NVIDIA at CES 2025: Jensen Huang’s Blueprint for the Next Era of AI, Robotics, and Gaming

At CES 2025, NVIDIA CEO Jensen Huang took the stage to deliver a keynote packed with massive hardware unveilings and profound insights into the trajectory of artificial intelligence. From the highly anticipated RTX 50 series GPUs to groundbreaking developments in physical AI and robotics, Huang's vision positions NVIDIA not just as a chipmaker, but as the foundational engine of the future computing landscape.

Here is a summary of the key announcements and insights from Jensen Huang's keynote.

1. The RTX 50 Series "Blackwell" GPUs Revolutionize Gaming and AI

NVIDIA officially unveiled the GeForce RTX 50 Series, powered by the new Blackwell architecture. The flagship RTX 5090 boasts a staggering 92 billion transistors and delivers twice the performance of the RTX 4090.

• Neural Rendering: Huang emphasized that modern computer graphics are now inseparable from AI. Using advanced AI and DLSS, the GPU can accurately compute a fraction of the pixels and predict the rest, resulting in mind-blowing rendering performance and efficiency.

• Unprecedented Value: To the excitement of PC builders, the RTX 5070 offers the performance of the previous-generation flagship RTX 4090, but at a price point of just $549.

  
  

2. The Three Scaling Laws of AI

Huang offered a masterclass on the evolution of AI, explaining that the industry's rapid advancement is being driven by three "scaling laws":

• Pre-training Scaling Law: The traditional approach of training base models on massive datasets (like the entire internet).

• Post-training Scaling Law: Using reinforcement learning, human feedback, and synthetic data to refine and fine-tune AI for specific reasoning tasks.

• Test-time Scaling Law: Allowing AI to "think," reflect, and reason before answering, which requires allocating vastly more compute during the actual inference phase.

  
  

3. Agentic AI & The Enterprise Digital Workforce

The next major leap in software is "Agentic AI"—systems of models that can reason, break down complex missions into actionable tasks, use tools, and retrieve data independently. Huang envisions a future where every company's IT department acts as the "HR department" for onboarding and managing AI agents.

To support this, NVIDIA announced new developer blueprints and the Llama-Neotron suite of models—fine-tuned versions of Meta's Llama 3.1 optimized explicitly for robust enterprise use.

4. Native AI on Every PC

While the AI boom started in the cloud, NVIDIA wants to bring it natively to local PCs. Huang announced a major push to optimize NVIDIA's AI stack for Windows WSL2 (Windows Subsystem for Linux). This bridging technology allows developers and everyday users to run sophisticated AI models locally on their RTX-equipped PCs right out of the box, without relying on cloud servers.

5. NVIDIA Cosmos: The "World Foundation Model"

One of the most significant architectural announcements of the night was NVIDIA Cosmos. While large language models (LLMs) understand text, Cosmos is designed to natively understand the physical world—including properties like gravity, friction, and object permanence.

By connecting Cosmos with NVIDIA Omniverse, developers can create grounded, realistic "digital twins" to train autonomous vehicles and robots safely in simulations before they ever touch the real world. To accelerate the global robotics industry, NVIDIA is open-sourcing Cosmos.

6. The "Three Computers" Strategy for Robotics & AVs

Huang detailed NVIDIA's comprehensive strategy for the multi-trillion-dollar robotics and autonomous vehicle markets, which relies on three fundamental computers working in harmony:

1. DGX: The supercomputer used to train the AI models.

2. Omniverse (Digital Twin): The simulation system used to test models and generate synthetic training data.

3. AGX (Edge): The autonomous computer inside the robot or car.

   
   

To power the edge, Huang announced that NVIDIA DRIVE Thor, their next-generation robotics and AV processor, is now in full production.

7. Project Digits: A Desktop AI Supercomputer

To close the show, Huang surprised the audience with "Project Digits". It is a compact, beautifully designed desktop AI supercomputer built specifically for researchers and developers. Created in collaboration with MediaTek, the device houses a new GB110 chip that packs NVIDIA's entire cloud computing and AI stack into a quiet, cute, desktop-friendly form factor.

Final Thoughts

Jensen Huang's CES 2025 keynote painted a vivid picture of a world rapidly transitioning from chat-based generative AI to physical, agent-driven AI. Through breakthroughs in consumer GPU hardware, accurate world simulation via Cosmos, and localized desktop supercomputers, NVIDIA continues to cement its position as the absolute bedrock of the AI revolution.

Watch the full keynote here: https://www.youtube.com/watch?v=k82RwXqZHY8