NVIDIA has launched the NVIDIA Jetson T4000, a high-performance AI module designed to accelerate edge and robotics applications with advanced real-time reasoning capabilities. Built on the Blackwell architecture, the T4000 delivers up to 1,200 FP4 sparse TFLOPs of AI compute and features 64 GB of LPDDR5x memory with a 273 GBps bandwidth, offering a powerful balance of performance, efficiency, and scalability. Optimized for tight power and thermal constraints, the module is ideal for intelligent machines such as autonomous robots, smart infrastructure, and industrial automation systems. The Jetson T4000 includes one NVENC and one NVDEC hardware video codec engine, enabling real-time 4K video encoding and decoding. It shares the same form factor and pin compatibility with the higher-end Jetson T5000, allowing developers to design common carrier boards for both modules while accounting for differences in thermal and power requirements. Key specifications include a 12-core Arm Neoverse-V3AE CPU, support for up to 8 lanes of PCIe Gen5, and extensive I/O options including multiple UART, SPI, I2C, and CAN interfaces. The module operates within a 40W to 70W power range, making it suitable for production-ready edge deployments. In benchmark tests, the Jetson T4000 shows up to 2x performance improvement over the previous generation Jetson AGX Orin platform across a range of models. It delivers strong results for large language models (LLMs), text-to-speech (TTS), and vision-language-action (VLA) systems. For example, it achieves 218 tokens per second on Qwen3-30B-A3B and 1,100 tokens per second on Kokoro TTS, demonstrating its capability for real-time AI inference. The module runs on NVIDIA JetPack 7.1, the latest software stack for Jetson, which introduces critical enhancements for edge AI. A key addition is the NVIDIA TensorRT Edge-LLM SDK, an open-source C++ toolkit that enables efficient, low-latency inference of LLMs and vision-language models on edge devices. Unlike cloud-optimized Python-based frameworks, TensorRT Edge-LLM is designed for real-time robotics and embedded systems, with support for quantization to FP8, NVFP4, and INT4, reducing memory usage and latency while maintaining high accuracy. The SDK integrates seamlessly with existing C++ codebases, supports fast model deployment from PyTorch via ONNX and TensorRT optimization, and avoids the overhead of Python services and background processes—making it ideal for safety-critical, resource-constrained environments. JetPack 7.1 also introduces full support for the NVIDIA Video Codec SDK on Jetson Thor platforms. This unified SDK provides high-performance APIs for hardware-accelerated video encoding and decoding, enabling developers to build sophisticated perception and media pipelines with fine-grained control over quality, latency, and throughput. Features include CABR workflows for bitrate optimization, spatial and temporal adaptive quantization, and access to reconstructed frames for iterative encoding. The SDK includes both C-style APIs and higher-level C++ classes, along with sample applications and documentation. A Python wrapper, PyNvVideoCodec, is also available for developers seeking rapid prototyping. The Jetson T4000 is supported by a robust ecosystem of production-ready systems from NVIDIA partners. These platforms come pre-validated with power, thermal, and I/O designs tailored for robotics and industrial AI. With support for up to 16 lanes of MIPI CSI and GMSL, they enable multi-camera vision pipelines essential for autonomous navigation and inspection tasks. Developers can quickly move from prototype to deployment using the JetPack SDK, CUDA, and the full NVIDIA AI software stack. Many partner systems offer lifecycle support, regional certifications, and customization services to help scale from pilot to fleet operations. With JetPack 7.1 and the Jetson T4000, NVIDIA extends Blackwell-class AI, real-time reasoning, and advanced multimedia capabilities to a wider range of edge and robotics applications. The platform empowers developers to build next-generation physical AI systems with high performance, low latency, and software maturity. The Jetson T4000 is now available, and developers can get started with the Jetson AGX Thor Developer Kit and the latest JetPack 7.1. Comprehensive documentation, tools, and support are available through the Jetson Download Center and the NVIDIA Developer Forum.