Nvidia CEO Says Market Got it Wrong On Software Stocks | Insight with Haslinda Amin 6/1/2026

Key Concepts

• Agentic AI: A new computing paradigm where AI models (the "brain") use a "harness" to orchestrate tools, memory, and reasoning to perform productive work autonomously.
• AI Factories: Large-scale, complex infrastructure (data centers) designed to generate "tokens" as profitable units of revenue.
• Vera Rubin: Nvidia’s next-generation, full-stack AI computing platform, designed for high-throughput, low-latency agentic AI.
• CUDA X Libraries: A collection of specialized software libraries (e.g., cuLitho, cuOpt) that act as "skills" for AI agents.
• Co-design: The methodology of integrating hardware (GPUs, CPUs, networking) and software from the ground up to maximize performance per watt.
• Compute as Revenue: The economic principle that in the AI era, compute capacity directly correlates to revenue generation.

---

1. Main Topics and Key Points

• The Arrival of Useful AI: Jensen Huang declared that "Agentic AI" has arrived. Unlike previous generative AI, these systems can observe, reason, plan, and use tools to execute complex tasks.
• Productivity Multiplier: Huang argued that AI is not a threat to software engineers but a massive multiplier. He cited GitHub data showing that while the number of software engineers is increasing, their output has tripled, effectively turning $3 trillion in salaries into $9 trillion in economic productivity.
• The Shift in Computing Patterns: The industry is moving from traditional applications (code running in an OS) to distributed, heterogeneous systems where an agent orchestrates various components (GPUs for thinking, CPUs for tool use, DPUs for security) across a data center.

2. Real-World Applications

• Software Development: AI agents generating code and managing pull requests.
• Manufacturing/Engineering: Using AI to generate CAD files from simple prompts for 3D printing replacement parts.
• Infrastructure: The rapid buildout of "AI Factories" by companies like CoreWeave, Nscale, and various regional cloud providers (e.g., Indosat, GMI) to serve global and local AI demand.

3. Methodologies and Frameworks

• The Agentic Framework: An agent consists of:
  • Model: The "brain" (Large Language Model).
  • Harness: The "body" (orchestration software).
  • Tools/Skills: Libraries (CUDA X) that the agent learns to use.
  • Runtime: The environment where the agent executes tasks.
• Infrastructure Design (DSX): Nvidia’s approach to building AI infrastructure, moving from individual chips to full-stack systems (RTX for GPUs, DGX for systems, DSX for infrastructure).

4. Key Arguments

• Compute is Revenue: Huang emphasized that because tokens are now profitable, compute demand is the primary constraint for global industry. Therefore, performance per watt is the most critical metric for profitability.
• Software Longevity: By using Nvidia’s CUDA ecosystem, developers ensure their software assets remain useful across hardware generations (Hopper, Blackwell, Vera Rubin), lowering the Total Cost of Ownership (TCO).

5. Notable Quotes

• "Agentic AI has arrived. Useful AI has arrived." — Jensen Huang
• "People talk about AI reducing jobs. Complete nonsense. It’s causing more software engineers to be hired." — Jensen Huang
• "Compute is revenue. Performance per watt is your revenue." — Jensen Huang

6. Technical Terms

• MVLink 72: A high-speed interconnect technology that allows massive clusters of GPUs to act as a single, unified system.
• KV Caching: A memory management technique used in LLMs to store context, crucial for the "working memory" of agents.
• Omniverse: A digital simulation platform used to design and test AI factories in a virtual environment before physical construction.
• BlueField DPU: A data processing unit used for security and offloading infrastructure tasks from the CPU.

7. Data and Research Findings

• Taiwan Economic Growth: Mentioned that Taiwan’s annual GDP is expected to grow by nearly 10% due to the AI boom.
• Assembly Efficiency: The assembly time for a server rack has been reduced from two hours (Grace Blackwell) to five minutes (Vera Rubin) due to improved design and supply chain integration.
• Market Impact: The "AI trade" has broadened significantly, with server makers and physical AI/robotics companies seeing 20%+ growth in equity markets.

8. Synthesis and Conclusion

The keynote at Computex Taipei marked a transition from the "Generative AI" hype cycle to the "Agentic AI" implementation phase. Jensen Huang positioned Nvidia not merely as a chip manufacturer, but as the architect of the world’s new industrial infrastructure—the "AI Factory." By integrating hardware and software into a cohesive, secure, and highly efficient stack (Vera Rubin), Nvidia aims to solve the bottleneck of compute demand, framing AI as a fundamental driver of global economic productivity rather than a replacement for human labor. The focus on "performance per watt" and "time to first token" underscores the shift toward treating AI compute as a critical, revenue-generating utility.