Nvidia’s CES 2025 showcase impressed with RTX 50 GPUs, Project Digits, and a push into physical AI. Yet, the absence of an AI PC processor raised questions, and DeepSeek’s disruptive AI model claims rattled the market, wiping 18% off Nvidia’s stock (before the inevitable rebound). While DeepSeek’s long-term impact is unclear, Nvidia faces fresh challenges to its dominance.
What do we think? Jensen Huang’s CES 2025 keynote had all the hallmarks of an Nvidia spectacle—cutting-edge hardware, bold AI ambitions, and a future-forward vision. Yet, despite unveiling the GeForce RTX 50 series, Project Digits, and the Nvidia Cosmos platform, the presentation felt oddly disjointed. The innovations were significant, but the absence of an Arm-powered AI PC processor left a noticeable gap in Nvidia’s road map. Demos at the show were sometimes impressive but far from perfected. Subsequent news makes the wobbly start to the year seem oddly apt with DeepSeek undermining some of Nvidia’s glory.
Thoughts on Nvidia post-CES 2025
The RTX 50 series, built on the Blackwell architecture, showcased impressive advancements, with the RTX 5090 pushing gaming visuals to unprecedented levels through DLSS 4’s Multi Frame Generation and AI-assisted rendering. No, that’s not hyperbolic—I’m really not sure I’ve seen a better-looking gaming demo than the RTX Kit one.

Another standout announcement was Project Digits, a compact AI supercomputer designed to bring powerful AI capabilities to every developer’s desk, democratizing AI accessibility beyond traditional data centers. Digits contains a 20-core Arm-based CPU (10X Coretex-X925 and 10X Cortex-A725) co-designed with MediaTek.
At a competitor’s party on the night of the keynote, all the chat was about when Nvidia would just bite the bullet and announce an actual AI PC part to compete with Qualcomm and Intel. No one questioned the assumption such a part will also be designed with MediaTek, but how it will be sold is a matter of contention between the two companies, with MediaTek not thrilled about the junior partner status it has enjoyed in its ventures with Nvidia in the automotive space.
In comments at the show, Huang suggested a happy world where MediaTek might sell the part, or provide it to Nvidia to sell, characterizing it as “a great win-win.” Our understanding is the discussions about who will sell and who will get what portion of revenue are far less “win-win” and yet to be resolved. Still, we expect the announcements in 2025 with something concrete at May’s Computex with machines on show from Lenovo and other partners and shipping by the holiday season.

Nvidia took over the conference center at the Fountainebleu hotel for its press and analyst briefings, filling it with cars, robots, etc. Despite some pop-up issues, for me, the standout was Nvidia’s RTX Kit demo, possibly the loveliest looking gaming demo I’ve ever seen, enabled by RTX Mega Geometry creating hundreds of millions of animated triangles through real-time subdivision surfaces.

Beyond gaming, Nvidia is steering AI into the physical world with its Cosmos platform, a foundation model designed to enable robots and autonomous vehicles to predict, simulate, and act with near-human foresight. Nvidia positions this as the next phase of AI, integrating training, simulation, and real-world deployment through its three-computer architecture. Supercomputers like DGX train foundational AI models, the Omniverse platform provides digital twins for testing, and the Jetson Thor runtime system enables real-time AI execution in the physical world.
Autonomous vehicles remain a flagship example of physical AI in action, with Nvidia’s Drive AGX and Drive Hyperion platforms gaining traction among automakers like Toyota and Hyundai. These vehicles blend perception, generative intelligence, and real-time decision-making, demonstrating how AI can transition from digital efficiency to real-world impact. For Nvidia, this shift is a crucial business strategy. The move from software-driven AI to hardware-centric AI creates long-term revenue opportunities through chips, platforms, and infrastructure.
While Nvidia champions physical AI as the pinnacle of AI development, questions remain. Digital AI optimizes massive systems, and agentic AI operates autonomously in digital environments. Physical AI, though slower to scale due to real-world constraints, has the potential to redefine human-machine interaction. Whether the future is dominated by intelligent digital agents or autonomous physical systems, physical AI may prove a necessary hook for Nvidia’s future growth. We wondered why it was positioned so prominently. Perhaps Nvidia had a premonition of the upcoming DeepSeek disruption (or even intelligence).
DeepSeek’s depth charge
DeepSeek, a Chinese AI start-up, claims to train advanced models at a fraction of the cost incurred by US giants like OpenAI. Using techniques like “mixture of experts,” it suggests a future with less reliance on Nvidia’s GPUs. Here at JPR, we have some skepticism about the robustness of the media story on DeepSeek, but there is something there for Nvidia to worry about beyond just the market sentiment that saw Nvidia’s stock dropping 18% after DeepSeek’s announcements (before a partial rebound the next day).
We broadly agree with the very smart Professor Neil Lawrence, DeepMind professor of machine learning within the Department of Computer Science and Technology at the University of Cambridge. He said, “I think the progress is unsurprising, and I think it’s just the tip of the iceberg in terms of the type of innovation we can expect in these models. History shows that big firms struggle to innovate as they scale, and what we’ve seen from many of these big firms is a substitution of compute investment for the intellectual hard work. I’ve been suggesting that this has made the conditions ideal for a Dreadnought moment, where current technology is rapidly rendered redundant by new thinking. I don’t think DeepSeek is it, because the innovations deployed are relatively incremental, but it shows that we’re still in the age of the Newcomen engine, there’s plenty of space for budding James Watts to emerge, and that they are less likely to come from established players.”
AI breakthroughs often come with hidden trade-offs in performance and scalability. Details on DeepSeek’s methods are limited, making it unclear if its cost savings can be replicated at scale. Beyond technical concerns, geopolitical factors could limit adoption. Many Western enterprises will be justifiably cautious about relying on Chinese AI solutions due to security and regulatory concerns. Nvidia’s entrenched CUDA ecosystem also gives it a strong competitive edge.

LIKE WHAT YOU’RE READING? INTRODUCE US TO YOUR FRIENDS AND COLLEAGUES.