CIOInsider India Magazine

One of the highly anticipated presentations of CES 2026 took place in January at the Mandalay Bay Convention Center in Las Vegas. Featuring NVIDIA CEO Jensen Huang, the event saw his keynote which focused on various aspects of modern AI development. This included discussing the philosophical foundations of a revolutionized computing industry and the much awaited introduction of NVIDIA's latest Vera Rubin supercomputing platform.

Rubin, Nvidia's first platform designed for extreme coding, will consist of six AI chips, alongside various networking technologies and system software, operating in harmony as a unified computing unit. Named in honor of the astronomer who provided proof for the existence of dark matter through observations of galactic rotation curves, this system embodies Nvidia's solution to the profound challenge highlighted by Jensen. During his extensive keynote address lasting around two hours, Jensen emphasized that AI is expanding into all fields and devices. Through Rubin, Nvidia seeks to propel AI into new frontiers while significantly reducing the cost of generating tokens to only one-tenth of the previous platform, thereby making large-scale AI deployment far more cost-effective.

Also Read: IIT Bombay Inks Agreement with the US-based Columbia University

The Engineering Power of Vera Rubin
The computational requirements for AI are expanding rapidly beyond what traditional scaling can keep up with. Every year, the size of models is growing significantly. During testing, scaling now involves more complex reasoning processes rather than just quick responses, with the generation of tokens increasing by about five times each year. At the same time, competition is pushing token costs down by roughly ten times annually as businesses rush to reach new milestones.

Vera Rubin's details exemplify this fact. The device provides 100 petaflops of AI power, which is five times greater than before. The technology consists of six innovative chips that necessitated a total overhaul of all parts and a complete rewrite of the software. Each Vera Rubin MDL 72 rack contains 220 trillion transistors and weighs close to two tons. With 1,152 GPUs spread over 16 racks, the total system is viewed by NVIDIA as a significant advancement in the field of AI.

The system is filled with impressive engineering accomplishments. Each GPU is provided with a scale-out bandwidth of 1.6 terabits per second by ConnectX9. The storage and security operations are taken care of by BlueField 4 DPUs, freeing up compute resources for AI tasks. The NVLink switch facilitates the movement of large amounts of data, connecting compute nodes and allowing up to 72 Rubin GPUs to function as one cohesive unit.

Jensen stressed the importance of breaking the internal rule at NVIDIA that limits the number of chip changes per generation for the Vera Rubin project. It is essential to accelerate hardware development beyond the traditional semiconductor industry pace to keep up with the rapid advances in AI technology. Jensen made it clear that the Vera Rubin system is currently in production. The GB200 systems started shipping 18 months ago, while the GB300 systems are now being manufactured on a large scale.

Vera Rubin: Beyond Blackwell
Rubin is being hailed as the next-generation successor to NVIDIA Blackwell, promising to decrease AI inference expenses by a significant factor. Through a collaborative effort between its GPUs and CPUs, Rubin has improved upon its previous chip designs to better accommodate the demands of AI deployment on a larger scale. GPUs are responsible for carrying out extensive mathematical calculations in order to facilitate the training of exceptionally large AI models.

Also Read: Sarvam AI Takes on Giants with Indic LLM

On the other hand, Vera CPUs execute intricate instructions by utilizing their Olympus cores, and they are able to exchange data between themselves at a remarkable speed, all without encountering any bottlenecks. The greater the number of transistors, the quicker Rubin can handle larger datasets while consuming less power, and can easily scale up - envision a scenario where Lego blocks are stacked together to build a planet-sized artificial intelligence system without causing servers to overheat due to excessive heat generation.

Physical AI: From Digital Intelligence to Physical Action
Possibly the most innovative part of Jensen's talk focused on the concept of physical AI, which involves developing artificial intelligence that can comprehend and engage with the real world. Jensen detailed a three-computer system for advancing physical AI. The initial computer is responsible for AI training, while the second manages inference in robotics applications like vehicles, robots, or edge settings. The third computer, which Huang proposed as groundbreaking, is designated for simulation. Jensen emphasized the importance of simulating how the physical world reacts to AI actions in order to assess the system's performance accurately.

Spectrum-6 Ethernet: Engineering the AI Network Core
Cutting-edge Ethernet networking and storage play a crucial role in the AI infrastructure, ensuring that data centers operate optimally, enhance productivity, and reduce expenses. The NVIDIA Spectrum-6 Ethernet is a cutting-edge technology designed for AI networking, specifically tailored to support Rubin-based AI factories more effectively and with increased durability. It features advanced 200G SerDes communication circuitry, co-packaged optics, and AI-optimized fabrics to enhance performance and reliability.

The Spectrum-X Ethernet Photonics optical switch systems, based on the Spectrum-6 architecture, offer enhanced reliability and longer uptime for AI applications compared to traditional methods. With improved power efficiency, these systems maximize performance per watt. The Spectrum-XGS Ethernet technology within the Spectrum-X Ethernet platform allows distant facilities to operate as a unified AI environment, even when separated by hundreds of kilometers. These advancements collectively represent the future iteration of the NVIDIA Spectrum-X Ethernet platform, designed with thorough codesign for Rubin to facilitate the operation of vast AI facilities and lay the groundwork for upcoming million-GPU systems.

Setting the Pace for the AI Ecosystem
The presentation for quantum computing firms highlights the competitive environment and possible ways for integration. With NVIDIA leading in AI infrastructure, quantum systems will need to show clear benefits in specific areas rather than being generally capable. The focus on combining different architectures and systems hints at a future where quantum processors could be used as specialized accelerators in AI workflows controlled by classical systems.

Also Read: Flipkart to Enter into India's Food Delivery Market

NVIDIA's plan outlines a determined effort to constantly enhance AI infrastructure, as Vera Rubin is expected to surpass Blackwell in both factory throughput and cost efficiency for token generation. By choosing to release significant platform upgrades annually instead of following conventional semiconductor development timelines, the company is setting a pace that will impact the entire AI ecosystem.

NVIDIA's approach to an open model strategy goes beyond being just a hardware vendor, positioning them as a key player in democratizing AI. They are expanding their focus beyond language models to delve into physical AI as their next big area of development. The collaboration with Siemens implies that physical AI technology is transitioning from being confined to research labs to being utilized in real-world industrial settings.

Jensen Huang's presentation at CES 2026 described an industry undergoing rapid transformation, rendering traditional development timelines irrelevant. The underlying message conveyed throughout the presentation was that staying ahead in AI requires more than just access to efficient models - it also requires access to the necessary computational power to innovate faster than others. NVIDIA's perspective views AI not just as a software layer but as a revolutionary shift in the way computing operates, how applications are created, and how digital systems interact with the world around us.