RISC‑V: The paradigm shift and the difference between Silicon Valley
Explore how RISC-V challenges traditional CPU architecture, and analyze its implications for the paradigm differences between China's Internet and Silicon Valley
Looking at the differences between China's Internet and Silicon Valley from the rise of RISC‑V
RISC‑V is like China's independent gateway, forcing Western CPU giants to reshuffle their cards. Traditionally, the CPU market has been monopolized by two closed instruction sets, ARM and x86, and the ecological chain revolves around patent fees; while RISC‑V opens the door with an open instruction set, directly reducing the "design cost" from hundreds of millions to several million, injecting new blood into the high concurrency and low latency needs of China's Internet. I bet that China's core computing power will be dominated by RISC‑V in the next five years because it takes back "control" from the United States.
At the technical implementation level, the openness of RISC-V is reflected in three key points: the instruction set itself can be directly forked on GitHub and can be freely implemented by any manufacturer; the complete tool chain (LLVM, GCC, Spike) has been implemented in 2021. Full platform compilation has been achieved; and the cost of silicon implementation-in 2023, SiFive's E21 core costs about US$0.3 per chip on a 28nm process, which is about 40% lower than the same generation ARM Cortex-M0. Among global new designs in 2022 - 2024, the proportion of RISC‑V has climbed from 5% to 12%, and in the China market, it has exceeded the threshold of 20%.
At the product level, the most direct beneficiaries are domestic cloud computing and content platforms. In 2021, Alibaba's Dharma Institute announced its self-developed "Black Iron" series. AI inference cards based on RISC‑V are 12% faster than ARM cards of the same generation at the same power consumption, while the procurement cost of Cloud Virtual Machine has dropped by about 15%; ByteDance's recommended system-specific chips invested in 2022 also use RISC‑V, increasing the throughput of a single machine by 18%. Behind these numbers lies a common pain point: highly concurrent user portrait computing requires a large number of customized operators, while licensing fees and technical barriers to closed instruction sets force business costs to rise. RISC‑V has transformed "software and hardware collaboration" from buyout to joint innovation.
On business accounts, RISC‑V is not driven by a giant, but by a group of "small money" scattered among funds, companies and universities. In 2022, China's direct investment in RISC-V will exceed US$250 million, exceeding the US$190 million invested by the United States in the same year; and most of these funds come from national integrated circuit projects and internal funds of companies such as Alibaba and Huawei. The business model has also shifted from "authorization +IP" to "service + customization". For example, the "soft core +IP+ maintenance" package provided by Huawei's Hisilicon department has an annual fee of about 1 million yuan, but it allows customers to complete the closed loop from design to mass production within one year. For Silicon Valley, the traditional patent revenue model is no longer an absolute advantage in the China market.
Golden Sentence
The era of > behind closed doors is over, and the open source instruction set is the new track.
Technology Implementation and Ecological Chain
From an ecological perspective, the success of RISC‑V is inseparable from the "open source + collaboration" mechanism. In 2020, the Linux kernel marked RISC‑V as an official support platform for the first time in its official release. Subsequently, mainstream frameworks such as Qt and TensorFlow completed the porting of RISC‑V in 2021. In 2023, the "RISC-V Software and Hardware Complete Practice Manual" jointly released by domestic universities has been adopted by more than 30 universities, cultivating tens of thousands of engineers familiar with the open instruction set. In contrast, ARM's training resources in China are mainly concentrated in the hands of a small number of authorized partners, with high costs and narrow coverage.
Opposing views often amplify the risks of RISC‑V-commercial cases where the ecosystem is not mature enough, the tool chain is not stable enough, and there is a lack of high-end CPUs. Indeed, actual shipments on the mobile side at the beginning of 2021 are still less than 1%, and the prices of IP providers in the high-performance field (such as SiFive's U74) are still higher than those of the same class ARM Cortex‑A55.
I don't deny these shortcomings, but it would be too conservative to regard them as "insurmountable gaps." In the past two years, the number of optimizations of LLVM in RISC‑V has tripled, and the compilation success rate of GCC has increased from 85% to 98%. Moreover, the domestic "Xuan Tie X1" has already scored TOP‑5 in a single card in the Beijing AI Computing Power Competition. The maturity of the technological ecosystem is an exponential curve. The key lies in the scale of the open source community and the continuous irrigation of funds, which is the unique advantage of the China market.
Cross-domain analogies can help understand: The rise of RISC‑V is similar to the Web browser competition of the 1990s. Netscape quickly occupied the market with its open plug-in system, and was eventually suppressed by Microsoft's Internet Explorer with closed standards; but then Chrome regained its dominance with the open source Blink engine. The competition for instruction sets is essentially a game of "standardization versus closure". Whoever can involve more developers, chip manufacturers, and end users will create a network effect. RISC‑V is on this path, and its community is inherently global-from Huawei Labs in Beijing to the RISC‑V Foundation in San Francisco, everyone contributes code on the same source code.
I make a bet: Before 2027, at least 30% of China's mainstream Cloud Virtual Machine will be customized chips based on RISC‑V, and Silicon Valley CPU suppliers will have to renegotiate licensing fees in front of China giants such as Huawei and Alibaba. The days of building cars behind closed doors are over. The real competition lies in who can turn "openness" into "advantages."
Golden Sentence
In Silicon Valley, code is the lock; in China, RISC‑V is the key.