NVIDIA advances AI frontiers with CES 2025 announcements

NVIDIA CEO and founder Jensen Huang took the stage at CES 2025 to present the company’s vision for the future of AI across gaming, autonomous vehicles (AVs), robotics, and beyond.

“AI has been evolving at an astonishing rate,” Huang remarked. “We started with perception AI—understanding images, words, and sounds. Then came generative AI, which creates text, images, and audio. Now, we are stepping into the era of ‘physical AI’—AI that can perceive, reason, plan, and take action.”

Huang highlighted how NVIDIA’s platforms and GPUs continue to drive innovation across multiple industries, unveiling key advancements such as the Cosmos platform, next-generation GeForce RTX 50 Series GPUs, and the compact AI supercomputer, Project DIGITS.

RTX 50 Series: “A Beast of a GPU”

A major highlight of CES 2025 was the launch of NVIDIA’s GeForce RTX 50 Series, powered by the Blackwell architecture. The flagship RTX 5090 GPU, with 92 billion transistors and a staggering 3,352 trillion AI operations per second (TOPS), made its debut.

“GeForce brought AI to the masses, and now AI is becoming an integral part of GeForce,” Huang said.

Holding the sleek, blacked-out GPU, Huang described it as “a beast,” emphasizing its advanced capabilities, such as dual cooling fans and AI-driven real-time graphics enhancements.

The RTX 50 Series will have a staggered release in early 2025, starting with the RTX 5090 and RTX 5080 on January 30, followed by the RTX 5070 Ti and RTX 5070 in February. Laptop versions will launch in March.

Additionally, NVIDIA introduced DLSS 4 with ‘Multi-Frame Generation’ technology, boosting gaming performance up to eight times by generating three extra frames for every frame rendered.

Cosmos: Advancing Physical AI

At CES 2025, NVIDIA unveiled the Cosmos platform, a significant leap forward for robotics, industrial AI, and AVs. Huang compared Cosmos’ impact to that of large language models in generative AI, signaling a transformative shift in how AI interacts with the physical world.

“The ChatGPT moment for general robotics is approaching,” Huang asserted.

Cosmos integrates generative models, tokenizers, and video-processing frameworks, enabling robots and AVs to simulate potential outcomes and predict optimal actions. By processing text, images, and video inputs, Cosmos generates “virtual world states,” ideal for real-world robotics and AV applications.

Leading robotics and automotive companies, including XPENG, Hyundai Motor Group, and Uber, are among the first adopters. Cosmos is available on GitHub under an open license.

Pras Velagapudi, CTO at Agility, highlighted its potential: “Data scarcity and variability are major challenges in robotic learning. Cosmos’ ability to generate photorealistic scenarios from text, images, and video allows us to train models without the need for extensive real-world data collection.”

AI Models for Developers

NVIDIA also introduced AI foundation models for RTX PCs, designed to enhance content creation, productivity, and enterprise applications. These models, branded as NVIDIA NIM (Neural Interaction Model) microservices, are optimized for integration with RTX 50 Series hardware.

“These AI models run on every major cloud platform because NVIDIA GPUs are now available everywhere,” Huang noted.

To further empower developers, NVIDIA launched AI Blueprints—pre-configured frameworks for building AI agents tailored to tasks like content generation, fraud detection, and video analysis.

“They’re fully open-source, so developers can customize them to their needs,” Huang explained.

Additionally, NVIDIA introduced Llama Nemotron, a framework for building and deploying AI agents. Ahmad Al-Dahle, VP and Head of GenAI at Meta, emphasized its significance: “Agentic AI is the next evolution of AI development, and optimizing across a full-stack system is essential. NVIDIA’s Llama Nemotron family, built on Llama, will help enterprises develop customized AI agents.”

Philipp Herzig, Chief AI Officer at SAP, added: “AI agents collaborating to solve complex business problems will drive a new wave of productivity. With SAP’s Joule, millions of enterprise users will interact with AI agents to achieve their goals faster than ever. NVIDIA’s open Llama Nemotron model will accelerate the development of specialized AI agents for business applications.”

Safer, Smarter Autonomous Vehicles

NVIDIA also revealed advancements in AV technology with its DRIVE Hyperion platform, built on the new NVIDIA AGX Thor system-on-a-chip (SoC). This platform enhances functional safety and autonomous capabilities using generative AI models.

“The autonomous vehicle revolution is here,” Huang stated. “Like all robots, AVs require three key computing systems: NVIDIA DGX to train AI models, Omniverse to generate synthetic data, and DRIVE AGX, a supercomputer inside the car.”

Huang highlighted the role of synthetic data in AV development, as it significantly improves training datasets. NVIDIA’s AI data factories—powered by Omniverse and Cosmos—create synthetic driving scenarios, dramatically enhancing real-world training effectiveness.

Project DIGITS: A Compact AI Supercomputer

To close his keynote, Huang unveiled “one more thing”: Project DIGITS, NVIDIA’s smallest yet most powerful AI supercomputer, equipped with the GB10 Grace Blackwell Superchip.

“This is NVIDIA’s latest AI supercomputer,” Huang announced, revealing its compact size, small enough to be “almost pocketable.”

Project DIGITS enables developers and engineers to train and deploy AI models from their desks, providing full access to NVIDIA’s AI ecosystem in a portable format.

A Vision for the Future

Looking back on NVIDIA’s journey since pioneering the programmable GPU in 1999, Huang reflected on the past decade’s AI-driven transformations.

“Every layer of the technology stack has been redefined,” he said.

With breakthroughs in gaming, AI-driven agents, robotics, and autonomous vehicles, Huang envisions an exciting road ahead.

“All the technologies we’ve discussed today will lead to remarkable advancements in general robotics and AI in the coming years,” he concluded.