Xpeng’s new chip powers its push toward full driving autonomy

In the race to develop end-to-end large models for autonomous driving, car manufacturers are picking up the pace on in-house chip development. Xpeng Motors has now pulled ahead by completing the tapeout of its self-developed artificial intelligence chip, named “Turing.” The announcement, made during the launch of Xpeng’s new Mona M03 model, confirms rumors that have been circulating for some time.

Sources familiar with the matter told 36Kr the Turing chip is designed to meet specific AI demands, supporting end-to-end (E2E) large models. It integrates cockpit and driving functionalities within a central computing architecture, delivering AI computing power said to nearly match three of Nvidia’s Drive Orin X chips.

Tapeout is a crucial checkpoint in chip development, marking the finalization of a chip’s design before it can be manufactured. Xpeng’s Turing chip hit this stage on August 23, and if upcoming tests go smoothly, mass production will soon be underway. Built for Xpeng’s needs, the Turing chip is a keystone in the company’s R&D strategy. CEO He Xiaopeng emphasized that this chip is vital for closing the loop on data handling for both autonomous driving and robotics.

This breakthrough also marks Xpeng as the second Chinese EV maker, after Nio, to make strides in self-developed chips.

Xpeng’s investment in chip development dates back to 2020. Today, the company’s chip team has grown to between 200–300 members. The focus on chip computing power is closely linked to Xpeng’s work on deploying E2E large models in autonomous driving. Back in May, Xpeng rolled out an E2E model built around three core components: the XNet neural network, XPlanner regulation and control model, and XBrain large language model (LLM).

Xpeng has set an audacious target: to rival the performance of Waymo’s robotaxi technology within 18 months. Achieving this goal requires substantial investment in both E2E large models and the chip platforms that support them, as Xpeng aims to leap from Level 2 to Level 4 autonomy.

Tesla, too, is navigating similar waters. In August, Tesla released version 12.5 of its Full Self-Driving (FSD) software, built on E2E large models. The new version boasts five times the parameters of its predecessor, V12.4, and was initially rolled out to models equipped with HW4, Tesla’s fourth-generation intelligent driving chip. Owners of HW3-equipped vehicles had to wait an additional two weeks for the update.

Elon Musk has admitted that Tesla needs more time to optimize its code for HW3, acknowledging that the hardware is nearing its limits. But Tesla isn’t stopping there. The company is already gearing up for its next-generation hardware, AI 5, which is expected to debut in December 2025 and be ten times more powerful than HW4. Given that HW4 is already five times more powerful than HW3, AI 5 promises a significant leap forward.

For Chinese automakers like Xpeng, the endpoint of E2E technology and chip computing power remains elusive, making it hard to predict a clear winner in the near term. This prolonged competition in both software and hardware is a true test of these companies’ commitment to sustained investment. Lang Xianpeng, vice president of smart driving R&D at Li Auto, told 36Kr in an interview that playing in the autonomous driving space is impossible without USD 1 billion in net profits.

This financial reality looms large for Xpeng. In Q2, the company sold over 30,000 vehicles. While the numbers are modest, Xpeng’s gross profit margin rebounded to 14%, a significant year-on-year improvement.

Volkswagen has been a key player in helping Xpeng restore its profit margins. Through joint procurement efforts, Xpeng’s per-vehicle cost dropped from RMB 240,000 (USD 33,758) in the first quarter to RMB 211,000 (USD 29,682) in the second. Additionally, Volkswagen’s contributions were crucial to Xpeng’s nearly RMB 1.3 billion (USD 182.9 million) in service-related revenue reported in its financial statements. With an estimated quarterly increase of around RMB 500 million (USD 70.3 million), Xpeng’s gross profit margin could see further gains.

The Mona model’s launch is also a crucial part of Xpeng’s strategy. With a presale price capped at RMB 135,900 (USD 19,106), it targets BYD’s core market segment. Xpeng expects Mona to achieve monthly sales of 10,000 units, helping the company hit 20,000 monthly vehicle sales.

Video featuring the Mona M03 electric sedan developed by Xpeng Motors. Video and header image source: Xpeng Motors.

Boosting vehicle sales while enhancing technical services is key to Xpeng’s continued investment in AI R&D, moving the company closer to its ultimate goal of full autonomous driving.