NVIDIA’s Big Reveal at CES 2026: Rubin Platform Enters Mass Production, Accelerating AI-Powered Autonomous Driving

TradingKey - On January 5, 2026, Eastern Time, at the CES exhibition in Las Vegas, USA, NVIDIA CEO Jensen Huang once again took center stage in the global technology industry. Dressed in his signature black leather jacket, he delivered a 90-minute keynote speech, attempting to define a future where AI will not merely 'see,' but truly 'understand, reason, and act' .

In this highly anticipated speech, Huang, with his "All in AI" and "All in Physical AI" resolute stance, outlined a grand blueprint spanning from transistor-level architectural design to embedded control systems, and further to the integrated software and hardware deployment of autonomous driving platforms and humanoid robots. This was not only a moment to comprehensively showcase NVIDIA's full-stack AI capabilities but also a powerful response to capital market skepticism regarding the slowing of Moore's Law and the potential for an AI industry bubble.

Vera Rubin Enters Mass Production

"I can tell you that Vera Rubin has entered full production," Huang stated, a declaration that undoubtedly became the highlight of the day's presentation.

He further emphasized that this represents an unprecedented architectural innovation, with all six core modules—encompassing central processing units (CPUs), graphics processing units (GPUs), network interconnects, and entire system designs—being completely re-engineered. This not only overturns traditional approaches but also establishes a new direction for NVIDIA's next-generation AI infrastructure.

The Rubin platform is named in honor of the renowned astronomer Vera Rubin, and its core characteristic is the adoption of Extreme Co-Design, a philosophy where hardware and software are deeply integrated at every level, from chip to system, thereby forming a truly full-stack AI computing architecture.

Huang noted: "Artificial intelligence is reshaping industries at an unprecedented pace, and the fundamental challenge lies in acquiring sufficient computing power to support increasingly massive training and inference demands. Competition is intensifying, as the shorter the time to complete the same task, the greater the lead."

He added: "The introduction of Rubin is perfectly timed, as the demand for AI computing for both training and inference is experiencing explosive growth."

Compared to previous iterations which only updated a few chips at a time, this occasion saw a complete overhaul of six key hardware components in one go, considered NVIDIA's most groundbreaking product strategy step in recent years.

The core processor, Vera CPU, achieves a twofold performance increase under the same power consumption. The Rubin GPU, on the other hand, boosts computing power to five times its original level, with only a 60% increase in transistor count.

Regarding data transmission, the platform features the new NVLink 6 interconnect architecture, enabling data bandwidth of up to 260TB/s within a single rack. Concurrently, it is equipped with the next-generation ConnectX-9 SuperNIC network interface card, BlueField-4 Data Processing Unit (DPU), and Spectrum-6 Ethernet switch, collectively building a highly integrated, high-throughput, and technologically advanced AI computing infrastructure.

Rubin's introduction of a co-design strategy is a response to real-world manufacturing challenges such as the slowing of Moore's Law and limitations in transistor density growth. In other words, relying solely on increasing transistor count can no longer sustainably improve efficiency; instead, a simultaneous restructuring is required across the entire system architecture layer, network communication pathways, and even energy consumption management.

For thermal management, the system supports operation under 45°C warm water cooling conditions, maintaining stable temperature control without traditional cooling mechanisms.

Concurrently, even with doubled power consumption, the entire platform achieves a 100% improvement in energy efficiency, estimated to help reduce global data center electricity expenditure by approximately 6%.

From a practical application perspective, taking a 10-trillion-parameter (10T) language model as an example, using the Rubin system during the training phase can save approximately 75% in resource costs compared to a Blackwell deployment.

Tesla (TSLA) CEO Elon Musk stated that the Rubin platform "will be the rocket engine for AI" and "is the infrastructure for large-scale model deployment."

Anthropic CEO Dario Amodei added: "The efficiency gains of NVIDIA's Rubin platform represent an infrastructure advancement that enables longer memory, more powerful inference capabilities, and more reliable outputs."

Autonomous Driving Models Capable of 'Reasoning'

Unlike traditional autonomous driving systems, the Alpamayo model, an AI system based on a Vision-Language-Action (VLA) architecture, for the first time imbues vehicles with human-like "causal reasoning" capability, enabling them to handle extreme scenarios never encountered in training data.

