the range of regulatory options available to the US government. While export controls focus mainly on the transfer of hardware and technology, the US lacks sufficient means to regulate an abstract design blueprint in the form of a specification document — and this is where the significance of this institutional design lies. By combining a royalty-free instruction set that anyone can implement with a nonprofit governance structure based abroad, RISC-V became, in principle, unstoppable by any single company's license suspension or any single country's export controls. Membership has grown roughly 19-fold, from 236 companies in 2019 to over 4,600 in 2025. The full-time staff actually running this massive ecosystem remains quite small; what holds the real governing power isn't the size of the staff but the composition of the member companies gathered on the board and technical committees.
A Second Bet: Infiltration From Within the Governance Structure
On September 30, 2025, US Senator Mark Warner, in a letter to Kessler, an undersecretary at the Bureau of Industry and Security (BIS), noted that Chinese companies hold roughly half the seats on RISC-V International's board. The Entity List is a roster of companies designated by the US Commerce Department over national security concerns; exports of US-origin technology to listed companies generally require individual authorization. Phytium and Huawei, both flagged in the letter, are registered on this list yet reportedly hold seats on the board. Checking the roster on RISC-V International's official website as of August 2026 shows that, of 24 total board seats, 6 to 7 (roughly 25-29%) belong to directors affiliated with mainland Chinese companies or institutions — fewer than Warner's "roughly half" claim, but the core fact remains: Entity List-registered companies do hold board seats. The "unstoppable by anyone" institutional design apparently did not extend to protecting the inside of this board itself.
Within the standards body, the Technical Steering Committee (TSC) and its subordinate technical committees handle technical decisions such as which extension specifications get approved as official profiles and which technology areas take priority in the next roadmap, while the Board of Directors holds the final authority over ratifying specifications and organizational operations. It's a two-tier structure where member company representatives are involved both on the technical front and in the organization's ultimate decision-making — and beyond board seats, participation in the TSC and various technical committees also shapes substantive influence over the direction of standards. Even without holding IP, gaining a seat on these committees or the board allows one to exert influence over the direction of specifications — a practical form of involvement distinct from controlling a licensing model. China's control of this channel carries far greater long-term significance than shipping 2.5 billion chips. The RVA23 standard profile introduced earlier is itself a product of consensus-building through these technical committees and the board — and C930 and C950's rush to comply reflects an effort not to fall behind on this foundation.
When the Chinese government and state media describe RISC-V, the consistent framing is "zizhu kekong" (self-reliant and controllable) — a sanctions-resistance frame centered on technological independence unaffected by outside license suspensions or export bans. No official statement has been found acknowledging the reality of governance control — that China is engaging with the substance of the standards themselves through board seats. The likely reason China's official messaging says almost nothing about this participation in governance is that repeating only the sanctions-resistance narrative avoids clashing with the premise of neutrality — "not for any particular country" — used to justify the move to Switzerland.
This kind of concern didn't originate with Senator Warner. In 2023, a bipartisan group of lawmakers, including Republican Representative Mike Gallagher, sent a letter to Commerce Secretary Raimondo requesting consideration of export controls on RISC-V. As of 2026, while the Trump administration is pursuing a deregulatory approach in the semiconductor sector, Congress — as reflected in Senator Warner's letter — is growing more wary, with the administration and Congress not aligned in their stance.
The think tank CSIS concluded in an April 28, 2025 analysis that a US strategy of "disengagement" — withdrawing entirely from RISC-V's standardization process — would actually be counterproductive, and that maintaining influence requires continued active engagement. If US and allied companies withdraw, the relative weight of voice in standard-setting shifts proportionally toward the remaining players. The "deemed export" regulation that Senator Warner's letter calls for is an authority that could subject certain technical information to regulation if it's effectively deemed to have reached foreign national employees or board members — but as of August 2026, no export control legislation explicitly naming RISC-V has been enacted.
The nonprofit, foreign-based structure designed to shield RISC-V International from intervention by any specific country also equally prevents the US government from directly reshaping the internal balance of power through regulation. The only remaining countermeasure is for US and allied companies to keep participating on the board and technical committees, maintaining balance through numbers and presence. CSIS's call for active engagement reflects exactly this asymmetric dynamic — and this structure itself is what creates the impasse for the United States.
The State Strategy's Implementation and Where the Technology Stands Now
In March 2025, Reuters reported that eight Chinese government departments, including the Cyberspace Administration of China, were jointly drafting guidance to promote RISC-V nationwide. A summary article published by the Chinese Academy of Sciences in January 2026 states that this guidance was formally issued jointly by the eight departments sometime in 2025, positioned as China's first nationwide guidance on RISC-V chip development and elevated to the status of a "national strategic-level technology direction." The specific date of issuance and the full text of the provisions have not yet been identified through primary sources, but the situation has progressed from drafting stage to formal issuance. Financial backing is already in motion as well: the third phase of the "National Integrated Circuit Industry Investment Fund," a state fund for the semiconductor industry, was established on May 24, 2024 with registered capital of 344 billion yuan. This is a massive fund covering domestic semiconductor substitution broadly, and while it could serve as one foundation for driving capital into RISC-V-related companies in step with the issued guidance, the allocation of individual investments has not been disclosed.
