The semiconductor industry's most contentious topic recently centers on Xiaomi's launch of its 3nm SoC chip—the Xuanjie O1.

On May 22nd, when benchmark scores for Xiaomi's Xuanjie O1 chip surfaced showing performance approaching Apple's A18 Pro, the prevailing public discourse wasn't about the chip's impressive capabilities. Instead, it ignited a fierce online battle over the "authenticity of Xiaomi's self-developed chip" claims.

Critics unearthed deleted posts from ARM's official website, questioning whether the Xuanjie O1 was merely "minor modifications to ARM's reference architecture." Others countered: 99% of global smartphone chips rely on ARM—does that make all mobile processors ARM's exclusive domain? Some went further, circulating doctored images to falsely claim that Huawei's Kirin chips were also based on ARM-provided design architectures.

Despite Xiaomi's continuous efforts to dispel these rumors, public opinion proves difficult to reverse, as people's perspectives vary based on their underlying beliefs.

Suddenly, ARM—this British company that never manufactures chips—found itself at the center of the "authentic vs. fake chip development" controversy.

What gives ARM such influence?

ARM: The "Android" of the Chip World

source: ARM China

To understand this controversy, we must start with the "language" of chips—the instruction set.

If we compare a chip to a precision machine, the instruction set serves as its "genetic code"—it defines which commands the chip can understand and how it executes those commands.

Take a simple calculation like "1+2." In ARM's instruction set, this translates into binary code (such as 10001011000), instructing the chip's transistors on how to switch and compute.

Assembly language acts as the "intermediate translator" between human engineers and machines, converting "1+2" into instructions like "ADD X9, X10, X11," which compilers then transform into machine language.

ARM architecture's dominance fundamentally stems from defining a globally universal set of "language rules."

ARM's business model represents textbook-level monopolization: it doesn't sell chips, only design blueprints.

For every four chips sold globally, one carries ARM's "DNA"—from Apple's A-series to Huawei's former Kirin lineup, from MediaTek's Dimensity to Qualcomm's Snapdragon, without exception.

This "asset-light, high-penetration" model has given ARM over 95% market share in mobile devices (2024 data), earning it the moniker "the Android system of the chip world."

But why can't everyone bypass ARM?

ARM's Three Pillars of Dominance

Starting with cost considerations, I've recently observed extensive online discussions about the expense of developing a SoC chip and whether Xiaomi's R&D budget suffices.

Comparing the most accessible data points: Xiaomi's Xuanjie O1 requires single tape-out costs in the hundreds of millions, with four years of cumulative investment reaching 13.5 billion yuan. Huawei HiSilicon invested a decade and burned through 160 billion yuan to establish its foothold in the premium market. (Note: HiSilicon's chips extend beyond Kirin and encompass multiple chip varieties—please interpret these comparisons rationally.)

Currently, most mainstream flagship SoCs utilize TSMC's 3nm process, which achieves only 55% yield rates. According to Siemens' 2024 statistics, each chip tape-out has merely a 14% success rate.

This demands that companies ensure their chip designs are mature and reliable before beginning design work, rather than continuously failing tape-out validations (Xiaomi's previous Surge S2 reportedly failed due to excessive tape-out failures).

ARM serves not merely as an instruction set developer but also as a microarchitecture developer. They package and sell the latest architectures and instructions to chip design companies like Qualcomm, Apple, MediaTek, Huawei, and Xiaomi.

ARM's reference architecture licensing fees start as low as $0.50 per chip (approximately 3.5 yuan), effectively offering companies a ticket to the premium chip club for the price of a cup of coffee.

This makes ARM particularly attractive to the market.

Note: ARM offers two licensing models. The first provides access to all latest instruction sets and microarchitectures with per-chip royalties. The second offers perpetual licensing for specific instruction set versions, allowing companies to develop their own microarchitectures based on those instruction sets.

Companies do attempt self-developed microarchitectures—Apple, Qualcomm, and Huawei have all tried—but "failures" and "inability to circumvent ARM patents" remain commonplace. Previously, Qualcomm acquired a company with perpetual ARM licensing, leading to a lawsuit with ARM that Qualcomm ultimately won.

"Completely self-developed instruction sets? That's equivalent to demanding automakers invent the wheel before building cars." In reality, no one pursues this path unless absolutely necessary.

Would you choose so-called "true self-development, complete self-development, pure self-development" with continuous tape-out failures and ultimate abandonment, or accept being labeled an "assembly plant" (which would categorize Apple, MediaTek, and Qualcomm similarly)?

Cost represents one factor, but ARM's foundation also stems from its massive customer base.

Take the Armv9.2 architecture powering the Xuanjie O1. Its memory security mechanisms and AI instruction set optimizations have been validated across hundreds of millions of devices. Even Apple, despite its strength, must refine microarchitectures based on ARM instruction sets.

ARM's dominance derives not only from technology but from weaving a global ecosystem network.

