At CES 2026, NVIDIA increases product and partnership announcements, emphasizing a continuous technology chain, from computing infrastructure to embedded AI in vehicles, robots and industrial machines, while relying on a network of leading industrial, cloud and software partners.
The cornerstone of the 2026 edition is the launch of the platform Rubin. Presented as an “AI supercomputer”, Rubin marks an evolution of NVIDIA’s positioning, which no longer highlights an isolated chip but a complete architecture, designed by code design around six components: Vera CPU, Rubin GPU, NVLink 6 interconnect, ConnectX-9 SuperNIC, BlueField-4 DPU and Spectrum-6 Ethernet. The goal is to reduce the overall cost of next-generation AI, both for training and inference, to facilitate large-scale adoption. NVIDIA claims significant gains compared to Blackwell, notably on the cost per token and the efficiency of Mixture of Experts models, in a context where agentic and long reasoning workloads are becoming dominant.
This announcement comes with broad support from the ecosystem, including major cloud providers (Amazon Web Services, Microsoft, Google Cloud And Oracle Cloud Infrastructure) which confirm the arrival of instances based on Rubin from 2026. On the user side, major players in AI such as OpenAI, Anthropic, Meta, Mistral AI Or xAI are part of this trajectory, validating the centrality of NVIDIA infrastructure for the next phase of AI scale-up.
Following on from Rubin, NVIDIA also highlighted BlueField-4 and the Inference Context Memory Storage Platforma new brick intended to respond to the explosion of context memory needs. As models reason over multiple rounds and AI agents become persistent, storing and sharing the KV cache becomes critical. NVIDIA offers here an AI-native approach, capable of increasing token throughput and energy efficiency, relying on an ecosystem of storage and infrastructure partners, including Dell Technologies, HPE, IBM, Pure Storage Or WEKA.
In terms of uses, NVIDIA has broadened its spectrum with Alpamayoa family of open models and tools intended for autonomous driving. The challenge is to deal with rare and complex scenarios that still slow down level 4 autonomy. By combining Vision-Language-Action models with explicit reasoning, simulation and open data sets, NVIDIA intends to accelerate the development of more robust autonomous stacks. Nvidia counts on its partners like Jaguar Land Rover, Lucid Motors And Uberas well as academic research with Berkeley DeepDriveto accelerate this adoption.
Another major focus of the announcements was robotics, an area where NVIDIA demonstrated new open models, with Cosmos, Isaac and GR00T, as well as open source frameworks intended to simplify the simulation, training and evaluation of robots capable of reasoning and learning multiple tasks. This approach aims to transform specialized machines into more general robots, while reducing development costs and complexity. Industrial players like Boston Dynamics, Caterpillar, NEURA Robotics Or LG Electronics will take advantage of CES to present robots integrating the NVIDIA stack.
Finally, NVIDIA illustrates the convergence of these bricks in mass-produced automobiles with Mercedes-Benz. The new CLA, the first model based on the MB.OS platform, includes the software DRIVE AV from NVIDIA for advanced Level 2 driver assistance features, with an architecture combining end-to-end AI and redundant safety stack via the Halos system. This partnership shows how software-defined AI can quickly translate into real-world features on the road.
Through these product and partnership announcements, NVIDIA is delivering at CES 2026 affirming its ability to unite partners and customers around the same technological stack, intended to serve as a reference for industrial AI.
