Nvidia CEO introduces powerful next-gen chips: Jensen Huang Unveils “Vera Rubin” Architecture and Reasoning AI at CES 2026
The “race for AI” just shifted into a higher gear.
In a high-octane keynote kicking off CES 2026 in Las Vegas, Nvidia CEO Jensen Huang proved the company isn’t just leading the AI revolution—it’s accelerating it. Surpassing even the loftiest expectations, Huang officially pulled back the curtain on Vera Rubin, the successor to the record-breaking Blackwell architecture, and announced a breakthrough in “Physical AI” that will put reasoning cars on U.S. roads this quarter.
The Rubin Revolution: Six Chips, One AI Supercomputer
While the tech world was still digesting the rollout of Blackwell, Nvidia has already moved the Vera Rubin platform into full production. Huang described the platform as an achievement in “extreme co-design,” where six specialized chips are architected to function as a single, massive unit of compute.
Breaking Down the Rubin Stack
Named after the pioneering astronomer Vera Rubin, this platform is built to handle the next generation of 10-trillion-parameter models.
- Vera CPU: Featuring 88 custom-designed “Olympus” cores, this CPU is built specifically to orchestrate data movement in AI factories, delivering twice the performance of its predecessor.
- Rubin GPU: The heart of the system, delivering a staggering 50 petaflops of inference performance. It features HBM4 memory with 22 TB/s of bandwidth—nearly tripling the speed of Blackwell.
- Next-Gen Networking: The stack includes the NVLink 6 Switch, providing 260 TB/s of bandwidth per rack, and the Spectrum-X Ethernet Photonics system, which uses light to move data with 5x better power efficiency.
Efficiency That Defies the Odds
Perhaps the most shocking metric shared by Huang was the platform’s efficiency. Despite doubled power consumption, Rubin achieves a 100% improvement in energy efficiency and slashes the cost of generating AI tokens to one-tenth of prior systems.
The “ChatGPT Moment” for Physical AI: Alpamayo
“The race is on for physical AI,” Huang declared as he introduced Alpamayo, an open-source family of AI reasoning models.
Unlike traditional self-driving systems that rely on pattern matching, Alpamayo uses Vision-Language-Action (VLA) architecture. It doesn’t just detect a ball in the street; it “reasons” that a child might follow it, using step-by-step thinking (chain-of-thought) to make safe decisions.
Mercedes-Benz Leads the Charge
Nvidia confirmed that its long-standing partnership with Mercedes-Benz is finally hitting the pavement. The all-new Mercedes-Benz CLA will be the first production vehicle fully equipped with the Nvidia DRIVE platform and Alpamayo reasoning capabilities, launching in the U.S. in Q1 2026.
Tech Insights & FAQs
Q: Why is the Rubin platform such a big deal? A: It shifts AI from “perceiving” to “reasoning.” For a 10-trillion-parameter model, Rubin cuts training costs by 75% compared to Blackwell.
Q: Is Alpamayo really open-source? A: Yes. Nvidia has released the Alpamayo 1 model weights on Hugging Face and the AlpaSim simulation framework on GitHub to accelerate industry-wide Level 4 autonomy.
Q: When will Rubin chips be available? A: They are in full production now, with shipping to key customers like Microsoft, Amazon, and Google expected in the second half of 2026.
🚀 Pro-Tips for Tech Enthusiasts
- Watch the “MoE” Shift: Rubin is optimized for Mixture-of-Experts (MoE) models, requiring 4x fewer GPUs to train them than previous generations.
- Liquid Cooling is the Future: The Rubin platform supports warm-water cooling up to 45°C, signaling a major shift in data center design.
- Autonomous Edge: Beyond cars, look for the Isaac GR00T N1.6 model, which uses this reasoning tech to allow humanoid robots to manipulate objects and move simultaneously.
Related News
- Microsoft’s Fairwater Superfactory: Microsoft will be among the first to deploy Vera Rubin NVL72 racks in its next-gen AI data centers.
- The Global Rollout: While the Nvidia-powered Mercedes-Benz launches in the U.S. this quarter, it is slated for Europe in Q2 and Asia later in 2026.
- Nvidia-Intel Synergy: Following a major stake in Intel last year, the companies are reportedly collaborating on specialized manufacturing for the Rubin ecosystem.
Nvidia’s CES 2026 keynote has made one thing clear: the age of machines that “understand and reason” isn’t coming—it’s already here.
The “race for AI” just shifted into a higher gear.