Alibaba is not the sole bearer of this technology. The "XiangShan" processor, developed by a research team affiliated with the Chinese Academy of Sciences, recorded 16.5 points per GHz on SPEC CPU2006 according to figures announced by the Academy in March 2026, with Bao Yungang, deputy director of the Academy's Institute of Computing Technology, calling it "the world's strongest open-source RISC-V processor." A talent development program for RISC-V led by the Chinese Academy of Sciences reportedly involves over 27,000 participants from more than 1,100 universities, forming a two-pronged approach with commercial development at private companies proceeding alongside talent cultivation at academic institutions.
Despite this state-led push, Chinese semiconductor companies are not betting uniformly on RISC-V. Major CPU maker Loongson continues using "LoongArch," its own proprietary ISA announced in 2021, and has not joined the RISC-V camp. At the core of Loongson's stated reasoning, beyond avoiding licensing dependence on Arm or x86, is a concern that RISC-V, too, originates in the West and could someday carry control risk through its governance structure. China, then, is simultaneously running two distinct "self-reliant and controllable" tracks in parallel: the RISC-V strategy, which bets on an open international standard, and the LoongArch strategy, which relies on proprietary self-sufficiency with no dependence on any outside governance body whatsoever.
In April 2024, roughly 10,000 Loongson 3A5000-equipped PCs were trial-deployed to schools in Hebi City, Henan Province, and the company's high-performance processors (the specific model has not been officially disclosed) reportedly power the "Tianzhou-8" cargo spacecraft resupplying China's "Tiangong" space station. The scale of LoongArch camp deployment remains this limited for now. While the RISC-V camp is trying to break into the server market primarily through Alibaba, a major private cloud provider, the LoongArch camp's main battleground is the closed market of government procurement and national projects — making the two "self-reliant and controllable" strategies less like competitors and more like the state running a self-sufficiency track in parallel as a hedge, in case its bet on governance participation in an open standard proves uncertain.
Implications for Japan and the Next Point of Change
Renesas partnered with SiFive in April 2021 to jointly develop high-performance RISC-V for automotive use, and in March 2024 became the first in the industry to unveil a microcontroller (MCU) with a self-designed 32-bit RISC-V CPU core built in-house. What stands out is that the company is pursuing both external IP partnerships and in-house design in parallel, trying to bring RISC-V into the demanding automotive application space; Denso is also pursuing proprietary development for automotive and industrial equipment applications. Especially notable is the move by leading semiconductor foundry Rapidus, which partnered in November 2023 with Canada's Tenstorrent, a RISC-V-based AI chip maker, on IP core development, and in February 2024 welcomed Tenstorrent as the first customer under the framework of the NEDO-led research and development infrastructure LSTC, using it as a contract manufacturing client. Leading RISC-V design firm SiFive of the US also established its Japanese subsidiary SiFive Japan in August 2022, with Sam Rogan serving as president. Facing the same RISC-V standard, Japan's approach stands in stark contrast to China's.
Outside the corporate sphere, groundwork for research and adoption is also progressing. RISC-V Alliance Japan, founded in 2017 and incorporated in 2021, continues to hold its "RISC-V DAY TOKYO" event, while research labs such as Ryota Shioya's at the University of Tokyo and the "RISC-V Design Center" at the Institute of Science Tokyo advance processor research with NEDO support. Kyoto Microcomputer has released "SOLID" v4.0, a RISC-V-compatible development platform, showing that alongside company-level adoption, a research and tooling community is also building up.
There is no cross-government unified strategy here comparable to the guidance issued by China's eight ministries. It cannot be confirmed that RISC-V is explicitly positioned within the Ministry of Economy, Trade and Industry's industrial strategy in Japan; efforts remain at the level of NEDO supporting research at the University of Tokyo and the Institute of Science Tokyo, and individual companies' development infrastructure. This contrast — a state moving with an eye toward participation in governance structures in China, versus a Japan where companies each choose RISC-V based on individual practical judgment — illustrates how the same open instruction set can carry vastly different strategic weight. Japan's practical advantage lies in the fact that regardless of which camp ultimately becomes mainstream, leading-edge manufacturing hubs like Rapidus have already secured a position to capture contract manufacturing demand either way.
On RISC-V International's board, of 24 total seats, 6 to 7 (roughly 25-29%) belong to mainland Chinese companies or institutions, including Entity List-registered companies. What China is orchestrating is building influence through these seats to secure, in advance, a position that could sway which extension specifications become industry standard in the future.
CSIS's analysis holds that if the US withdraws from RISC-V's standardization process, the relative weight of influence shifts proportionally toward the remaining players, and the think tank recommends active engagement — maintaining and increasing participation on the board and technical committees. If US, European, and allied companies follow this recommendation and continue increasing their board seats, the outcome of this contest will likely settle into a practical, working-level competition over the balance of votes on the board. What determines the outcome ahead comes down to two things: how far China's eight-department guidance is actually implemented with concrete budget and enforcement power, and whether US concerns — symbolized by Senator Warner's letter — crystallize into effective legislation like "deemed export" controls. More than the figure of 2.5 billion units, it's the outcome of this contest over board seats that deserves the closest attention going forward.