Xiaomi builds its own 3nm AI chip to give future EVs powerful onboard intelligenceXiaomi has unveiled its first self-developed high-performance chip for intelligent driving, marking a major expansion of the Chinese technology giant's push to design the critical computing hardware behind its growing electric vehicle business.Called the Xring D100, the new processor is built on a 3-nanometer process and is designed specifically for automotive AI workloads. Xiaomi says it is China's first high-computing-power intelligent-driving chip based on a 3 nm process, and the company plans to begin commercial use in 2027.The announcement places Xiaomi alongside Chinese automakers including BYD, Nio, Li Auto and XPeng, all of which have been developing proprietary processors for increasingly sophisticated driver-assistance and autonomous-driving systems.A 3 nm brain for Xiaomi's future carsAccording to reports, the Xring D100 features a 20-core high-performance CPU and a 16-core neural processing unit, or NPU. It also supports up to 160 GB of unified memory.AdvertisementAdvertisementPerhaps more significantly, Xiaomi says the chip can support the local deployment of AI models containing up to 200 billion parameters. That could give future Xiaomi vehicles considerably more onboard AI capacity, reducing the need to send certain computing workloads to external servers. Large models could potentially be used to process information from cameras and other sensors while supporting Xiaomi's increasingly ambitious intelligent-driving software.However, Xiaomi has not yet disclosed one of the D100's most important performance figures. Its computing power in TOPS, or trillions of operations per second, remains undisclosed. Estimates placing the chip between roughly 700 and 1,000 TOPS remain unconfirmed.Xiaomi moves beyond NvidiaXiaomi's existing electric vehicles have relied on Nvidia hardware for their intelligent-driving systems. The company's SU7 and YU7 models have used Nvidia's automotive computing platforms, but the D100 could eventually give Xiaomi greater control over a much larger part of its technology stack.AdvertisementAdvertisementThe advantage of designing both the hardware and software is not simply independence from a supplier. Xiaomi can theoretically optimize the D100 around the data flows, memory requirements and AI models used by its own intelligent-driving system. The company has also said that controlling the chip, domain controller and underlying software stack could help reduce component costs as production volumes increase. That kind of vertical integration is becoming an increasingly important strategy across China's EV industry.China's automakers are entering the silicon raceBYD introduced its 4 nm Xuanji A3 intelligent-driving chip earlier this year, while Nio, Li Auto and XPeng have also developed proprietary automotive processors. XPeng, for example, is using its in-house Turing chips in vehicles and has deployed four of them in its GX robotaxi platform, delivering a claimed 3,000 TOPS of onboard computing power.But developing custom silicon is expensive. Xiaomi says it has invested more than 21 billion yuan, or roughly $3.1 billion, in its semiconductor program and now has a chip team of nearly 3,000 people.AdvertisementAdvertisementThat scale of investment means self-designed chips may remain an option primarily for automakers with sufficiently large vehicle volumes and deep financial resources.The next challenge: putting it on the roadThe D100 has reportedly completed validation, but Xiaomi has not yet identified the first production vehicle that will use it. Designing a powerful chip is only the beginning. It must still prove reliable across automotive operating conditions and be deeply integrated with the vehicle's sensors, software, and safety systems before reaching customers.If Xiaomi meets its stated 2027 commercialization target, the D100 could become a key part of the company's effort to control everything from the AI model to the silicon running it. The race to build smarter cars may increasingly be decided not just by who builds the best vehicle, but by who builds the best computer inside it.
+73 others