Manufacturing benefits from TSMC's 3nm process optimizations, design leverages Synopsys' mature EDA tool compatibility, and systems integrate with Google's Android adaptations—all these giants serve as ARM's "allies."

When users choose ARM architecture, they're actually accessing a technology alliance encompassing thousands of companies—from chip design to software adaptation, every link has been pre-established.

This reveals how substantial ARM's barriers truly are.

But this raises a question: if ARM architecture is so pervasive, how can Chinese chip companies "start from scratch"?

Chinese Chips' Dilemma and "Alternative Solutions"

Today, x86 and ARM instruction set architectures remain the two dominant instruction set architectures, while MIPS, ALPHA, SPARC, POWER, and other instruction set architectures have declined.

The emerging open-source instruction set architecture RISC-V has become the preferred choice—as an open-source architecture, it requires no licensing fees and allows companies to freely modify designs. Alibaba's T-Head Yitian 710 server chip, Huawei's IoT processors, and even Qualcomm's experimental projects all attempt to use RISC-V to circumvent ARM's patent walls.

This sounds simple—just switch architectures, right? Reality proves more challenging.

Currently, RISC-V's ecosystem maturity is less than one-tenth of ARM's, concentrated primarily in IoT and other low-computing-power domains. Even Alibaba must incorporate ARM instruction set compatibility in RISC-V chips to support Android systems.

"It's like installing a Linux virtual machine in a Windows system—it never feels quite right."

While using RISC-V architecture means avoiding starting from scratch and can leverage existing software and hardware ecosystems for rapid development while meeting certain domestic autonomous and controllable requirements, Chinese companies have another approach: true "self-developed instruction sets."

Examples include Huawei's exposed "Lingxi instruction set" and Loongson's "LoongArch instruction set," both attempting to forge new paths outside x86 and ARM ecosystems.

However, this technology often requires building software ecosystems from zero, even forcing developers to rewrite code from scratch to catch up with others. Given such difficulty, why pursue this path?

Globalized division of labor represents the best direction for the semiconductor industry's collective development. No one wants to spend "unnecessary money" reinventing existing "wheels."

Unfortunately, as Loongson founder Hu Weiwu stated: whether x86, ARM, RISC-V, or the MIPS instruction set architecture Loongson previously adopted, all are foreign instruction set architectures. Particularly under the backdrop of US trade and technology wars against China, all carry uncontrollable risks.

Currently, based on cost and ecosystem considerations, we cannot completely eliminate dependence on x86 and ARM.

But this doesn't mean we shouldn't use available options, nor does it mean we shouldn't continue pursuing autonomous R&D when capable. We must pursue both paths—they're not contradictory.

Instruction systems and basic industry (process materials and equipment) constitute the two most important foundations of the information industry. We have reason to believe globalization remains the trend, but we won't miss opportunities to master "root technologies" ourselves.

Just as we can write articles in English for external output, we cannot develop our own national culture based on English. Similarly, we can use foreign instruction systems to create products, but we cannot expect to build our own ecosystem using foreign instruction systems.

Conclusion

Returning to the Xiaomi Xuanjie controversy, the core issue may not be "whether ARM was used" but rather "how to define autonomy within ARM's framework."

Apple uses ARM reference architecture but achieves consumer electronics industry profit leadership through hardware-software synergy. Huawei, once restricted from accessing the latest ARM instructions, still built differentiation barriers through self-developed modules.

True "self-development" shouldn't involve shouting "closed-door development" slogans in self-admiration, but rather striving to excel in autonomous design, thereby securing irreplaceable ecological positions in global division of labor. After all, whoever becomes difficult to replace in ecosystem chains controls the discourse.

Like NVIDIA, TSMC, Google, and Microsoft, none achieve 100% full-chain self-development, and none can thrive independently outside global ecosystems. Yet they dominate their respective fields as industry leaders, wielding global influence.

Adding to Xiaomi's momentum, the company announced on May 27th that both revenue and profits hit all-time highs in Q1 2025. Total revenue reached 111.3 billion yuan (historic high), up 47.4% year-over-year. The smartphone and AIoT division generated 92.7 billion yuan (up 22.8%), while the smart electric vehicle and AI innovation division contributed 18.6 billion yuan. Adjusted net profit soared to 10.7 billion yuan—another record high—representing a 64.5% year-over-year increase.

This financial performance underscores a crucial point: in 2024, Xiaomi's global smartphone market share ranked third (14%, trailing Samsung's 20% and Apple's 18%, with vivo and OPPO at 8% each). Domestically, it sold 42 million units, internationally 120 million units. Xiaomi is a latecomer, but as I mentioned in my previous article, latecomers always have opportunities to challenge—and these numbers prove it.

Ultimately, Chinese chips started late, destined to be "dancing in shackles." However, this debate over "authentic vs. fake self-development" may prove more worthy of contemplation than the Xuanjie chip itself.