BYD reveals first self-driving chip developed in-house

China’s largest electric car seller joins other Chinese brands in strategy to reduce dependence on Nvidia

unveiled its first in-house developed computing chip, joining a growing group of Chinese EV (electric vehicle) companies designing proprietary chips for autonomous driving, aiming to reduce dependence on external suppliers.

BYD President Wang Chuanfu presented on Thursday (May 28, 2026) the 4-nanometer Xuanji A3 chip, which has a processing capacity of around 700 TOPS. Although mass production has already begun, Wang did not specify which vehicle model will use the chip nor provide a timeline for implementation.

The initiative highlights a broader shift among Chinese EV automakers to secure supply chains and increase profit margins through the development of proprietary chips, challenging global semiconductor giants like .

BYD is the latest Chinese automaker to enter the custom chip race. It began deploying its in-house developed Turing chip in July 2025, migrating almost all of its major products to the proprietary chip by April 2026. The Shenji chip debuted in its vehicles in March 2025, with plans to use it in more cars. And recently launched an SUV model with its own Mahe 100 chip.

Custom chip development has traditionally been a high-cost, capital-intensive endeavor reserved for well-funded automakers. Generally, Nvidia processors are very popular among automakers, while Qualcomm and national companies such as AEA Horizon Robotics also capture a share of the market.

Because semiconductor developers typically amortize R&D costs through large shipment volumes, automakers often have difficulty justifying the investment unless they can achieve scale.

One startup founder estimated that an automaker needs to deploy at least 1 million chips to make in-house development economically viable. BYD easily reaches this level. The company sold more than 4.6 million new energy vehicles in 2025 and its sales exceeded 1 million units in the first 4 months of 2026.

Even at lower volumes, proprietary chips can provide significant cost benefits. Nio President William Li previously stated that his company has spent more than $300 million on Nvidia chips in 2024.

Nio Chief Financial Officer Yu Qu said in June 2025 that using the company’s internally developed chips in upgraded models had reduced costs by about 10,000 yuan ($1,477) per vehicle.

In addition to cost reduction, automakers emphasize the advantage of developing hardware and software simultaneously. XPeng President He Xiaopeng said commercial chips include redundant features to serve diverse customers, while proprietary chips can be customized specifically to work with a company’s internal AI models more efficiently.

In response to skepticism about automakers’ high investments in chip development, Li Auto Chairman Li Xiang said in May that competition in the AI ​​era is no longer about isolated components, but rather about systemic capabilities that integrate chip architecture, operating systems and algorithmic models.

BYD has a long history in chip development. Its semiconductor team, founded in 2002, has evolved into BYD Semiconductor, which currently employs more than 7,000 R&D employees and operates 5 wafer fabs, including the largest 12-inch wafer fab for the automotive sector in mainland China.

However, BYD’s wafer fabs primarily focus on less advanced manufacturing processes for power semiconductors. Because computing chips like the Xuanji A3 require cutting-edge manufacturing nodes, BYD designed and tested the chip in-house but outsourced its production to an overseas foundry.

Historically, BYD has lagged the competition in smart driving technology, prompting it to launch an aggressive upgrade campaign in February 2025. Its smart driving systems currently integrate technologies from third-party vendors such as Momenta. The company has not clarified whether its new chip will be supplied exclusively with its own software or whether it will be compatible with third-party systems.


This report was originally in English by Caixin Global on June 1, 2026. It was translated and republished by Poder360 under mutual content sharing agreement.