What did Xiaomi unveil about its Xring D100 autonomous driving chip—including its 3 nanometer manufacturing process, 20 core CPU, 16 core NPU, support for up to 160 GB of unified memory and on device AI models with up to 200 billion parameters, its claimed status as mainland Chin

Xiaomi unveiled the Xring D100 as its first in-house “smart-driving” AI chip: a 3 nm design intended for commercial deployment in 2027 after completing development and validation. The announcement signals a push to own more of the silicon and software stack across Xiaomi’s consumer devices, AI systems, computers, and EVs—not yet a demonstrated performance win over established automotive platforms. 1012

• D100 specifications and claims: Xiaomi says the chip has a 20-core high-performance CPU, 16-core NPU, and support for up to 160 GB of unified memory. It says this enables local execution of AI models with up to 200 billion parameters. 12

• Process-node claim: Xiaomi calls D100 mainland China’s first high-compute smart-driving AI chip built on a 3 nm process. That is Xiaomi’s claim; the company did not disclose a TOPS figure, the foundry, a first vehicle model, or a precise vehicle-installation timetable beyond planned 2027 commercialization. Consequently, its raw inference performance cannot yet be compared credibly with rivals’ published TOPS ratings. 1012

• AI Cube demonstration: Xiaomi also showed a 150 W AI Cube Prototype combining three Xring chips: the O3 mobile SoC, O100 AI accelerator, and D100 automotive processor. The demonstrated unit reportedly had 80 GB unified memory and ran a local “120B + 3B” model configuration; it was an engineering prototype, not a commercial product announcement. 5

Competitive context

• Nio: Its 5 nm Shenji NX9031 is already deployed in Nio and Onvo vehicles, with more than 300,000 units shipped; Nio has previously claimed over 1,000 TOPS per chip. D100’s 3 nm node is nominally more advanced, but Xiaomi’s undisclosed TOPS, power, bandwidth, safety architecture, and real-world driving results prevent a meaningful performance ranking. 7

• Li Auto: Li Auto’s 5 nm M100 has a reported 1,280 TOPS figure, giving it a disclosed throughput benchmark that Xiaomi has not supplied. 67

• BYD: BYD’s 4 nm Xuanji A3 is reportedly already in mass production, so BYD holds an execution and deployment advantage even if D100’s claimed node is smaller. 6

• XPeng: XPeng is likewise pursuing proprietary autonomous-driving silicon, making Xiaomi part of a wider Chinese EV-industry move to reduce reliance on merchant processors. Publicly comparable details remain uneven across these companies. 6

• Nvidia and Tesla: Xiaomi’s current EVs have used Nvidia hardware, and D100 is positioned as a route to reduce that dependence from 2027. 11 Strategically, the effort resembles Tesla’s in-house FSD-computer approach: integrate vehicle software, data, models, and compute rather than buy the entire platform externally. Nvidia, however, remains the better-established supplier ecosystem and benchmark platform; Xiaomi has not released enough D100 performance, efficiency, software-toolchain, or safety-validation data to establish parity.

What it says about Xiaomi’s strategy—and risk

• Xiaomi has invested more than RMB 21 billion in semiconductors and built a chip team of nearly 3,000 engineers. 12
• The spending comes while its Smart EV, AI and other new-initiatives segment recorded an RMB 2.6 billion operating loss in the second quarter of 2026. 9
• The rationale is vertical integration: use common in-house Xring capabilities across smartphones, local AI, home/desktop computing, and cars, potentially improving feature integration, supply control, and long-term unit economics. Reuters reported that Xiaomi is betting greater control of key components will strengthen its business. 1
• The principal caveat is execution: D100 is validated but not yet shipping in production vehicles, and a 3 nm process designation alone does not establish autonomous-driving capability, power efficiency, reliability, tool support, or cost competitiveness.