At CES 2025, NVIDIA CEO Jensen Huang showcased groundbreaking advancements in AI technology, particularly agentic AI, which enhances real-world applications like coding and customer service, alongside the introduction of the RTX Blackwell GPUs and the NVIDIA Cosmos platform for generating synthetic training data. He also unveiled Project Digits, a compact AI supercomputer aimed at democratizing access to powerful AI capabilities for researchers and startups.
At CES 2025, Jensen Huang, the CEO of NVIDIA, delivered an impressive keynote that showcased the company’s advancements in AI technology, particularly focusing on agentic AI and its applications across various industries. He emphasized that 2025 is set to be the year of agentic AI, which combines the capabilities of large language models with real-world functionalities, enabling tasks such as coding assistance, customer service, and patient care. Huang highlighted the transition from traditional computing to AI-driven computing, predicting that AI will increasingly predict and generate interactions in real-time, fundamentally changing how software applications operate.
One of the standout demonstrations during the keynote was NVIDIA’s new rendering technology, which utilizes AI to generate 90% of video game frames in real-time. This innovative approach allows for more efficient graphics rendering, where traditional ray tracing is only used for a small fraction of the frames. Dr. Jim Fan, a research scientist at NVIDIA, explained that this method leverages neural networks to fill in details based on a rough sketch, marking a significant shift in how video games and graphics are produced.
Huang also introduced the RTX Blackwell series of GPUs, which promise to deliver top-tier performance at a significantly lower price point compared to previous models. The keynote discussed the scaling laws of AI, emphasizing the importance of data and computational power in training models. Huang noted that as the internet continues to generate vast amounts of data, the AI industry is racing to harness this information to improve model capabilities, with a focus on pre-training, post-training, and test-time scaling.
A major announcement was the introduction of NVIDIA Cosmos, a world foundation model designed to understand and simulate the physical world. This platform aims to generate synthetic data for training AI models, addressing the challenges of capturing real-world data. Huang explained that Cosmos could create virtual environments for reinforcement learning, enabling AI to learn and adapt without the need for extensive real-world data collection, which is often costly and time-consuming.
Finally, Huang unveiled Project Digits, a compact AI supercomputer designed for consumer use, which integrates NVIDIA’s entire AI software stack. This device aims to democratize access to powerful AI computing, allowing researchers and startups to leverage advanced AI capabilities from their homes or offices. With impressive specifications and a focus on privacy and security, Project Digits represents a significant step towards making AI technology more accessible to a broader audience, further solidifying NVIDIA’s position as a leader in the AI industry.