Multiple delays in the production of Nvidia's flagship Thor smart driving chip have forced XPeng to reconsider its adoption for 2025 vehicle models. Wallstreetcn reports indicate that the setbacks are raising concerns about Nvidia losing key automaker partners.
According to 36Kr, Thor's mass production, initially slated for mid-2024, has been repeatedly delayed due to architectural issues. Sources familiar with the matter note that "Thor is now delayed to the point where competitors' in-house chips are ready."
XPeng accelerates Turing development
The production setbacks have compelled XPeng to abandon plans for Nvidia's Thor chip in its 2025 models, including the P7+, which now uses Nvidia's Orin. XPeng is accelerating the development of its proprietary Turing smart driving chip.
XPeng's Turing chip has reached the tape-out stage and is undergoing stability and performance tests. The company confirmed that its full-stack Navigation Guided Pilot (NGP) is now operational on its internally codenamed XP5 chip.
Nio opts for in-house solution
According to 36Kr, Nio has not ordered Nvidia's Thor chip for next year. Instead, the company announced in July that its NX9031 chip had entered the tape-out stage, with new models set to feature this in-house smart driving technology.
Nio's 2025 models will incorporate the NX9031 chip, Horizon Robotics solutions, and Nvidia's Orin, but will exclude Thor.
Both XPeng and Nio declined to comment.
Early adopters shift strategy
XPeng and Nio were early adopters of Nvidia's automotive AI chips. In 2020, XPeng's P7 used Nvidia's first-generation Xavier chip, delivering highway Navigation on Autopilot (NOA) with 30 TOPS of computing power.
Nvidia's Orin chip, released in 2021, was integrated into Nio's ET7 in 2022, using four chips to deliver a combined 1,000 TOPS of computing power.
While Nvidia's Orin is now a mainstream choice, Thor has struggled to secure XPeng and Nio's launch slots. Still, automakers like BYD, Zeekr, and Li Auto have announced plans to adopt the Thor chip, designed for AI and large models.
While positioned as an early adopter of Nvidia's Thor, Li Auto is developing its own "Schumacher" smart driving chip. The company is also advancing its next-generation end-to-end Visual-Language-Action (VLA) model, aiming to pair it with its 2026 in-house chip to lessen dependence on Thor.
Production issues compound market concerns
Nvidia's Orin chip remains its cornerstone automotive AI product. Nio CEO William Li revealed that Nio purchased 640,000 Orin chips in 2023, accounting for 46% of Nvidia's global shipments. With Nio selling 160,000 vehicles, each equipped with four Orin chips, Nvidia's total Orin sales are estimated at 1.39 million units.
China's smart driving sector is surging, with Nvidia's Thor automotive AI chip attracting attention. Launched at the 2022 Fall GTC, Thor offers 750 TOPS and 1,000 TOPS of computing power, securing partnerships with XPeng, Zeekr, BYD, and Li Auto.
Thor chip production, initially planned for mid-2024, has been delayed by at least a year. An XPeng P7+ executive noted that "2026 would be more realistic." The delays are tied to issues with Nvidia's Blackwell GPU architecture, designed for Transformers, LLMs, and generative AI workloads.
The Blackwell chip faces production setbacks, including yield issues with TSMC's 4nm process, challenges with CoWoS-L packaging, and thermal design flaws. While some issues are being resolved, production delays continue to disrupt automakers' vehicle development timelines.
In response to the delays, automakers like XPeng, Nio, and Li Auto are accelerating the development of self-developed smart driving chips, set for mass production by 2025. In-house chips boost efficiency and flexibility, following a path similar to Tesla's HW3, which remains upgradeable since its 2019 debut.
Intense competition has shortened product cycles to 12-18 months, with automakers demanding 10% annual cost cuts. Thor's delays and rising pressure are testing patience, as Li highlights the need to curb supply chain profit margins.
Market dynamics challenge Thor's position
While Thor meets the growing demand for high-performance automotive AI, domestic alternatives maturing in 2025 and stricter cost controls among automakers are posing mounting challenges for Nvidia.
Nvidia's Orin lineup includes the flagship Orin X (254 TOPS) for urban NOA and end-to-end driving, the entry-level Orin N (80 TOPS), and the lower-cost Orin Y (200 TOPS) at half the price of Orin X. Despite driving revenue, Nvidia's automotive segment still represents just 1% of total earnings.
Domestic firms like Horizon Robotics and Momenta are gaining market share. Momenta's end-to-end solutions have already been adopted by Mercedes-Benz, BYD Denza, and others, further intensifying competition as Nvidia struggles to catch up in this segment.
Despite setbacks, Thor remains a vital AI computing platform for automakers without self-developed chips, particularly those competing with Tesla. To solidify its position and grow automotive revenue, Nvidia must deliver a fully integrated hardware-software solution.
