From General-Purpose Robotics Models to AI Computers: NVIDIA's Vision for the Future of AI

NVIDIA CEO Jensen Huang delivered a sweeping keynote at GTC 2025 in San Jose, California, on March 18, laying out the company's roadmap for artificial intelligence (AI) computing and unveiling a series of groundbreaking products and collaborations. Despite the ambitious announcements, Wall Street reacted coolly, with NVIDIA's stock (NVDA) closing down 3.43% on Tuesday and slipping an additional 0.56% in after-hours trading.

Four Waves of AI: From Perception to Physical

Huang began by framing AI's evolution into four distinct waves:

1. Perception AI: Early-stage AI focused on speech recognition and basic functionalities.

2. Generative AI: The dominant wave over the past five years, enabling the creation of text, images, and other media.

3. Agentic AI: The current era, where reasoning models allow AI to interact autonomously in digital environments.

4. Physical AI: The emerging frontier, bringing AI to real-world applications like humanoid robots.

Huang emphasized the surging demand for computational power, noting that generative AI requires 100 times more resources than initially expected. He projected that data center investments in AI infrastructure will surpass $1 trillion by 2028, fueled by rapid adoption of NVIDIA's chips.

Product Roadmap: Blackwell, Vera Rubin, and Beyond

In a highly anticipated announcement, Huang confirmed the launch of Blackwell Ultra GPUs in late 2025. These will be followed by the GB300 superchip, which combines two Blackwell Ultra GPUs with a Grace CPU.

Looking ahead, NVIDIA plans to roll out the Vera Rubin AI superchip in 2026 and its upgraded Vera Rubin Ultra in 2027. The company also unveiled its next-generation chip, Feynman, slated for a 2028 release, continuing its tradition of naming chips after renowned scientists.

Compact AI Computers for Developers

NVIDIA introduced two new AI computers designed for developers and researchers:

• DGX Spark: Dubbed the "world's smallest supercomputer," this compact device features the GB10 Grace Blackwell superchip, delivering up to 1,000 TOPS of AI performance. Priced at $3,000, it is designed for developers, researchers, and students, with preorders open now and shipments expected this summer.

• DGX Station: A more powerful system featuring the GB300 Grace Blackwell Ultra, offering 20,000 TOPS of AI performance and up to 784GB of memory. Pricing details remain undisclosed, but the product is expected to launch later this year.

Dynamo: Scaling AI Factories

Huang also unveiled NVIDIA Dynamo, an open-source software platform aimed at scaling inference models for AI factories. Dynamo claims to boost inference performance by up to 30 times under the same architecture and GPU count, positioning it as a critical tool for AI-driven industries.

GR00T N1: A Leap in Robotics

In a significant move toward addressing global labor shortages, NVIDIA launched the GR00T N1, the world's first open, fully customizable general-purpose foundational model for humanoid robotics. The model is part of NVIDIA's collaboration with Google DeepMind and Disney on the Newton robotics platform, which was showcased through "Blue," an advanced humanoid robot.

Strategic Partnerships

NVIDIA announced several high-profile collaborations across key industries:

• Autonomous Driving: Expanded its partnership with General Motors to optimize factory planning and enhance in-vehicle AI systems using platforms like Omniverse and DRIVE AGX.

• AI-Native 6G Networks: Partnered with T-Mobile, Cisco, and others to develop hardware and software for next-generation 6G networks.

• Quantum Computing: Established the Accelerated Quantum Research Center (NVAQC) in Boston to integrate quantum hardware with AI supercomputers, advancing quantum technologies.

Cosmos: A Foundation for Physical AI

NVIDIA revealed the Cosmos™ World Foundation Model (WFM), an open, customizable inference model for physical AI. Powered by NVIDIA's Omniverse and Cosmos platforms, it enables large-scale synthetic data generation for training robots and autonomous vehicles. Companies like 1X, Agility Robotics, and Uber have already adopted Cosmos to accelerate innovation.

"Cosmos represents a breakthrough for Physical AI," Huang said. "It opens the door to unprecedented advancements in robotics and physical industries."

Conclusion

With a robust product roadmap, strategic partnerships, and a bold vision for the future, NVIDIA's GTC 2025 keynote underscored its leadership in AI computing, robotics, and quantum technologies. Despite short-term market reactions, the company is setting the stage for transformative advancements across industries, addressing global challenges with cutting-edge innovation.