For instance, when a vehicle approaches an intersection, the system can not only identify abnormal traffic lights but also comprehend the complex logic of "at this moment, one should follow the traffic officer's gestures or yielding rules," and make safety decisions accordingly.

The VLA model adopted in the Alpamayo series not only enhances the system's intelligent processing capabilities for sudden and long-tail scenarios (i.e., low-frequency but high-risk events) but also significantly improves the explainability of its decision-making logic.

A car equipped with this system can "think step-by-step" like a human and maintain clear judgment when facing the unknown, making its actions more trustworthy to users. Furthermore, the model integrates NVIDIA's self-developed Halos AI safety system, providing stability verification and dynamic risk control monitoring support for the entire inference pathway.

Huang stated: "The ChatGPT moment for physical AI has arrived—machines are beginning to understand, reason, and act in the real world. Robotaxis are among the first beneficiaries. Alpamayo brings reasoning capabilities to autonomous vehicles, enabling them to contemplate rare scenarios, drive safely in complex environments, and explain their driving decisions—this is the foundation for safe, scalable autonomous driving."

Notably, NVIDIA has decided to open-source the core Alpamayo model to the developer community free of charge.

This series is positioned as a large benchmark model with educational functions, allowing developers to fine-tune it based on its general capabilities and independently refine it into a set of core autonomous driving solution modules adapted to their product architecture needs. This 'common foundation + personalized evolution' model will significantly accelerate the pace at which companies enter L4 deployment, while simultaneously reducing the costs of algorithm accumulation and software-hardware adaptation.

Thomas Müller, Executive Director of Product Engineering at Jaguar Land Rover, stated: "Open and transparent AI development is crucial for responsibly advancing autonomous driving. By open-sourcing models like Alpamayo, NVIDIA is helping accelerate innovation across the entire autonomous driving ecosystem, providing new tools for developers and researchers to safely navigate complex real-world scenarios."

Autonomous Driving Commercialization Imminent

The Alpamayo model has gained high recognition within the industry, and NVIDIA's AI autonomous driving technology is progressing towards full commercialization. Huang stated that a mass-produced vehicle fully equipped with its full-stack DRIVE platform, the Mercedes-Benz CLA, is set to officially hit U.S. roads in the first quarter of 2026. This marks a milestone as NVIDIA's autonomous driving software and hardware solutions are fully integrated into a mainstream automotive product for the first time.

Furthermore, the company also plans to initiate L4-level Robotaxi (autonomous taxi) platform deployment by 2027, conducting field test operations with partners. This service will commence on specific urban routes, aiming to achieve an advanced "driverless" autonomous driving experience, and ultimately build a global next-generation mobility infrastructure.

L4 level represents a critical node in current high-level automation systems—within defined operational design domains, vehicles can autonomously complete the entire process from perception to decision-making and execution, ensuring safe passage even without human intervention.

This level of intelligence relies on pioneering technologies, including multi-modal foundational AI models, end-to-end computing architectures, and multi-scenario inference engines, enabling it to flexibly adapt to various complex road conditions and unforeseen events.

Xinzhou Wu, Vice President of Automotive at NVIDIA, explained that the Robotaxi project will target L4-level key performance indicators, launching in its early stages through "limited scope validation" and involving a large, undisclosed partner in joint operations.

He emphasized: "The core is not merely technological breakthroughs, but rather ensuring that products meet long-term reliability and commercial deployment standards within a controllable scope."

This move also signifies that NVIDIA, previously focused on chip manufacturing and computing power output, is now steadily transitioning into a new role of actively participating in the construction of intelligent mobility ecosystems, posing indirect competition to existing industry giants such like Waymo and Cruise.

Although NVIDIA's revenue contribution from the overall automotive sector remains modest—data shows that as of the last fiscal quarter, its revenue from automotive and robotics-related businesses totaled $592 million, accounting for merely about 1% of its total income—the company has nevertheless explicitly designated intelligent transportation as the next most promising growth driver after artificial intelligence, as AI further penetrates from data centers into edge devices.

Jensen Huang has stated on multiple occasions that he envisions a future where over a billion vehicles globally will achieve high-level or fully autonomous driving, encompassing both personally owned vehicles and shared mobility models such as on-demand rentals.