Pony.ai Inc. is deepening its collaboration with Nvidia Corp., announcing a next-generation autonomous driving controller with 4,000 trillion operations per second of computing power to accelerate its push for large-scale Level 4 robotaxi deployment. The new system, built on Nvidia's DRIVE Hyperion platform, will use two of the upcoming DRIVE Thor systems-on-a-chip (SoCs) linked together.
"The next-generation domain controller built on NVIDIA DRIVE Hyperion will be a key enabler for the continued evolution of our L4 autonomous driving products and help accelerate large-scale commercialization," Dr. James Peng, Founder and CEO of Pony.ai, said in a statement.
The new controller boasts a combined maximum performance of 4,000 FP4 TFLOPS, a substantial leap from the company's current-generation systems. Pony.ai's Gen-7 robotaxis, which began mass production in 2025, are powered by a domain controller using four Nvidia DRIVE AGX Orin SoCs. While a direct comparison is complex, a single Orin chip provides around 254 TOPS, highlighting the massive performance increase the dual-Thor setup will offer for processing complex AI models and sensor data from cameras, lidar, and radar.
The announcement signals a move to consolidate computing power and increase efficiency as the industry grapples with the high cost of L4 autonomous systems. By building on the DRIVE Thor platform, which is designed for automotive-grade applications, Pony.ai aims to create a scalable system that can be deployed across a wide range of vehicles and applications, from robotaxis to logistics and robosweeping, supporting its growing domain controller business.
The heart of the new platform is Nvidia's DRIVE Thor, the successor to the widely adopted Orin SoC. Thor integrates the Blackwell GPU architecture, designed for transformer and large language models, which are becoming increasingly critical for complex scenario understanding in autonomous driving. The use of NVLink to connect two Thor SoCs allows them to function as a single, powerful computer, enabling high-speed data sharing with low latency. This architecture is crucial for the multi-sensor fusion and real-time decision-making required for fully driverless operation in dense urban environments. Pony.ai expects the platform to support various configurations, including single-chip versions for less demanding applications, providing a portfolio of solutions for different market tiers.
This development is a critical step in Pony.ai's strategy to achieve profitability and scale its robotaxi service. The company has already reached unit-economics breakeven in two major Chinese cities and aims to expand its fleet to over 3,000 vehicles by the end of 2026. The increased computing power and potential cost efficiencies of the Thor-based system are essential for improving the safety, reliability, and operational domain of its autonomous fleet. The collaboration also puts Pony.ai in a strong position against competitors like Baidu's Apollo and autonomous trucking firm TuSimple, which are also developing their own high-performance computing solutions. The successful mass production and deployment of this new controller will be a key milestone to watch in the race to commercialize autonomous driving technology.
This article is for informational purposes only and does not constitute investment advice.
