Pony AI Debuts Nvidia-Powered L4 Domain Controller

Per Pony.ai's press release, Pony.ai announced a new-generation autonomous driving domain controller built on the NVIDIA DRIVE Hyperion platform and powered by DRIVE AGX Thor with NVIDIA NVLink, claiming a combined maximum computing performance of 4000 FP4 TFLOPS. The company described the platform as supporting single-chip and multi-chip configurations, multiple cooling solutions, and features intended to address L4 requirements including multi-sensor fusion, full-scenario perception and high-complexity scenario understanding (Pony.ai press release, Apr 25, 2026).

Per the announcement, the platform will use DRIVE AGX Thor SoCs linked with NVIDIA NVLink to provide high-speed, low-latency communication between chips. Pony.ai's release highlights a flexible portfolio spanning different compute tiers and cooling options and says the architecture is engineered for enhanced safety redundancy and deployment robustness (Pony.ai press release). Pony.ai also noted it previously scaled to a four-DRIVE AGX Orin X domain controller for its mass-produced seventh-generation robotaxi in 2025 (Pony.ai press release).

Editorial analysis: Companies building next-generation in-vehicle compute platforms are converging on multi-chip designs to accommodate larger foundation models for perception and planning. Such architectures deliver headroom for increased model size and sensor fusion, but industry-pattern observers note they typically raise integration complexity for thermal management, power budgeting, and vehicle-level safety validation.

Per Pony.ai's press release, cumulative shipments of the company's "Fangzai" domain controller rose by more than 500% year-over-year in 2025. PR Newswire carried Pony.ai CEO Dr. James Peng's direct quote: "Our collaboration with NVIDIA has supported several critical milestones in Pony.ai's autonomous driving journey," and NVIDIA's Rishi Dhall said, "Autonomous driving systems are rapidly increasing in complexity, driving the need for scalable, high-performance compute platforms." Reporting by StockTitan adds that Pony.ai is targeting more than 3,000 robotaxis and operations in over 20 cities by the end of 2026 and reports the company reached unit-economics breakeven in two Chinese markets (StockTitan reporting).

• •Deployment metrics: observers should track verified fleet counts, live-service city rollouts, and third-party telemetry for robotaxi operations to validate the StockTitan targets.
• •Thermal and power integration: engineering disclosure or teardown reporting that details vehicle-level cooling and power budgets for multi-Thor configurations will indicate practical feasibility for mass deployment.
• •Software/model support: published compatibility notes or SDK updates showing support for larger perception/planning models on DRIVE AGX Thor will clarify how compute headroom maps to application capability.

Editorial analysis: For AV system engineers and platform teams, the announcement underscores continuing demand for higher-performance edge compute to host larger perception and planning models. Integrating multi-chip SoC solutions typically shifts development effort toward hardware-in-the-loop testing, deterministic latency profiling across NVLink interconnects, and more rigorous fault-tolerant design to satisfy L4 safety envelopes.