NVIDIA founder and CEO Jensen Huang kicked off CES 2025 with a 90-minute keynote that included new merchandise to advance gaming, autonomous automobiles, robotics, and agentic AI.
AI has been “advancing at an unbelievable tempo,” he mentioned earlier than an viewers of greater than 6,000 packed into the Michelob Extremely Enviornment in Las Vegas.
“It began with notion AI — understanding photographs, phrases, and sounds. Then generative AI — creating textual content, photographs and sound,” Huang mentioned. Now, we’re coming into the period of “bodily AI, AI that may proceed, motive, plan and act.”
NVIDIA GPUs and platforms are on the coronary heart of this transformation, Huang defined, enabling breakthroughs throughout industries, together with gaming, robotics and autonomous automobiles (AVs).
Huang’s keynote showcased how NVIDIA’s newest improvements are enabling this new period of AI, with a number of groundbreaking bulletins, together with:
Huang began off his discuss by reflecting on NVIDIA’s three-decade journey. In 1999, NVIDIA invented the programmable GPU. Since then, trendy AI has essentially modified how computing works, he mentioned. “Each single layer of the know-how stack has been reworked, an unbelievable transformation, in simply 12 years.”
Revolutionizing Graphics With GeForce RTX 50 Sequence
“GeForce enabled AI to succeed in the lots, and now AI is coming residence to GeForce,” Huang mentioned.
With that, he launched the NVIDIA GeForce RTX 5090 GPU, probably the most highly effective GeForce RTX GPU thus far, with 92 billion transistors and delivering 3,352 trillion AI operations per second (TOPS).
“Right here it’s — our brand-new GeForce RTX 50 sequence, Blackwell structure,” Huang mentioned, holding the blacked-out GPU aloft and noting the way it’s in a position to harness superior AI to allow breakthrough graphics. “The GPU is only a beast.”
“Even the mechanical design is a miracle,” Huang mentioned, noting that the graphics card has two cooling followers.
Extra variations within the GPU sequence are coming. The GeForce RTX 5090 and GeForce RTX 5080 desktop GPUs are scheduled to be obtainable Jan. 30. The GeForce RTX 5070 Ti and the GeForce RTX 5070 desktops are slated to be obtainable beginning in February. Laptop computer GPUs are anticipated in March.
DLSS 4 introduces Multi Body Era, working in unison with the entire suite of DLSS applied sciences to spice up efficiency by as much as 8x. NVIDIA additionally unveiled NVIDIA Reflex 2, which may cut back PC latency by as much as 75%.
The newest era of DLSS can generate three further frames for each body we calculate, Huang defined. “In consequence, we’re in a position to render at extremely excessive efficiency, as a result of AI does quite a bit much less computation.”
RTX Neural Shaders use small neural networks to enhance textures, supplies and lighting in real-time gameplay. RTX Neural Faces and RTX Hair advance real-time face and hair rendering, utilizing generative AI to animate probably the most reasonable digital characters ever. RTX Mega Geometry will increase the variety of ray-traced triangles by as much as 100x, offering extra element.
Advancing Bodily AI With Cosmos|
Along with developments in graphics, Huang launched the NVIDIA Cosmos world basis mannequin platform, describing it as a game-changer for robotics and industrial AI.
The following frontier of AI is bodily AI, Huang defined. He likened this second to the transformative influence of enormous language fashions on generative AI.
“The ChatGPT second for normal robotics is simply across the nook,” he defined.
Like massive language fashions, world basis fashions are basic to advancing robotic and AV improvement, but not all builders have the experience and sources to coach their very own, Huang mentioned.
Cosmos integrates generative fashions, tokenizers, and a video processing pipeline to energy bodily AI techniques like AVs and robots.
Cosmos goals to deliver the ability of foresight and multiverse simulation to AI fashions, enabling them to simulate each attainable future and choose optimum actions.
Cosmos fashions ingest textual content, picture or video prompts and generate digital world states as movies, Huang defined. “Cosmos generations prioritize the distinctive necessities of AV and robotics use circumstances like real-world environments, lighting and object permanence.”
Main robotics and automotive firms, together with 1X, Agile Robots, Agility, Determine AI, Foretellix, Fourier, Galbot, Hillbot, IntBot, Neura Robotics, Skild AI, Digital Incision, Waabi and XPENG, together with ridesharing large Uber, are among the many first to undertake Cosmos.
As well as, Hyundai Motor Group is adopting NVIDIA AI and Omniverse to create safer, smarter automobiles, supercharge manufacturing and deploy cutting-edge robotics.
Cosmos is open license and obtainable on GitHub.
Empowering Builders With AI Basis Fashions
Past robotics and autonomous automobiles, NVIDIA is empowering builders and creators with AI basis fashions.
Huang launched AI basis fashions for RTX PCs that supercharge digital people, content material creation, productiveness and improvement.
“These AI fashions run in each single cloud as a result of NVIDIA GPUs are actually obtainable in each single cloud,” Huang mentioned. “It’s obtainable in each single OEM, so you can actually take these fashions, combine them into your software program packages, create AI brokers and deploy them wherever the purchasers wish to run the software program.”
These fashions — supplied as NVIDIA NIM microservices — are accelerated by the brand new GeForce RTX 50 Sequence GPUs.
The GPUs have what it takes to run these swiftly, including help for FP4 computing, boosting AI inference by as much as 2x and enabling generative AI fashions to run domestically in a smaller reminiscence footprint in contrast with previous-generation {hardware}.
Huang defined the potential of recent instruments for creators: “We’re creating an entire bunch of blueprints that our ecosystem may benefit from. All of that is utterly open supply, so you can take it and modify the blueprints.”
