Nvidia Reveals RTX 50 GPUs, Thor Auto Chip, and AI Supercomputer at CES 2025

Nvidia Product Announcements at CES 2025

On January 7, Jensen Huang, CEO of Nvidia, delivered the opening keynote at the 2025 Consumer Electronics Show (CES) in the United States.

During a two‑hour presentation, he unveiled the RTX 50 series GPUs based on the Blackwell architecture, the new automotive AI chip "Thor", and the personal computing platform Project Digits. He also hinted that the era of humanoid robots is approaching.

RTX 50 Series GPUs

The RTX 50 series, featuring the Blackwell GPU, contains 92 billion transistors—three times more than the previous generation—and supports a trillion floating‑point operations per second. It introduces DLSS 4, which can predict future frames and generate results beyond the native frame rate.

Huang explained that the new GPUs can blend AI workloads with graphics workloads, using programmable shaders that now carry neural networks, enabling "neural textures", compression, and neural material shading for stunning visual quality.

Pricing starts at $1,999 for the RTX 5090, $999 for the RTX 5080, $749 for the RTX 5070 Ti, and $549 for the RTX 5070, with shipments slated for January.

Nvidia also highlighted its NVLink 36 and NVLink 72 interconnects, which serve most data centers worldwide, and announced a massive Grace Blackwell NVLink72 chip that will combine 72 Blackwell GPUs, surpassing the performance of the world’s fastest supercomputers.

Humanoid Robotics

Huang announced Nvidia’s focus on humanoid robots, citing collaborations with domestic partners such as Yushu Technology H1 and Xiaopeng Laotie.

The Nvidia Isaac Groot platform will support developers with four key components: a base robot model, data pipelines, a simulation framework, and the Thor robot computer.

He also introduced Nvidia Cosmos, a world model that converts Omniverse‑created 3D scenes into photo‑realistic images, enabling AI agents to learn tasks more efficiently.

Automotive Innovations

The "Thor" processor is described as a revolutionary robot computer for cars, delivering 20× the compute power of the previous generation.

Huang announced that Nvidia Drive OS has become the first certified programmable AI computer for functional safety (ASIL‑D) and listed partners such as BYD, Li Auto, Xiaomi, Zeekr, Tesla, Mercedes, Toyota, Aurora, and Continental.

He projected Nvidia’s automotive revenue to reach $5 billion in fiscal year 2026.

Project Digits – Personal AI Supercomputer

Project Digits, powered by the Nvidia GB10 Grace Blackwell super‑chip, offers exaflop‑scale AI performance for researchers, data scientists, and students.

The GB10 chip integrates a Blackwell GPU with the latest CUDA cores and fifth‑generation Tensor cores, linked via NVLink‑C2C to a Grace CPU featuring 20 energy‑efficient Arm cores.

Each Project Digits unit provides 128 GB unified memory, up to 4 TB NVMe storage, and can run models with up to 200 billion parameters. Two units can be networked via Nvidia ConnectX to handle models up to 405 billion parameters.

Future AI Agents and Blueprint

Huang outlined Nvidia’s upcoming AI agents, built on Llama‑based models (Nemotron Nano, Super, Ultra) covering PCs, edge devices, and data centers.

Nvidia introduced a preview of the Nvidia AI Blueprint, leveraging the Metropolis platform, Cosmos Nemotron vision‑language model, Nemotron LLM, and NeMo Retriever to enable developers to create and deploy AI agents for video and image analysis.

The blueprint integrates Nvidia AI Enterprise services, including NIM micro‑services for VLM and LLM, and advanced AI frameworks for retrieval‑augmented generation, achieving video processing speeds up to 30× real‑time.

These AI agents can perform chain‑of‑thought reasoning, task planning, and tool invocation, and can be combined with other agents to deliver complex AI services across edge and cloud environments.

Huang concluded that in the future, every company’s IT department will function like an AI‑agent‑driven human‑resources unit.