What Is NVIDIA Jetson? The Edge AI Platform Explained

NVIDIA Jetson is a family of small, power-efficient computers designed to run AI at the edge, close to where data is generated rather than in a data centre. Each Jetson is built around a system-on-module that combines an NVIDIA GPU, an Arm CPU, and memory on a single board, paired with the JetPack software stack. They power robots, smart cameras, autonomous vehicles, and industrial machines that need to see, decide, and act in real time.

NVIDIA Jetson is NVIDIA’s platform for edge AI: it brings the GPU-accelerated computing used to train and run AI models into a compact, low-power form that can be embedded in a product. Where a server runs AI in a controlled rack, a Jetson runs it on a moving robot, a factory line, or an outdoor camera.

The platform spans a wide range of performance, so the same software and tools scale from a pocket-sized module to one powerful enough for humanoid robotics. To choose between them, see our guide to the Jetson lineup.

If the cloud can run any AI model, why run it on a small computer in the field? Because for many real-world systems, sending data away and waiting for an answer does not work. Edge AI keeps the inference local, which matters when latency, bandwidth, privacy, or connectivity rule out the cloud.

A robot or vehicle has to react in milliseconds, not after a round trip to a server. Streaming many camera feeds off a remote site is expensive or impossible. Keeping video on the device avoids sending sensitive data off-site. And the system has to keep working when the network does not. Jetson exists to make that local inference fast enough, and efficient enough, to embed in the product.

At the heart of every Jetson is a system-on-module, or SOM: a single board carrying the GPU, CPU, memory, and power management. Putting all of that on one validated module is what makes Jetson practical for product designers — the hard, expensive silicon engineering is done, and the module drops onto a carrier board that provides the connectors, I/O, and form factor a specific product needs.

That split is also why a Jetson developer kit is not what you ship. The kit is a reference board for evaluation; a product pairs the bare module with a carrier board built for the job. We cover the module versus a developer kit in full. Our guide to choosing between an off-the-shelf, adapted or custom Jetson carrier board covers the other half of that pairing.

The current Jetson range runs from entry-level to data-centre-class, all sharing one software stack:

The earlier Jetson Xavier and TX2 generations are still in fielded products, but NVIDIA has brought them to end-of-life, so new designs start on Orin. At the entry end of Orin, the Super software update lifted the Orin Nano from 40 to 67 TOPS without any change to the silicon.

Jetson is not new. NVIDIA has shipped edge AI modules since the Jetson TK1 in 2014, through the TX1 and TX2, the Xavier generation, and the current Orin family, to the new Thor. Each generation has roughly tracked the GPU architecture of its era — Maxwell, Pascal, Volta, Ampere, and now Blackwell — bringing a large step in AI performance per watt.

For designers, the practical effect is a long, predictable platform with a clear upgrade path, which is part of why Jetson is trusted in industrial and defense programs. The trade-off is that older modules eventually reach end-of-life, as the Xavier and TX2 generations are now doing. At the newest end of that timeline, what Jetson Thor adds over AGX Orin for robotics and sensor fusion shows how large the latest generational step is.

Jetson is not the only way to run AI at the edge, but it occupies a useful middle ground.

Versus an industrial PC with a discrete GPU, a Jetson is far smaller and lower-power, and is built to be embedded, where a PC-plus-GPU is bulky and power-hungry for a moving or sealed product.

Versus a small AI accelerator, such as a Coral or Hailo module, Jetson carries a full GPU, CPU, and a mature software stack, so it runs a wider range of models and complete applications rather than a fixed-function accelerator.

Versus a general single-board computer, such as a Raspberry Pi, Jetson adds the GPU acceleration that makes real-time vision and deep learning practical, which a general SBC cannot match.

The right choice depends on the workload, but for GPU-accelerated AI in a compact, embeddable, long-life form, Jetson is the default.

Jetson goes wherever AI has to run in real time, on the device. In our own deployments that has meant autonomous agriculture, factory human-robot safety, smart surveillance for transit, and counter-drone tracking.

More broadly it powers autonomous mobile robots, machine vision and inspection, smart retail, medical devices, and rugged in-vehicle systems — anywhere latency, bandwidth, privacy, or connectivity rules out sending data to the cloud.

A Jetson module is the start, not the finish. To become a product it needs a Jetson carrier board for the right I/O, an enclosure for the environment, a configured software image, and validation.

That integration is what we do. We take the module suited to the workload and build it into a rugged Jetson computer, so what you deploy is a finished platform rather than a board on a bench. If you are choosing where to start, our guide to choosing a Jetson module walks the options.