Written by: Techub News

Recently, NVIDIA founder and CEO Jensen Huang appeared on the renowned technology podcast All-In Podcast for a deep conversation lasting an hour. As a key driver and observer of the global AI wave, Huang not only elaborated on NVIDIA's strategic evolution from a hardware supplier to an "AI factory" operator during the interview, but also provided insightful judgments on critical topics such as the rise of agents, the explosion of inference computing demands, the future of open-source versus closed-source models, geopolitical challenges, and the profound impact of AI on employment. This conversation serves as a valuable window for understanding NVIDIA's future layout and the next wave of trends in the AI industry.

From GPU to AI Factory: NVIDIA's Strategic Reconstruction

At the beginning of the interview, Huang clearly stated NVIDIA's identity transformation: "We have evolved from a GPU company into an AI factory company." The core of this transformation is "Dynamo" — which he describes as "the operating system for AI factories." Dynamo is named after the machine invented by Siemens that converts mechanical energy into electrical energy, symbolizing the infrastructure driving the next industrial revolution.

The key technology behind Dynamo is "decoupled inference." Huang explained that the processing flow of modern AI inference is extremely complex and involves mathematical computations of different scales and forms. NVIDIA's idea is to decouple this process, allowing different parts to run on different types of processors to maximize the efficiency of heterogeneous computing. This concept has also led to NVIDIA's acquisition and integration of Mellanox (network technology) and Groq (inference chips).

Today, NVIDIA's computing ecosystem spans a diverse range of chips including GPUs, CPUs, switches, network processors, and Groq LPUs. Huang revealed that in the future data center architecture, up to 25% of space will be allocated to a combination of Groq LPUs and GPUs to handle increasingly complex and diverse AI workloads, particularly in agent computing.

This “decoupled” and “heterogeneous” thinking is NVIDIA's answer to the "inference explosion." Huang recalled that when he predicted last year that the demand for inference computing power would see a thousandfold, millionfold, or even billionfold growth, many found it too exaggerated, as the focus in the industry was still on training at that time. Now, inference has become the bottleneck for computing power. He announced that the next generation inference factory will have a throughput ten times that of current levels.

In response to external doubts about the high costs of NVIDIA's solutions (rumored to be between $40 billion and $50 billion), Huang made a strong rebuttal. He emphasized that the construction costs of the factory should not be confused with the final cost of generating each token. A $50 billion factory, due to its extremely high efficiency, may produce the lowest-cost tokens. He calculated that in data center costs, expenses for land, power, shell, storage, network, CPU, servers, cooling, etc., are fixed costs, and the cost differences of GPUs themselves make up a much smaller percentage of total costs than imagined. When a $50 billion solution can provide ten times the throughput, its unit cost advantage is decisive. "Even if chips are free, if they can't keep up with the pace of technological iteration, they are still not cheap enough," he concluded.

The Agent Revolution: The Paradigm Shift in Computing and the Power of Open Source

Huang views the rise of agents as the third major turning point following generative AI and inferential AI. He pointed out that while ChatGPT popularized generative AI, the real qualitative change was triggered by Claude Code and the recent hit Open Interpreter (referred to as "Open Claw" in the show).

He particularly emphasized the cultural and technological significance of Open Interpreter: it not only popularized the concept of AI agents among the public but more importantly, it redefined the form of computers. Open Interpreter features a memory system (short-term memory/file system), resource management, scheduling capabilities (like cron jobs), an I/O subsystem (connecting to WhatsApp, etc.), and various application "skills" through API calls. These four elements together create a blueprint for a "personal AI computer," a modern computing operating system that is open-source and can run anywhere.

The agent paradigm will lead to exponential growth in computing demand. Huang estimates that the demand for computing power will grow about 100 times from generative to inferential; and potentially another 100 times from inferential to agent-based. This means that within two years, the demand for computing could grow ten thousandfold. Moreover, people who pay for information will pay even more for "completing tasks." Agents are the tools for completing work, which will unlock tremendous economic value.

He illustrated this shift with an internal example: NVIDIA has about 43,000 employees, of which about 38,000 are engineers. He posed a thought experiment: an engineer with an annual salary of $500,000, if he only consumes $5,000 worth of AI tokens a year, would be quite dissatisfied. He believes that such high-value knowledge workers should consume at least $250,000 worth of tokens, akin to how chip designers must use CAD tools, representing a new productivity paradigm. In the future, engineers will no longer write large amounts of code by hand but will organize and manage hundreds or thousands of agents to accomplish tasks by writing ideas, architectures, and specifications.

