NVIDIA Thor under Yole Group’s microscope

The automotive industry is undergoing a fundamental shift in computing. As vehicles evolve into software-defined platforms running ADAS and infotainment, the silicon at their core has never been more strategic.

In Yole Group’s Automotive Computing and AI 2025 report (2026 edition coming soon), analysts estimated that NVIDIA Orin captured approximately 60% of the automotive central processor market, ahead of competitors including Tesla, Huawei HiSilicon, and Horizon Robotics. Yole Group’s visit to the Beijing Auto Show 2026 confirmed that NVIDIA Thor is rapidly emerging as the new benchmark for ADAS computing. At the same time, the Chinese market, NVIDIA’s largest automotive market, is accelerating the development of in-house alternatives, including Xpeng Turing (See new Yole Group’s report: Xpeng Turing Central ADAS SoC), NIO Shenji NX9031, Huawei HiSilicon MDC, and Horizon Robotics solutions.

Designed to run end-to-end transformer AI models across cameras, radars, and LiDARs on a single SoC, NVIDIA Thor consolidates workloads that previously required multiple ECUs and processors.

In this article, Yole Group’s analysts Felix Rocheteau and Hugo Antoine provide an inside look at NVIDIA Thor through an in-depth teardown and silicon analysis of the Thor-U extracted from Xiaomi’s latest domain controller: NVIDIA -Thor SoC report.

This analysis is part of Yole Group’s broader automotive computing and AI activities and complements other advanced teardown investigations conducted in Yole Group’s dedicated laboratory, including those of Xpeng Turing and emerging Chinese automotive AI processors. Together, these analyses provide unique insights into the technologies, architectures, and competitive dynamics shaping the future of intelligent mobility. More information

Enjoy reading this new computing snapshot.

The chip analyzed in this NVIDIA Thor report is the Thor-U. It was extracted from the XIAOMI 4-in-1 Domain Controller for the Xiaomi YU7 and offers a representative view of NVIDIA’s latest high-performance compute platform deployed in a next-generation Chinese smart EV architecture.

NVIDIA Thor is based on TSMC’s 5 nm-class FinFET node and integrates a Blackwell GPU architecture, with up to 14 Arm Neoverse V3AE cores, Arm’s first automotive-grade Neoverse CPU IP based on the Armv9-A architecture.

The SoC delivers up to 1,000 INT8 TOPS, a 4x increase over the Orin-X (254 TOPS), and 273 GB/s of memory bandwidth within AEC-Q100 qualification constraints. Thor reflects a deliberate convergence between data center and automotive compute architectures.

NVIDIA offers two versions of its Thor SoC for automotive applications: the Thor-U at 730 INT8 TOPS, the version mainly used in vehicles today and analyzed in this report, and the Thor-X at 1,000 INT8 TOPS.

This floor plan analysis, detailed in Yole Group’s Thor SoC report, notably reveals an almost twofold increase in CPU area compared to Orin, bringing the GPU-to-CPU area ratio closer to parity, from a nearly 3-to-1 ratio on Orin.

The die measures 391.2mm², slightly smaller than its predecessor, Orin, at 429.2mm², yet the transition from Samsung’s 8nm node to TSMC’s 5nm-class process, combined with architectural improvements, accounts for the performance uplift. It nonetheless remains one of the largest dies used in automotive applications, almost twice the size of the Tesla FSD AI4. The clearest evidence of its performance uplift is that many OEMs have replaced two Orin SoCs with a single Thor.

There are no major surprises on the packaging side, still FC-BGA, confirm Yole Group’s analysts. But this time assembled in South Korea, whereas Orin’s packaging was assembled in Taiwan.

NVIDIA Thor marks a new era in automotive AI computing, combining data center-class performance with automotive-grade integration to power next-generation software-defined vehicles. As OEMs shift toward centralized architectures, competition is accelerating with emerging in-house solutions from Xpeng, Huawei, and NIO.

Through detailed teardown, floorplan, process, and packaging analysis, Yole Group’s NVIDIA Thor report delivers exclusive insights into one of the industry’s most advanced automotive SoCs.

👉 Contact Yole Group to explore NVIDIA Thor’s technology choices in greater depth and discover our latest teardown analyses of the most advanced automotive and AI processors shaping the future of mobility.

Felix Rocheteau is a Technology & Cost Analyst for Computing at Yole Group where he is a member of the Computing and Memory team. With more than two years of experience in the electronics industry, Felix specializes in the analysis of advanced processors, including SoCs, MCUs, and integrated circuits.

On a daily basis, he conducts reverse engineering and cost analysis of semiconductor devices and supports custom projects in close collaboration with Yole Group’s laboratory analysts. He also monitors technology developments, market trends, and manufacturing innovations in the computing and semiconductor industry, with a particular focus on the intersection of IC analysis and artificial intelligence.

Felix holds a specialization in biomedical and embedded systems, along with a Master’s degree in Electronic and Numeric Technologies.

Hugo Antoine is a Technology & Market Analyst, Computing and Software at Yole Group.

Hugo develops technology & market analyses covering computing hardware, software, and Artificial Intelligence (AI).

He holds a master’s degree from Ecole des Mines de Saint-Etienne (France), with a focus on microelectronics and computing at the Centre of Microelectronics in Provence (France). In addition, he pursued an AI specialization at Ecole Polytechnique de Montreal (Canada). Furthermore, he completed a dual-degree program in innovation management at emlyon business school, highlighting his expertise at the intersection of technology and business.

Related technology, process, and cost reports:

Related market reports:

Source: www.yolegroup.com