At NVIDIA’s (NASDAQ:NVDA) GTC 2025, CEO Jensen Huang delivered on his promise to detail the company’s latest advancements in artificial intelligence (AI) and hardware.
Key announcements included the Blackwell Ultra AI chip, the next-generation Vera Rubin platform and a glimpse into future product roadmaps.
The keynote emphasized the “tipping point of accelerated computing,” marked by a shift from retrieval to generative AI and driven by a combination of agentic and physical AI.
NVIDIA’s AI-powered future
Huang’s speech, delivered without a script, highlighted NVIDIA’s focus on the transformative power of AI, particularly in robotics and generative computing, while also touching on NVIDIA’s advancements in quantum computing with CUDA-Q, a platform for hybrid quantum-classical computing.
For self-driving cars, he showed how NVIDIA’s technology is used to train and simulate autonomous vehicles, explaining how the company will provide the complete system from the data center to the car itself.
Speaking of data centers, Huang addressed the critical role they will play in supporting the next stage of AI advancements. The company is focusing on optimizing data centers to handle the massive computational demands of AI, particularly for AI inference.
This involves a balance between speed and accuracy in token generation, crucial for cost-effectiveness. To support these needs, NVIDIA will deploy powerful configurations of its Blackwell GPUs. Each rack—an enclosure designed to hold multiple electronic equipment modules—is equipped with 8 Blackwell GPUs. This dense configuration will allow for a high concentration of processing power within a compact footprint in modern data centers.
NVIDIA also introduced the Dynamo operating system, designed to manage and optimize large-scale AI infrastructure like data centers and “AI factories”, which are designed to produce AI models and capabilities at scale with intensive computation, data processing and model training. Huang mentioned NVIDIA’s collaboration with Perplexity, one of his “favorite, favorite partners”, on this project, but didn’t provide specific details.
The Omniverse and Cosmos software, which together will create simulated environments for training robots on synthetic data, is intended to leverage the Dynamo operating system for efficient deployment and execution within these AI factories.
The unveiling of NVIDIA Groot N1 – a dual-system architecture for humanoid robots – and its open-sourcing, were significant highlights. Groot N1 allows robots to perform complex tasks, like handling objects and following multi-step instructions, addressing anticipated labor shortages by 2030.
In terms of graphics, Huang demonstrated improvements in real-time ray tracing, a technique for creating more realistic images. He also hinted at future GeForce graphics cards, suggesting that they will be smaller, use less power and perform better than current models.
Blackwell and Vera Rubin: NVIDIA’s next-generation hardware platforms
Hardware advancements were also central, with updates on the production of the Blackwell system highly anticipated. Huang stated that the Blackwell system is now in full production with architectural improvements, including increased transistor density and optimized data pathways for AI workloads to deliver 1 exaflop of FP4 performance, a 25x increase over the previous Hopper architecture.
NVIDIA also unveiled Blackwell Ultra, a higher-performance variant of the Blackwell GPU designed for demanding AI workloads, slated for release in H2 2025. Later, Huang detailed the Vera Rubin platform, NVIDIA’s next-generation platform that will succeed Blackwell in H2 2026. The Vera Rubin platform features the Rubin GPU, which will utilize HBM4 memory, and the Vera CPU. An enhanced version of the Rubin GPU, Rubin Ultra, utilizing HBM4e memory, is also planned for 2027.
Forging strategic partnerships for future technologies
Partnerships were another key theme, with announcements including:
A partnership with the telecom industry to develop “AI-native” wireless network hardware for upcoming 6G networks.
Collaborations with DeepMind and Disney Research on the Newton physics engine, aimed at improving AI training through real-time simulation.
How did NVIDIA’s share price perform?
NVIDIA’s share price has fluctuated following a record-shattering run in 2024 that saw the company briefly surpass Microsoft (NASDAQ:MSFT) and Apple (NASDAQ:APPL) as the world’s most valuable on more than one occasion.
NVIDIA’s record high of US$149.43, recorded on January 6, has been in decline since due to macroeconomic factors combined with speculation that the company could be past its peak. Its customers have turned to competitors like Broadcom (NASDAQ:AVGO) or are working to develop chips of their own. The company also faced setbacks rolling out its Blackwell product line and is challenged by export restrictions to China, a large customer base.
Despite this, NVIDIA reported strong financial results for the fourth quarter of 2025, exceeding analyst expectations with significant revenue growth driven by high demand for its AI solutions. Following those results and Huang’s optimistic remarks about the demand for the Blackwell architecture, NVIDIA’s share price saw a 3.67 percent increase.
However, NVIDIA’s share price dropped over 3 percent in early trading on Tuesday, hours before Huang was set to take the stage, following a report from The Information on Amazon’s lowered cost of its AI chips.
As the keynote progressed, NVIDIA’s share price saw a slight uptick but declined by 3.35 percent to close at US$115.43, followed by an additional decrease in after-hours trading.
Securities Disclosure: I, Meagen Seatter, hold no direct investment interest in any company mentioned in this article.
