NVIDIA announced two new computing modules on July 15, 2026: the Jetson T3000 and the Jetson T2000. These are compact computers built to sit inside robots, factory machines, warehouse systems, and autonomous vehicles. They run AI models locally on the device, without needing a connection to a data center for every decision the machine makes.
The T3000 is the more powerful of the two. It uses 32GB of memory and delivers processing performance that NVIDIA says matches its more expensive flagship module for the AI tasks robots actually run. The T2000 is the entry-level option, with 16GB of memory, aimed at simpler machines like cameras with AI built in, mobile warehouse robots, and factory arms. Both modules are physically about half the size of NVIDIA's current high-end options.
The real story behind these chips is a memory price crisis that has nothing to do with robots directly. AI data centers around the world are consuming enormous amounts of high-bandwidth memory to run large AI models. That demand has pulled manufacturing capacity away from the standard memory chips used in everything else. LPDDR5X prices, the type of memory used in these robot modules, rose roughly 78 to 83% in a single quarter in 2026. For anyone building a product that contains memory, that is a serious cost problem.
NVIDIA's T3000 is designed specifically to address this. By using 32GB instead of 64GB or 128GB, robot builders spend less on memory per unit. NVIDIA says real-world AI performance for robot tasks is not meaningfully different because those workloads depend more on memory speed than on total memory size. In other words, you can get equivalent output from a smaller, cheaper chip if the chip is fast enough.
Alongside the hardware, NVIDIA released software tools that automatically find ways to reduce memory use in a robot's software stack. The process that used to take engineers weeks now takes days. Several robotics companies have already used these tools to cut memory use by up to 15GB per robot, allowing them to drop to lower-cost hardware without changing what the robot can do. One traffic management company reduced memory use by 30% on an existing device, making room to add new AI features without buying new hardware.
The humanoid robot market is growing quickly. Estimates from multiple research firms put global market size somewhere between $3 billion and $8 billion today, with projections suggesting it could reach $10 to $15 billion by 2030. Production surged tenfold in 2025, though most units are still in testing rather than full commercial deployment. Cost is the main barrier to wider adoption, and memory is one of the biggest cost items in every robot.
NVIDIA is also expanding its position by controlling more of the stack that robot builders depend on. Beyond the chips, the company now provides simulation software for testing robots before they exist physically, AI models that give robots the ability to see and reason about their environment, safety systems for robots that work near people, and deployment tools. A robotics company that builds on NVIDIA's platform uses NVIDIA hardware and NVIDIA software, trained on NVIDIA simulation environments.
This matters for any business evaluating robotics, whether as a buyer or a builder. A small team integrating an off-the-shelf robot system will likely be running on this infrastructure without knowing it. A larger operation considering custom automation will face a choice about how deep into this platform to go. The T3000 and T2000 ship in Q1 2027, but developers can begin working with them in simulation mode on existing hardware starting this month.
The memory price squeeze is not going away quickly. Supply growth for standard DRAM is projected at just 16% for 2026, well below historical norms, because manufacturers have shifted capacity to higher-margin AI server memory. Shortages are expected to persist into at least late 2027. For procurement teams building hardware bills of materials today, the amount of memory in a device has become a supply chain decision, not just a performance one.