Conversation with NVIDIA Founder and CEO Jensen Huang: Satya Nadella at Microsoft Build 2026

Key Concepts

• Agentic Systems: AI systems capable of autonomous operation, task execution, and iterative problem-solving without constant human intervention.
• RTX Spark: A breakthrough PC-based AI system featuring a petaflop of AI performance and 128GB of memory, designed to run large-scale models locally.
• Grace Blackwell: A GPU architecture optimized for post-training, reinforcement learning, and reasoning models.
• Vera Rubin: The next-generation architecture specifically engineered for agentic AI, focusing on extreme low latency and massive-scale concurrent processing.
• NVFP4: A specialized numerical format developed by NVIDIA and Microsoft to optimize memory usage and performance for large parameter models.
• Confidential Computing: A security framework ensuring data is encrypted in transit, in use, and in storage.

---

1. The Evolution of the PC: From Tool to Personal AI

Jensen Huang and Satya Nadella discuss the transformation of the PC over their 30-year partnership. The PC has evolved from a passive tool into an autonomous assistant.

• Functionality: Users can now delegate complex tasks (e.g., coding, design modifications) to their PC via text commands. The PC autonomously fires up necessary software, iterates on the task, and completes the work while the user is away.
• Technical Specs: The RTX Spark system utilizes the NVFP4 format, allowing it to host "couple-of-hundred-billion parameter models" locally, effectively bringing data-center-level intelligence to the desktop.

2. Data Center Innovation: From Pre-training to Agentic Workloads

The speakers trace the evolution of AI supercomputing through three distinct generations:

• Ampere/Hopper: Focused primarily on pre-training large language models.
• Grace Blackwell: Shifted focus to post-training and reinforcement learning, enabling reasoning models. The NVLink72 architecture allows an entire rack to function as a single computer.
• Vera Rubin: Designed specifically for the agentic era. Unlike previous CPUs designed for human patience, Vera Rubin is engineered for "impatient" agents that require extremely low latency to iterate rapidly.
• Sustainability: The "Fairwater" design (utilizing Grace Blackwell) is highlighted as a "miracle of engineering," featuring a closed-loop liquid cooling system that is highly energy-efficient and environmentally friendly. It achieves a 30x improvement in token generation cost-efficiency compared to Hopper.

3. The "Agentic" Computing Framework

The core argument presented is that AI has reached a tipping point where it is now "useful" and "profitable."

• Evidence: GitHub commits have increased by a factor of three in recent months, signaling that agentic systems are actively contributing to productive work.
• Methodology: To support these agents, NVIDIA and Microsoft are ensuring that the entire software stack—including data processing, SQL, Spark, and semantic/vector/graph-based tools—is fully GPU-accelerated.
• Rationale: Agents require high-speed iteration. By accelerating the underlying data infrastructure on Azure, the system reduces the time-to-token, which directly increases the profitability and intelligence of the AI output.

4. Strategic Collaboration and Ecosystem

The partnership between NVIDIA and Microsoft is characterized by "speed of light execution," where hardware and software are co-designed long before chips are taped out.

• Foundry Integration: NVIDIA’s models and tooling are being integrated into Microsoft’s Foundry, and NVIDIA software is being used to accelerate Microsoft’s internal data warehouse workloads.
• Security: A major focus is placed on confidential computing, ensuring that the "long-term memory" (storage) and "working memory" of these agentic systems remain encrypted throughout the entire lifecycle.

5. Notable Quotes

• Jensen Huang: "The PC evolved from being an incredible tool to now being a tool that's used autonomously by an AI assistant."
• Jensen Huang: "Vera Rubin is a revolutionary CPU designed for agents... past CPUs were designed for humans, and we're just more patient than agents are."
• Satya Nadella: "The speed of light execution between the teams is fantastic to see."

Synthesis and Conclusion

The transition from pre-training to agentic AI represents a fundamental shift in computing architecture. By moving from the "personal computer" to the "personal AI" at the edge (RTX Spark) and scaling to massive, low-latency agentic clusters in the cloud (Vera Rubin), NVIDIA and Microsoft are building an ecosystem where AI is no longer just a chatbot, but an autonomous worker. The success of this transition relies on full-stack GPU acceleration, extreme energy efficiency, and a commitment to confidential computing, all aimed at maximizing the profitability and speed of token generation for developers worldwide.