About the debate between open-source and closed-source models, Huang believes that they are not oppositional but coexist and thrive together. For the vast majority of consumers, general-level layered models (like ChatGPT, Claude, Gemini) will continue to flourish as service products. At the same time, vertical industry expertise must be captured and customized through open-source models. The open-source model ecosystem is thriving and is close to the technological frontier. He observed that many startups are adopting a "open-source first, then shift to proprietary models" strategy, while NVIDIA's "router" technology allows users to access both the best closed-source models and fine-tuned specialized models simultaneously.

Physical AI, Robotics, and National Competitiveness

Huang identifies "physical AI" as one of the core tracks for NVIDIA's long-term strategy. He believes that this is the first opportunity for the tech industry to enter a $50 trillion industry with previously low technological penetration. To this end, NVIDIA has developed three major computing systems: computers for training AI models, computers for assessing models in the virtual physical world (Omniverse), and edge computing for robotics (from self-driving cars to teddy bear toys).

In the field of robotics, he predicts that from the emergence of high-performance prototypes to the launch of mature products typically only takes "two to three cycles," or approximately 3 to 5 years, and robots will be ubiquitous. He particularly mentioned China's strong advantages in microelectronics, electric motors, rare earth magnets, and other foundational supply chains for robotics, emphasizing that the global robotics industry will be deeply reliant on this ecosystem.

Huang anticipates that robots will become a larger economic liquidity unlocker than cars. When everyone owns a robot, they can open online stores, create products, and complete tasks that individuals could not accomplish independently. This will not only address labor shortages but will also spawn new scenarios such as virtual existence and space exploration.

Regarding geopolitics and the supply chain, Huang expressed a pragmatic attitude. Concerning the Chinese market, he revealed that after obtaining the relevant export licenses, NVIDIA has received procurement orders from Chinese companies and is restarting its supply chain. However, he emphasized that from a national strategic security perspective, the U.S. must ensure that it maintains global leadership in the AI technology stack (from chips to computing systems to platforms) to avoid repeating mistakes seen in the solar, rare earth, telecommunications, and other industries.

On AI regulation, he urged policymakers to base their strategies on technological facts and avoid being influenced by "apocalyptic rhetoric" and extreme emotions. AI is software, not a biological entity or extraterrestrial life, and humans have a considerable understanding of it. He expressed concern that if the U.S. slows down AI application out of fear or controversy, leading to technological diffusion lagging behind other countries, that would pose a real national security risk.

In response to the recent AI safety controversy surrounding Anthropic, Huang affirmed its focus on safety and technology but suggested that industry leaders should be more cautious and balanced when issuing warnings. "It's possible to predict the future, but we need a bit more humility... Making extreme, disastrous claims without evidence may be more harmful than people realize," he stated.

Future Prospects: Opportunities, Moats, and the Role of People

Regarding the revenue prospects of the AI industry, Huang is more optimistic than many observers. He believes that not only companies like Anthropic and OpenAI will grow rapidly, but every traditional enterprise software company will become a value-added distributor of AI models and tokens, leading to exponential market expansion.

When asked about "moats" for application layer companies, Huang's answer is: deep verticalization. Future winners will be those companies with the deepest knowledge in specific fields. They can inject specialized knowledge into proprietary agents, creating a flywheel effect with customers. This disrupts the traditional software company model of first creating a general platform and then selling customized services, allowing vertical experts to directly leverage powerful AI tools to build core competitiveness.

Finally, regarding the most pressing issue of AI and employment, Huang again demonstrated his characteristic optimism and pragmatism. He reiterated the widely circulated judgment: "You won’t be replaced by AI but by people who can use AI." He cites radiologists as an example: ten years ago, top computer scientists predicted that computer vision would eliminate radiologists. Ten years later, computer vision is now 100% integrated into radiology equipment, yet the demand and number of radiologists have significantly increased. This is because advancements in technology have made scans faster and cheaper, allowing hospitals to serve more patients, generating higher revenue and subsequently requiring more doctors to interpret the surge in scan results and serve patients.

"Every job has its purpose and task. AI has taken over tasks, but the purpose of human work — helping patients diagnose diseases — has become more important and scalable." He extends this analogy to the field of education: a richer, more productive society can equip classrooms with more teachers, while AI offers each teacher customized courses for each student, transforming all educators into "augmented" teachers.

Huang advises young people to delve deeply into sciences, mathematics, and language skills (as language has become the programming language of AI), and to become experts in using AI. He emphasizes that learning how to issue commands to AI, giving it creative space while guiding it to achieve objectives, is a crucial new skill.

The core message Jensen Huang repeatedly conveyed throughout the conversation is this: AI is a revolution of empowerment and creativity. It eliminates restrictive mindsets such as "this is too difficult," "this will take too long," and "we need a lot of people," liberating human creativity from repetitive labor and pointing towards a future of limitless productivity and possibilities. As the architect of this revolution, he sees not replacement and threat, but an unprecedented blueprint for growth and prosperity.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。