Prime PC producers and system builders are launching NIM-ready RTX AI PCs with GeForce RTX 50 Sequence GPUs. “AI PCs are coming to a house close to you,” Huang mentioned.
Whereas these instruments deliver AI capabilities to non-public computing, NVIDIA can be advancing AI-driven options within the automotive business, the place security and intelligence are paramount.
Improvements in Autonomous Autos
Huang introduced the NVIDIA DRIVE Hyperion AV platform, constructed on the brand new NVIDIA AGX Thor system-on-a-chip (SoC), designed for generative AI fashions and delivering superior practical security and autonomous driving capabilities.
“The autonomous automobile revolution is right here,” Huang mentioned. “Constructing autonomous automobiles, like all robots, requires three computer systems: NVIDIA DGX to coach AI fashions, Omniverse to check drive and generate artificial information, and DRIVE AGX, a supercomputer within the automobile.”
DRIVE Hyperion, the primary end-to-end AV platform, integrates superior SoCs, sensors, and security techniques for next-gen automobiles, a sensor suite and an energetic security and degree 2 driving stack, with adoption by automotive security pioneers comparable to Mercedes-Benz, JLR and Volvo Automobiles.
Huang highlighted the important function of artificial information in advancing autonomous automobiles. Actual-world information is restricted, so artificial information is crucial for coaching the autonomous automobile information manufacturing unit, he defined.
Powered by NVIDIA Omniverse AI fashions and Cosmos, this strategy “generates artificial driving eventualities that improve coaching information by orders of magnitude.”
Utilizing Omniverse and Cosmos, NVIDIA’s AI information manufacturing unit can scale “a whole lot of drives into billions of efficient miles,” Huang mentioned, dramatically rising the datasets wanted for protected and superior autonomous driving.
“We’re going to have mountains of coaching information for autonomous automobiles,” he added.
Toyota, the world’s largest automaker, will construct its next-generation automobiles on the NVIDIA DRIVE AGX Orin, working the safety-certified NVIDIA DriveOS working system, Huang mentioned.
“Simply as laptop graphics was revolutionized at such an unbelievable tempo, you’re going to see the tempo of AV improvement rising tremendously over the subsequent a number of years,” Huang mentioned. These automobiles will provide functionally protected, superior driving help capabilities.
Agentic AI and Digital Manufacturing
NVIDIA and its companions have launched AI Blueprints for agentic AI, together with PDF-to-podcast for environment friendly analysis and video search and summarization for analyzing massive portions of video and pictures — enabling builders to construct, take a look at and run AI brokers wherever.
AI Blueprints empower builders to deploy customized brokers for automating enterprise workflows This new class of associate blueprints integrates NVIDIA AI Enterprise software program, together with NVIDIA NIM microservices and NVIDIA NeMo, with platforms from main suppliers like CrewAI, Each day, LangChain, LlamaIndex and Weights & Biases.
Moreover, Huang introduced new Llama Nemotron.
Builders can use NVIDIA NIM microservices to construct AI brokers for duties like buyer help, fraud detection, and provide chain optimization.
Obtainable as NVIDIA NIM microservices, the fashions can supercharge AI brokers on any accelerated system.
NVIDIA NIM microservices streamline video content material administration, boosting effectivity and viewers engagement within the media business.
Shifting past digital functions, NVIDIA’s improvements are paving the best way for AI to revolutionize the bodily world with robotics.
“The entire enabling applied sciences that I’ve been speaking about are going to make it attainable for us within the subsequent a number of years to see very speedy breakthroughs, shocking breakthroughs, typically robotics.”
In manufacturing, the NVIDIA Isaac GR00T Blueprint for artificial movement era will assist builders generate exponentially massive artificial movement information to coach their humanoids utilizing imitation studying.
Huang emphasised the significance of coaching robots effectively, utilizing NVIDIA’s Omniverse to generate hundreds of thousands of artificial motions for humanoid coaching.
The Mega blueprint permits large-scale simulation of robotic fleets, adopted by leaders like Accenture and KION for warehouse automation.
These AI instruments set the stage for NVIDIA’s newest innovation: a private AI supercomputer referred to as Venture DIGITS.
NVIDIA Unveils Venture Digits
Placing NVIDIA Grace Blackwell on each desk and at each AI developer’s fingertips, Huang unveiled NVIDIA Venture DIGITS.
“I’ve yet another factor that I wish to present you,” Huang mentioned. “None of this could be attainable if not for this unbelievable venture that we began a few decade in the past. Inside the corporate, it was referred to as Venture DIGITS — deep studying GPU intelligence coaching system.”
Huang highlighted the legacy of NVIDIA’s AI supercomputing journey, telling the story of how in 2016 he delivered the primary NVIDIA DGX system to OpenAI. “And clearly, it revolutionized synthetic intelligence computing.”
The brand new Venture DIGITS takes this mission additional. “Each software program engineer, each engineer, each artistic artist — everyone who makes use of computer systems immediately as a device — will want an AI supercomputer,” Huang mentioned.
Huang revealed that Venture DIGITS, powered by the GB10 Grace Blackwell Superchip, represents NVIDIA’s smallest but strongest AI supercomputer. “That is NVIDIA’s newest AI supercomputer,” Huang mentioned, showcasing the system. “It runs the complete NVIDIA AI stack — all of NVIDIA software program runs on this. DGX Cloud runs on this.”
The compact but highly effective Venture DIGITS is predicted to be obtainable in Might.
A Yr of Breakthroughs
“It’s been an unbelievable yr,” Huang mentioned as he wrapped up the keynote. Huang highlighted NVIDIA’s main achievements: Blackwell techniques, bodily AI basis fashions, and breakthroughs in agentic AI and robotics
“I wish to thank all of you to your partnership,” Huang mentioned.
See discover relating to software program product data.