In a high-octane keynote that kicked off CES 2026 in Las Vegas, Nvidia CEO Jensen Huang took the stage to prove that the company isn’t just leading the AI revolution—it’s accelerating it. Surpassing even the loftiest expectations of Wall Street and Silicon Valley, Huang officially pulled back the curtain on Vera Rubin, the successor to the record-breaking Blackwell architecture.
Named after the pioneering astronomer who discovered evidence of dark matter, the Rubin platform represents a paradigm shift from “perceiving” AI to “reasoning” AI. But the hardware was only half the story. Nvidia also dropped a bombshell in the automotive world: a new family of AI models called Alpamayo that will power the first truly “reasoning” autonomous vehicles, starting with Mercedes-Benz this quarter.
The Rubin Revolution: Six Chips, One Supercomputer
While the tech world was still digesting the rollout of Blackwell, Nvidia has already moved into full production of Rubin. Huang described the platform as an “extreme co-design” achievement—a tightly integrated system where six specialized chips function as a single, massive AI unit.
Breaking Down the Rubin Stack
The Rubin platform isn’t just a single processor; it’s an entire ecosystem built to handle the next generation of 10-trillion-parameter models.
- Vera CPU: Featuring 88 custom-designed “Olympus” Arm cores, this CPU is built specifically to orchestrate data movement and eliminate the bottlenecks that plague modern AI factories.
- Rubin GPU: The heart of the system, delivering a staggering 50 petaflops of inference performance. It features HBM4 memory with 22 TB/s of bandwidth—doubling the speed of its predecessor.
- NVLink 6 Networking: A scale-up interconnect providing 260 TB/s of bandwidth per rack, allowing thousands of GPUs to talk to each other as if they were one.
- Spectrum-X Ethernet Photonics: A breakthrough in networking that uses light to move data, offering 5x better power efficiency than traditional copper-based systems.
Efficiency That Defies the Odds
Perhaps the most shocking metric shared by Huang was the platform’s efficiency. Despite doubling the power consumption, Rubin achieves a 100% improvement in energy efficiency and slashes the cost of generating AI tokens to one-tenth of previous systems. For enterprise customers, this means training massive “Mixture-of-Experts” (MoE) models with four times fewer GPUs than Blackwell required.
The “ChatGPT Moment” for Physical AI: Alpamayo
“The race is on for physical AI,” Huang declared as he pivoted to Nvidia’s latest software breakthrough: Alpamayo.
Unlike traditional self-driving systems that rely on pattern matching and object detection, Alpamayo is a Vision-Language-Action (VLA) model. It doesn’t just “see” a ball rolling into the street; it “reasons” that a child might be following it and explains its decision-making process in real-time.
Mercedes-Benz Leads the Charge
Nvidia confirmed that its long-standing partnership with Mercedes-Benz is finally hitting the pavement. The all-new Mercedes-Benz CLA, launching in the U.S. in Q1 2026, will be the first production vehicle powered by the full Nvidia DRIVE stack and Alpamayo models.
“Alpamayo allows cars to think through rare scenarios, drive safely in complex environments, and explain their driving decisions,” Huang explained. “This is the foundation for safe, scalable autonomous driving.”
🚀 Pro-Tips for Investors & Tech Enthusiasts
- Watch the “MoE” Shift: As models move toward “Mixture-of-Experts” architectures, hardware that can handle fragmented, high-speed data movement (like Rubin’s NVLink 6) will be the gold standard.
- Look Beyond the GPU: Nvidia is no longer just a “chip company.” Their move into networking (Spectrum-X) and software (Alpamayo) means they are capturing the entire value chain of the data center.
- The Power Factor: Keep an eye on liquid cooling. Rubin is designed for warm-water cooling (up to 45°C), signaling that the future of the data center is liquid, not air.
Related Breaking News
- Microsoft’s Fairwater AI Superfactory: Microsoft has already signed on to be one of the first to deploy “Vera Rubin” NVL72 racks in its next-gen AI factories.
- Elon Musk Weighs In: The Tesla CEO expressed skepticism on X (formerly Twitter), noting that solving the “long tail” of autonomous driving is “super hard,” though he wished Nvidia success.
- Trump Administration Policy: Recent reports suggest the U.S. government has approved limited shipments of older H200 chips to China, but the high-end Blackwell and Rubin architectures remain strictly reserved for the U.S. market.
Nvidia’s CES 2026 keynote has made one thing clear: the age of “thinking” machines isn’t coming—it’s already here.
