In a surprising move that has sparked much conversation in the automotive and technology sectors, Xiaopeng Motors, a prominent Chinese electric vehicle manufacturer, has quietly exited the partnership list of tech giant NVIDIAThis development unfolded on the opening day of the 2025 Consumer Electronics Show (CES) in Las Vegas, where NVIDIA's influential founder and CEO, Jensen Huang, unveiled the company’s latest list of collaborators in the automotive domainThe roster featured heavyweight brands such as Tesla, BYD, Jaguar Land Rover, Li Auto, Mercedes-Benz, Toyota, Rivian, Xiaomi Auto, Volvo, Lucid, and Zeekr, along with several companies advancing Level 4 autonomous driving technologies.

However, the absence of Xiaopeng Motors, which had previously announced its intention to develop its own intelligent driving chips, caught the industry's attentionLast year, Xiaopeng's P7 series vehicles had already incorporated NVIDIA's Orin chips, marking the brand as one of its early customersThus, the company's decision to break ties with NVIDIA has raised eyebrows, especially in light of previous collaborations.

Experts suggest that this shift may be closely linked to ongoing delays in the mass production of NVIDIA's chipsThe P7+ model from Xiaopeng, which was slated to incorporate NVIDIA’s latest technology, has not been able to do so due to these setbacksThis situation significantly hampers Xiaopeng’s product development timeline and market readinessConcurrently, the company is hastening its development of the self-designed Turing chip, reflecting its ambition to replace NVIDIA’s technology with its own.

Wang Peng, a researcher at the Beijing Academy of Social Sciences, emphasized that should Xiaopeng opt to wait for the mass production of NVIDIA’s Thor chip before its launch, it may push back the delivery of the P7 series even furtherThis not only affects the vehicle's market launch schedule but could also lead to missing critical competitive opportunities

Advertisements

The pressure from a "chip shortage," as revealed by various market analyses, is becoming increasingly apparent.

NVIDIA's Thor AI chip, designed for automotive applications, has faced repeated production delaysThis phenomenon has raised concerns about losing core clients like Xiaopeng Motors, which was among the first companies to mass-produce vehicles equipped with NVIDIA's first-generation AI chipsBack in 2018, NVIDIA debuted its initial smart driving chip, Xavier, which Xiaopeng successfully integrated into its P7 model by 2020, leveraging the chip’s impressive computing power to facilitate advanced driving features.

The automotive sector in China is surging forward with its fascination for smart driving technologies, and all eyes are on the latest generation of NVIDIA’s automotive AI chipsIntended to support cutting-edge technologies like end-to-end AI models, the Thor chip has garnered significant anticipationHowever, the ongoing delays in its rollout have subjected Xiaopeng’s new models to configurations that fall short of expectations.

The Thor chip was first unveiled at NVIDIA's GTC 2022 conference, boasting an astonishing computing capacity of 2,000 TFLOPS—double that of its predecessor, the Orin chipOriginally slated for mass production in 2024 and deployment in vehicles by 2025, the timeline has now shifted multiple times, with the latest updates pushing production to as late as 2027.

In light of these developments, Xiaopeng’s response has been to modify its vehicle configurationsAfter the launch of the P7+, the product team disclosed that there were initial plans to utilize the Thor chip, but due to its delay, they opted for a dual Orin-X setup insteadWhile this adjustment seemed to momentarily solve technical challenges, it also exposed Xiaopeng's heavy reliance on NVIDIA's chips, raising alarms about future vulnerabilities.

Xiaopeng’s product manager, Nick, even took to social media to express caution regarding the uncertain timeline for the SOP (Small Outline Package) of the Thor chip, noting that merely seeing it available by 2026 would be a considerable achievement.

In an interview, Zhang Xiang, secretary-general of the International Smart Transport Technology Association, articulated that Xiaopeng and NVIDIA’s deep collaboration entails several risks

Advertisements

NVIDIA's positioning as a leading supplier of smart driving chips has only heightened the dependency of carmakers upon its technology, nesting them into a supply-demand dynamic that could undermine their negotiating power in the supply chain.

Official data from Xiaopeng indicates strong market demand for its P7+ vehicles, with deliveries surpassing 20,000 units by January 10, 2025, after the model’s launch on November 7, 2024. Nonetheless, the delays associated with the Thor chip have created substantial obstacles for Xiaopeng’s original product strategiesThe magnitude of these setbacks compels the company to recalibrate its designs and developmental timelines to avoid falling behind in delivery schedules and industry competitiveness.

Transitioning away from NVIDIA, Xiaopeng Motors has embarked on an ambitious path toward self-developed AI chipsThis decision reflects a broader industry trend aimed at creating proprietary technology ecosystemsIn August 2024, Xiaopeng’s chairman He Xiaopeng announced the successful tape-out of its Turing chipFurther details revealed during the November tech day showcased that the Turing AI chip underwent extensive feature validation in a remarkably short timeframe, indicating a robust performance.

He Xiaopeng highlighted the significance of this shift, noting that competition in the smart driving arena is not just occurring on the cloud but also at the vehicle level, demanding higher computing powerPublic chips often lead to inefficiencies in resource management, prompting Xiaopeng to prioritize the in-house chip development strategy to elevate resource utilization and technological control.

Rumored to be the world's first chip capable of simultaneous applications in AI vehicles, AI robotics, and flying cars, the Turing chip integrates two proprietary neural processing units (NPUs) and two independent image signal processors (ISPs), utilizing a specifically designed domain-specific architecture to optimize neural network computations

Advertisements

Advertisements

Advertisements