NVIDIA collected four awards at COMPUTEX 2026 in Taipei, Asia's largest tech trade show. Three products won: the Vera Rubin NVL72 rack, the Jetson Thor robotics module, and the Alpamayo self-driving platform. Each one targets a different layer of the AI supply chain, and together they tell you where NVIDIA thinks the next few years of AI spending will land.
The Vera Rubin NVL72 is the flagship. It is a single server rack that holds 72 AI processors, and NVIDIA claims it can run AI tasks at one-tenth the cost per unit of output compared to the previous generation, the Blackwell platform. That number matters because the cost of running AI is currently one of the main things stopping businesses from deploying it at scale. If the claim holds in practice, it changes the economics for anyone paying cloud providers to run AI workloads.
The rack itself is notable for its physical design. It has no fans, no cables connecting the trays, and runs entirely on liquid cooling. NVIDIA says a single compute tray now takes five minutes to install, down from two hours on the previous generation. That sounds like a detail for data center engineers, but it translates directly into faster deployments and lower maintenance costs for the companies that buy cloud AI services built on this hardware.
Pricing is significant context here. Industry estimates put Vera Rubin NVL72 racks at $5 million to $7 million per unit, compared to roughly $3 million for the previous Blackwell-based equivalent. The hardware is more expensive to buy, but NVIDIA's argument is that it delivers far more output per dollar spent over time. Volume availability is expected in the second half of 2026, through AWS, Google Cloud, Microsoft Azure, and others.
Jetson Thor is a different kind of product entirely. It is a compact computing module designed to sit inside physical machines, running AI without needing a constant connection to the internet. It is already shipping and in general availability, with companies like Amazon Robotics, Boston Dynamics, Caterpillar, Figure AI, and Medtronic among the first adopters. The practical applications span manufacturing floors, logistics warehouses, hospital operating rooms, and agricultural equipment.
What Jetson Thor enables is machines that can see, reason, and act in real time. Earlier generations of industrial robots followed pre-programmed rules. This chip lets them respond to things they have not seen before, which is the gap between a robot that works in a controlled factory and one that can handle the messiness of the real world.
Alpamayo sits in a different market: autonomous vehicles. The core idea is that self-driving systems have historically struggled with rare or ambiguous situations, scenarios that do not appear often enough in training data for a model to learn from. Alpamayo is designed specifically for these edge cases, handling situations like contradicting traffic signals or an emergency vehicle parked partly in a lane. The models are open-source and available on Hugging Face, meaning any AV developer can use them. JLR, Lucid Motors, and Uber have already expressed interest.
The broader picture from these three awards is straightforward. NVIDIA is not just a chip company anymore; it is building the full stack from data center hardware down to the module that sits inside a robot arm. Each layer it controls is a layer where competitors have to either partner with NVIDIA or build their own equivalent, which is expensive and slow. For businesses thinking about where to deploy AI in operations, the practical takeaway is that the hardware underneath the AI tools you use is getting significantly more capable in the next 12 months, and the cost of running those tools is set to fall.