In addition to the keynote speech at the heavyweight GTC conference, understand Huang Renxun's key points in his speeches to Wall Street and the Internet this week.

During the recently concluded GTC26, NVIDIA Corporation (NVDA.US) CEO Huang Renxun showcased the grand blueprint of this peak-valued enterprise to the outside world through an in-depth interview and financial Q&A. From "1 trillion dollars in order visibility" to "agents are the future of personal computers," Huang is not only selling chips but also reshaping the global IT industry's allocation logic. Regarding Token and value: If your engineer earns $500,000 a year and only spends $5,000 a year on tokens, I would go crazy. If that $500,000 engineer does not consume at least $250,000 worth of tokens, I would feel deeply disturbed. Even if chips are free, if they cannot keep up with technological advancements and our operational speed, they are still not cheap. Regarding agents and the future: Every engineer will have 100 agents. In the past, we wrote code, but in the future, we will write ideas, architectures, and specifications. Agent systems are systems that get the job done, they are helping our software engineers get the job done. Regarding Tokenomics: Computers used to be just tools, but in the future, they are manufacturing equipment. People buy these computers to produce tokens, and the efficiency of producing these tokens is crucial. You are buying the most expensive computer and producing the lowest-cost token at the same time. Regarding market demand and growth: We have strong visibility of demand, orders, and demand for Blackwell and Reuben of over 1 trillion dollars. Our growth rate is actually accelerating. Every software company, every company needs to have an OpenClaw strategy. Regarding competition and architecture: Anyone who says "my chip is 30% cheaper" is just proving that they do not understand AI. You simply have to breathe air until you finish breathing. After that, we will breathe compressed liquid air, but before that, how about air? It is free, and we have been using it for a long time. (Note: This refers to the limits of using copper cable/ wire technology) Financial nuclear bomb: 1 trillion dollars order wall Huang Renxun dropped a shocking number at the analyst conference: NVIDIA Corporation's demand visibility for Blackwell and Reuben architecture has exceeded 1 trillion dollars. Logic behind growth: This number is not inflated but based on confirmed purchase orders and factory construction pipelines. Huang emphasized that NVIDIA Corporation's advantage is its delivery cycle, which is much faster than companies designing ASIC chips themselves, even achieving "order placement in the same quarter, shipping in the same quarter." Margin moat: Faced with the questioning of "value being taken away by NVIDIA Corporation," Huang bluntly stated, "TSMC's wafers are the most expensive globally, but the most valuable, so I am willing to pay for them." He believes that customers aren't buying expensive computers, but the global cost-effective token production. Third turning point: From large models to "agents" Huang Renxun believes that AI has gone through two stages of generative and reasoning and is now at the third turning point - Agentic Systems. Token becomes the new salary: In the future, when companies hire engineers, they won't just provide laptops but also token budgets. If an engineer earning $300,000 per year does not consume tokens, then they are wasting productivity. Birth of Personal AI Computer: Systems like the open-source project OpenClaw, defined by Huang as "the world's first personal artificial intelligence computer." It has memory, scheduling, skills, and API, and is the operating system of the future IT industry. New hardware landscape: Vera Rubin and Groq's "marriage" NVIDIA Corporation is no longer just a GPU company but an "AI factory" company. Disaggregated Inference: This is the core of the Dynamo operating system. By breaking down inference tasks, chips of different capabilities can each play their role. Role of Groq: NVIDIA Corporation's acquisition and integration of Groq (LPX series) are not to replace GPUs but to utilize its extremely low-latency SRAM architecture for the "last step" of self-regressive reasoning. Trinity architecture: NVIDIA Corporation is the only company globally that can optimize HBM (high bandwidth memory), LPDDR5, and SRAM simultaneously, making competitors' single-point chips look like a "patchwork monster" through this "liquid-cooling rack" complete machine delivery. Physical AI: A $50 trillion blue ocean Huang is particularly bullish on Physical AI, believing that its eventual scale will surpass Digital AI. Reshaping traditional industries: This is a $50 trillion sector that has been almost a tech desert for the past 20 years. From Siasun Robot&Automation surgery, autonomous driving to intelligent base stations, Physical AI needs to comply with physical laws at the edge. Siasun Robot&Automation explosion in 3-5 years: Huang predicts that Siasun Robot&Automation will be everywhere in the next 3 to 5 years. Although China has a strong advantage in hardware supply chains like motors and rare earths, NVIDIA Corporation will provide the brain (training, simulation, three onboard computers). Industry symbiosis: NVIDIA Corporation is the "best seller" of cloud service providers In the face of threats from cloud giants' self-designed chips, Huang is extremely confident: Traffic brings sales: "AWS, Alphabet Inc. Class C, Microsoft Corporation have the largest booths at GTC because they want to promote services to my CUDA developers." NVIDIA Corporation guides developers to the cloud through the CUDA ecosystem, essentially serving as a customer acquisition engine for cloud service providers. Irreplaceability: 40% of business comes from non-cloud giants (regional clouds, enterprise private clouds), these customers buy the "full-stack platform" rather than just "chips." Without NVIDIA Corporation's complete solutions, these markets would be unreachable. This article is from "Wall Street Onlooker," author: Xu Chao, GMTEight editor: Chen Siyu.