Nvidia CEO Jensen Huang delivered the keynote last night at CES in Las Vegas: NVIDIA CEO Jensen Huang Keynote at CES 2025.
Key takeaways:
Blackwell is “in full production”… “enormous demand”. A focus on scaling and the enormous amount of compute each scaling (pre-training, post training and Test time inference reasoning) requires – driving “enormous demand” for Blackwell – FROM ONE TO THREE SCALING LAWS! Re-iterated Blackwell improves performance/Watt by 4x and performance/$ by 3x – so with AI datacenters power being capped, if perf/W is 4x better Blackwell, datacenters will be able to 4x their revenues by transitioning from Hopper to Blackwell. Remember Nvidia is now on an annual product cadence, which means that the next generation architecture Rubin is just around the corner.
New product announcements: 21 total press releases from NVDA worth going through all: Latest News | NVIDIA Newsroom. 1. COSMOS Platform –a physical AI platform with new foundation models and video data to run humanoid robots, autonomous vehicles (AV), and vision AI (Robots company Figure AI, Uber among early adopters & Hyundai adopting NVDA AI and Omniverse) 2. Blackwell-based GeForce RTX 50 Series GPUs – consumer GPUs for gamers, creators, and developers – Desktop and Laptop GPUs powered b by Blackwell architecture (92bn transistors 3,352 trillion AI operations per second (TOPS of compute) 3. NVIDIA NIM microservices – foundational models running locally on NVDA RTX AI PCs that is a vital for developers building AI agents and assistants using AI blueprints for tasks like customer support, fraud detection, supply chain optimisation etc; 4. NVIDIA DRIVE AGX Hyperion – end to end autonomous driving platform for passenger and commercial vehicles (Mercedes, JLR, Toyota, and Volvo said to be adopting the platform); 5. Project DIGITS – GB10 superchip SoC based on Grace Blackwell architecture for AI personal computers – with CUDA cores and Tensor cores connected via NVL Chip to Chip interconnect to high performance NVDA Grace CPU (20 power efficient cores with Arm architecture).
The next big thing in AI is Physical AI, where our physical world meets AI and where the real 3D world becomes fully simulated so that embodied AI can understand and operate in our analog world. The two largest use cases are AVs and humanoid robots, creating new multi-trillion dollar industries. Cosmos (robotic frontier models) x Omniverse (operating system for creating digital twins of our physical world) is Nvidia’s operating system software to enable this new industry, getting us closer to the world in the 1990s movie “MATRIX” where humans co-exist seamlessly in the real and also parallel digital Multiverses (Metaverse will surely make a comeback soon).
We continue to think that Nvidia is still a misunderstood business (despite its Market Cap of $ 3.7 trillion already) - it is so much more than just an AI chip company, with a full stack solution not only for Generative AI but also for agentic computing across the Enterprise and Physical AI. It is not far-fetched to think that Nvidia has all the cards to become the first company ever to reach a Market Cap of $ 10 trillion over the next 5 years, in our view.