Although the holiday gift-giving season may be over, Nvidia Corp. co-founder and Chief Executive Jensen Huang was in a very generous mood during his Monday keynote address at the CES consumer electronics show in Las Vegas. The leader in accelerated computing, which invented the graphics processing unit more than 25 years ago, still has an insatiable appetite for innovation.

Huang (pictured), dressed in a more Vegas version of his customary black leather jacket, kicked off this keynote with a history lesson on how Nvidia went from a company that made video games better to the AI powerhouse it is today. He then shifted into product mode and showcased his company’s continuing leadership in the AI revolution by announcing several new and enhanced products for AI-based robotics, autonomous vehicles, agentic AI and more. Here are the five I felt were most meaningful:

Cosmos for world-building

Nvidia’s Cosmos platform consists of what the company calls “state-of-the-art generative world foundation models, advanced tokenizers, guardrails and an accelerated video processing pipeline” for advancing the development of physical AI capabilities, including autonomous vehicles and robots.

Using Nvidia’s world foundation models or WFMs, Cosmos makes it easy for organizations to produce vast amounts of “photoreal, physics-based synthetic data” for training and evaluating their existing models. Developers can also fine-tune Cosmos WFMs to build custom models.

Physical AI can be very expensive to implement, requiring robots, cars and other systems to be built and trained in real-life scenarios. Cars crash and robots fall, adding cost and time to the process. With Cosmos, everything can simulated virtually, and when the training is complete, the information is uploaded into the physical device.

Nvidia is providing Cosmos models under an open model license to help the robotics and AV community work faster and more effectively. Many of the world’s leading physical AI companies use Cosmos to accelerate their work.

The Omniverse is expanding

Huang also announced new generative AI models and blueprints that expand and further integrate Nvidia Omniverse into physical AI applications. The company said leading software development and professional services firms are leveraging Omniverse to drive the growth of new products and services designed to “accelerate the next era of industrial AI.”

Companies such as Accenture, Microsoft and Siemens are integrating Omniverse into their next-generation software products and professional services. Siemens announced at CES the availability of Teamcenter Digital Reality Viewer, its first Xcelerator application powered by Nvidia’s Omniverse libraries.

New blueprints for developers

Nvidia debuted four new blueprints for developers to use in building Universal Scene Description (OpenUSD)-based Omniverse digital twins for physical AI. The new blueprints are:

Raising the bar for consumer GPUs

Nvidia announced the GeForce RTX 50 series of desktop and laptop graphics processing units. The RTX 50 series is powered by Nvidia’s Blackwell architecture and the latest Tensor Cores and RT Cores. Huang said it delivers breakthroughs in AI-driven rendering. “Blackwell, the engine of AI, has arrived for PC gamers, developers and creatives,” he said. “Fusing AI-driven neural rendering and ray tracing, Blackwell is the most significant computer graphics innovation since we introduced programmable shading 25 years ago.”

The pricing of the new systems gave rise to a loud cheer from the crowd. The previous generation GPU, RTX 4090, retailed for $1,599. The low end of the 50 series, the RTX 5070, which offers comparable performance (1,000 trillion AI operations per second) to the RTX 4090, is available for the low price of $549. The RTX 5070 Ti, 1,400 AI TOPS is $749, the RTX 5080 (1,800 AI TOPS) sells for $999, and the RTX 5090, which offers a whopping 3,400 AI TOPS, is $1,999.

The company also announced a family of laptops where the massive RTX processor has been shrunk down and put into a small form factor. Huang explained that Nvidia used AI to accomplish this, as it generates most of the pixels using Tensor Cores. This means only the required pixels are raytraced, and AI is used to develop all the other pixels, creating a significantly more energy-efficient system. “The future of computer graphics is neural rendering, which fuses AI with traditional graphics,” Huang explained. Laptop pricing ranges from $1,299 for the RTX 5070 model to $2,899 for the RTX 5090.

Project DIGITS

Huang introduced a small desktop computer system called Project DIGITS powered by Nvidia’s new GB10 Grace Blackwell Superchip. The system is small but powerful. It will provide a petaflop of AI performance with 120 gigabytes of coherent, unified memory. The company said it will enable developers to work with AI models of up to 200 billion parameters at their desks. The system is designed for AI developers, researchers, data scientists and students working with AI workloads. Nvidia envisions key workloads for the new computer, including AI model experimentation and prototyping.

Enabling agentic AI

Rev Labaredian, vice president of Omniverse and simulation technology at Nvidia, told analysts in a briefing before Huang’s keynote that the massive shift in computing now occurring represents software 2.0, which is machine learning AI that is “basically software writing software.” To meet this need, Nvidia is introducing new products to enable agentic AI, including the Llama Nemotron family of open large language models. The models can help developers create and deploy AI agents across various applications — including customer support, fraud detection, and product supply chain and inventory management optimization.

Huang explained that the Llama models could be “better fine-tuned for enterprise use,” so Nvidia used its expertise to create the Llama Nemotron suite of open models. There are currently three models: Nano is small and low latency with fast response times for PCs and edge devices, Super is balanced for accuracy and computer efficiency, and Ultra is the highest-accuracy model for data center-scale applications.

Final thoughts

If it’s not clear by now, the AI era has arrived. Many industry watchers believe AI is currently overhyped, but I think the opposite. AI will eventually be embedded into every application, device and system we use. The internet has changed how we work, live and learn, and AI will have the same impact. Huang did an excellent job of explaining the relevance of AI to all of us today and what an AI-infused world will look like. It was a great way to kick off CES 2025.

Photo: Nvidia/livestream