Xiaomi finally did it. After ten years, RMB 13.5 billion in cumulative R&D, and a team that ballooned past 2,500 engineers, the Xuanjie O1 (玄戒 O1) is shipping inside the Xiaomi 15S Pro. I watched the May 22 launch from a Shenzhen cafe and my first thought was: okay, now prove it wasn’t a one-off.
The specs are solid on paper. Second-generation 3nm process, 19 billion transistors, a ten-core CPU in a quad-cluster layout topping out at 3.9 GHz. AnTuTu scores north of 3 million. The Immortalis-G925 GPU beats Apple’s A18 Pro on some benchmarks, according to IT之家. But I’ve seen enough launch-day slides to know that benchmark bragging and real-world thermal behavior are two different conversations.
Here’s what matters. Xiaomi is now the second Chinese phone maker, after Huawei, to mass-produce a flagship smartphone SoC. That is not a small thing. OPPO’s Zeku division had 3,000 people and still shut down in May 2023. The difference? Xiaomi has cars, AIoT, and a revenue stream that hit RMB 111.3 billion in Q1 2025, up 47.4% year-on-year. OPPO had phones and not much else. When the smartphone market sneezed, Zeku caught pneumonia.
The 15S Pro itself starts at RMB 5,499 for the 16GB+512GB model, or RMB 4,999 with the national subsidy (国补). The top 1TB trim is RMB 5,999. Xiaomi is limiting sales to authorized stores and forcing in-store activation, the same playbook Huawei used for the Mate 60. Scarcity marketing, or genuine yield constraints? Probably both. A former Xuanjie engineer told 36kr that the team saw roughly 20% turnover this year. When your low-level staff are walking out before launch, that tells you something about the pressure inside Songjiang’s R&D towers.
Let’s talk about the asterisks. The CPU uses ARM’s Cortex-X925, A725, and A520 cores. The GPU is ARM’s Immortalis-G925. The 5G modem is Mediatek’s, not in-house. Critics on Weibo immediately called it “not truly self-developed.” That argument is tired. MediaTek, Rockchip, and even early Kirin chips all licensed ARM IP. The hard part is not buying the architecture manual. It is floor-planning 19 billion transistors into 109 mm², getting the cache hierarchy right, and making sure the ISP does not choke on night-mode photos. Xiaomi designed the NPU, the ISP, and the display pipeline. That counts.
What I do not buy yet is the yield story. Second-generation 3nm means either TSMC N3E or Samsung 3GAP. Both are subject to US export controls for Chinese firms, at least on paper. How Xiaomi secured volume allocation is a question nobody is answering directly. My sources say the wafer supply is “sufficient for limited flagships,” which is PR speak for “do not expect this chip in a Redmi Note.”
Meanwhile, the broader chip picture in China is grinding forward. SMIC reported Q4 2025 revenue of RMB 17.8 billion, up 11.9% year-on-year, with capacity utilization at 95.7%. The company added 110,000 wafers per month of 8-inch equivalent capacity in 2025. But 87.6% of SMIC’s revenue now comes from China, up from already-high levels. The localization switch (产业链回流) is real, and it is accelerating. Whether that helps Xiaomi’s long-term cost structure depends on whether SMIC can get anywhere near 3nm before the decade ends. Right now they are still pushing mature nodes.
On the EV side, BYD just overtook Tesla in global pure-electric sales for 2025, delivering 2.26 million BEVs against Tesla’s 1.64 million. BYD’s total vehicle sales hit 4.6 million, putting it in the global top five for the first time, ahead of GM, Ford, Honda, and Nissan. The overseas number is what caught my eye: 1.05 million units, up 145%. Wang Chuanfu is building a fleet of roll-on/roll-off ships and factories in Thailand, Brazil, and Hungary. This is no longer a domestic play.
Back to Xiaomi. Lei Jun says the company will invest at least RMB 50 billion over ten years in semiconductors. The Xuanjie O1 won Xiaomi’s 2025 internal technology grand prize, a RMB 10 million award. That is a nice trophy. But the real test is the 2026 refresh. Can they ship a second-generation SoC with an integrated 5G modem? Can they get yields high enough to put Xuanjie into the Xiaomi 17 series, not just an anniversary special edition?
I’ve seen this movie before. Huawei’s Kirin 970 was mediocre. The Kirin 980 was good. The 9000 was great. Chip design is a game of iteration, and iteration requires volume. If Xiaomi can only slot Xuanjie into niche models, the economics will not work. The break-even estimate I have heard is around 10 million units. The 15S Pro will not get there alone.
One more thing. The SIA reported global semiconductor sales of $59 billion in May 2025, up 19.8% year-on-year. The Americas grew 45.2%, China 20.5%. AI demand is sucking up all the advanced capacity it can find. In that environment, a Chinese fabless startup fighting for 3nm wafer starts is like a guy trying to buy a train ticket during Spring Festival. Possible, but you had better know someone.
Xiaomi has the money, the ecosystem, and now the proof of concept. What it still needs is time. And wafer supply. And a modem team that does not burn out.