BYD Launches 4nm AI Chip: On Par with NVIDIA in Process, Outperforms Tesla in Compute
BYD unveiled its first self-developed 4nm automotive-grade smart driving chip, Xuanji A3, achieving over 2100 TOPS with three chips combined. The dedicated NPU architecture offers 20% lower power per unit and 100% higher compute utilization compared to general-purpose GPUs. BYD also promises full compensation for accidents during city navigation.
Article intelligence
Key points
- BYD unveils fully self-developed 4nm smart driving chip Xuanji A3
- Dedicated NPU delivers 20% lower power and 100% higher compute efficiency
- Company promises unlimited accident compensation for city navigation
- Chip designed for broader physical AI applications beyond automotive
Why it matters
This matters because BYD unveils fully self-developed 4nm smart driving chip Xuanji A3.
Technical impact
May affect model selection, inference cost, product capability, and evaluation benchmarks.
BYD has officially entered the high-stakes AI chip race with the launch of its Xuanji A3, China's first automotive-grade 4nm smart driving chip. Unveiled by founder Wang Chuanfu, the chip marks a significant leap for the company, which was previously known primarily for electric vehicle powertrains. The Xuanji A3 is fully self-developed, from design to testing, and is already in mass production.
At the heart of the Xuanji A3 is a dedicated Neural Processing Unit (NPU) architecture, optimized for AI inference rather than the general-purpose GPU designs used by competitors like NVIDIA. Three chips combined deliver over 2100 TOPS of computing power, but BYD emphasizes that raw TOPS numbers don't tell the whole story. The chip offers 20% lower power consumption per unit of compute and a 100% improvement in compute utilization compared to generic GPU-based solutions. This is achieved by hardwiring common AI operations such as matrix multiplication and convolution, reducing overhead from general-purpose computing.
The chip features a 16-core CPU with 420K DMIPS for global scheduling, 273 GB/s memory bandwidth, and native support for Transformer large models. The dedicated NPU enables nanosecond-level data scheduling, significantly reducing latency in critical driving scenarios. In demonstrations, vehicles equipped with the Xuanji A3 exhibited smooth handling in complex urban environments, such as narrow streets and unexpected obstacles.
BYD's approach represents a strategic differentiation: instead of competing on theoretical peak performance, the company focuses on real-world effective compute. This efficiency allows the Xuanji A3 to not only meet current L2+ requirements but also to prepare for future L3/L4 capabilities, pending regulatory approval. Wang Chuanfu also made an unprecedented promise: BYD will fully compensate for any economic losses caused by its smart driving system during city navigation, with no cap. This bold guarantee underscores the confidence derived from the chip's robust architecture.
Beyond automotive applications, BYD positions the Xuanji A3 as a potential general-purpose computing platform for the physical AI era, including robotics, drones, and industrial automation. The chip's low latency, high efficiency, and support for large models make it suitable for any AI system that must perceive and act in the physical world. With vertical integration spanning chip design, manufacturing (through five wafer fabs), and a 7,000-strong chip R&D team, BYD has built a formidable moat. The Xuanji A3 is not just a smart driving chip; it could define the computing standard for the next wave of AI.