Nvidia CEO Jensen Huang dropped a bombshell at CES 2026: the company’s next-generation Vera Rubin AI platform is already in full production.
During his Monday keynote, Huang announced that Rubin, a six-chip, extreme-codesigned system built to slash the costs of training and running AI models, is ready for partners this year.
The move signals that Nvidia is accelerating its hardware roadmap ahead of schedule, cementing its dominance in the AI infrastructure race even as rivals AMD and Intel make aggressive pushes.
5 takeaways from Nvidia’s CEO speech at CES 2026
1. The first critical takeaway is that Rubin is not a standalone GPU upgrade.
Nvidia positioned Rubin as a complete rack-scale platform made of six tightly integrated components working as a single AI supercomputer.
The team includes a Vera CPU with 88 custom cores delivering double the performance of its predecessor.
Rubin GPUs capable of 50 petaflops of NVFP4 inference, a next-generation NVLink 6 switch providing 260TB/s of bandwidth per rack, the ConnectX-9 SuperNIC for high-throughput networking, the BlueField-4 DPU for offloading storage and security work, and Spectrum-X photonics-based Ethernet.
2. Huang declared that Rubin (also called NVL72) is in full production and being handed to partners now.
Partners, including Microsoft, CoreWeave, and Nebius, have already announced plans to integrate Rubin into their AI clouds in the second half of 2026.
This phased approach: partners first, then broader customer availability later in 2026, is crucial for investors tracking supply constraints and margin implications.
3. The third takeaway hinges on what Nvidia claims Rubin will deliver.
The company says Rubin GPUs offer up to 5x training performance improvements over Blackwell in some workloads and up to 3.5x training improvements overall.
More broadly, Nvidia says Rubin will reduce inference token costs by approximately 10x compared to Blackwell and allow enterprises to train mixture-of-experts models using 4x fewer GPUs.
These figures must be treated as company guidance, not independent benchmarks.
4. The next takeaway is Nvidia’s first major proof point in autonomous vehicles.
Huang announced that Mercedes-Benz’s new CLA will be the first production vehicle shipping with Nvidia’s complete DRIVE stack, including a new reasoning-based AI model called Alpamayo.
The system uses vision-language-action models to handle rare driving scenarios and explain its decisions, what Huang called the “ChatGPT moment for physical AI.”
The Mercedes rollout is critical because it demonstrates Nvidia moving beyond selling chips to selling full-stack solutions and reinforces the company’s pivot toward autonomous driving and robotics as growth engines beyond data centers.
5. Rubin is not Nvidia’s only big announcement.
Huang unveiled open-source AI models across six domains (healthcare, climate, robotics, reasoning, embodied AI, and autonomous driving) and introduced the DGX Spark desktop supercomputer for enterprise AI.
However, demos and commercial availability are different; Mercedes’ CLA will be Level 2+ on launch, and Nvidia’s Alpamayo robotaxi deployments are planned for 2027, not 2026.
For investors, the message is clear: Nvidia is betting hard on “physical AI” that not only sees but reasons and acts in the real world.
But expect staggered rollouts across different industries and geographies rather than immediate, universal deployment.
The post Nvidia CES 2026 announcements: 5 big takeaways from CEO Jensen Huang’s keynote appeared first on Invezz