Written by: Techub News整理
At a recent global conference held by the Milken Institute, NVIDIA founder and CEO Jensen Huang engaged in an in-depth dialogue with the host. As a leader at the forefront of the AI computing wave, Huang's insights not only pertain to NVIDIA's strategy but also outline the future evolution path, challenges, and opportunities of the entire AI industry. This conversation touched on core topics such as technological evolution, industry restructuring, safety ethics, and geopolitical competition, providing a valuable window for understanding the current critical junctures of the AI revolution.
The "Useful" Turn in AI and the Thousandfold Demand for Computing Power
Huang pointed out that AI has undergone a fundamental shift in the past few months: it has become "useful." This turning point began two years ago with the emergence of ChatGPT, which demonstrated generative capabilities. Generativity is the foundation of thinking and reasoning — one must generate ideas to think and must generate instructions to use tools. When AI possesses generative abilities, the door to understanding and reasoning is opened.
Subsequently, the industry rapidly moved towards "agent AI." Huang specifically mentioned Anthropic's Claude Code, considering it as the first agent system capable of performing truly productive work (such as software coding). Coding is deemed a key initial application scenario because it essentially involves programming tasks that people wish to automate. “How many companies and how many people in the world do not want to program repetitive tasks?” he countered. This means the "utility" of AI will far exceed that of software engineers, permeating all businesses and sectors.
However, the realization of agent AI comes at a high cost. Huang emphasized that compared to purely generative AI, completing the entire loop of understanding, reasoning, planning, using tools, and taking action requires a thousand times more computation. He used two vivid analogies: in two years, the number of cars needed worldwide has increased a thousandfold, and the number of aircraft has increased two thousandfold. This exponential growth in demand is the fundamental reason for the current surge in GPU consumption, with even older models from years ago appreciating in value like "fine wine."
This demand arises not only from the surge in computing power needed for single tasks but also because the user base will expand “one hundredfold.” The multiplying effect of both drives the foundational infrastructure for computing power to undergo unprecedented expansion. Huang believes this marks that AI is not merely an application but has fundamentally reshaped the computer industry, creating an entirely new industry.
He likened this new industrial ecosystem to a "five-layer cake": from the base level of energy, land, electricity, and facilities, to chips, system infrastructure, then to model layers, and finally to application layers. He particularly pointed out that people often focus only on models, but without the supporting foundational infrastructure, no useful models will emerge. The application layer — covering all industries such as healthcare, transportation, retail, and financial services — will be revolutionized by artificial intelligence.
Investment Bottlenecks and "Reindustrialization" Opportunities
Confronted with such immense demand, where do the bottlenecks lie? Huang admitted that the bottlenecks are dynamically changing. Two years ago, there was an abundance of energy but a shortage of chips. Now, the computing systems built by NVIDIA are extremely complex, with one rack valued at around four to five million dollars, weighing three tons, and containing one and a half million components; a data center is filled with these racks the size of soccer fields. These systems blend silicon photonics technology, state-of-the-art memory, 3D packaging, liquid cooling, and other cutting-edge technologies.
NVIDIA collaborates with almost all global chip companies and system manufacturers, possessing one of the largest supply chains in the world, but bottlenecks will always appear. Huang disclosed that NVIDIA's investment strategy is to strategically invest at every layer of the "five-layer cake" to activate a $100 AI ecosystem with a $1 investment. For instance, early investments in infrastructure companies like CoreWeave and Nebius were initially questioned as “circular trading,” but these investments have proven to be highly forward-looking, injecting confidence into the entire ecosystem.
He shared a key observation: in the past three to six months, gross margins for most AI-native companies, including OpenAI and Anthropic, have become “extremely considerable.” Once they start to become profitable, the goal is to expand production, which is why they are fiercely competing for computing capacity. Huang hopes this indicates that the AI ecosystem has begun to become self-sustaining.
Beyond commercial investments, Huang also sees a macro perspective on the AI-driven “reindustrialization” opportunity. He pointed out that AI is creating a vast number of jobs, and the best opportunities for American reindustrialization are rooted in this. This requires three types of factories: chip factories, computer factories, and AI factories. He estimates that this will lead to a reconstruction scale of trillions of dollars. Market forces are the strongest driver of reindustrialization. He cited the "Chip Act" as an example, pointing out that initially companies hesitated about building factories in the US, but when NVIDIA committed to offer hundreds of billions of dollars in orders, suppliers flocked to set up factories in the US.
Moreover, AI is also an excellent opportunity for modernizing the electrical grid. The American power grid is somewhat outdated, and now, for the first time, market forces have the opportunity to drive significant investments in sustainable energy (whether nuclear or other forms). Regarding the energy sector, Huang stated that if there are good ideas and NVIDIA can make a unique contribution, he is more than willing to invest, but currently focuses more on ensuring adequate funding support for land, power, and facilities.
AI Safety, Open Source, and Geopolitical Competition
When the topic shifted to concerns about AI, Huang positioned himself as a "pragmatist." He emphasized that ensuring AI safety is the responsibility of the tech industry, as the industry understands the technology best. Safety measures include redundant systems, diversified sensors, and protective technology. He acknowledged that today's protective systems are already very good but still not perfect and need constant improvements in real-world usage, just like airplanes, cars, and medical systems become safer over time through usage.
His greatest concern is not that other countries gain access to AI, but that Americans develop an excessive fear of AI. “Everyone should have AI,” Huang stressed, “it empowers people, enhances them, and gives them superpowers.” His biggest worry is that sci-fi-style fears will make AI unpopular in the United States, causing people to shy away from it, thus allowing the US to lose its leading position. He reminded that the US benefitted from the last industrial revolution not because it invented it but because it applied it.
Discussing the technological competition and export controls with China, Huang's viewpoint is clear and pragmatic: American companies should compete globally. The US should have the first, the most, and the best technologies, but should also maximize exports since exports increase revenue and taxation, thus enhancing national security and economic security. He posed a simple "test": everyone present needs AI, but no one needs nuclear bombs or F-35 fighter jets. AI is not a weapon.
In the debate on whether powerful models like Mythos (referring here to a certain advanced AI model, the transcription may be inaccurate) should be widely deployed, Huang provided a unique cybersecurity perspective: the way to defend against superpowers is not to possess another superpower but to have a large number of inexpensive powers. The best answer, in fact, is open source. Open-source models are now excellent and cost-effective. We can train a large number of open-source models to defend against cyber threats, forming a “white blood cell cluster” or "cybersecurity dome." Threat actors must decide which "front door" to attack, while defenders can possess more AI. One cannot expect their AI to be better than the opponent's, but one can expect their AI to be more numerous than the opponent's.
Regarding government regulation, Huang believes “it is absolutely necessary.” For example, future AI-assisted medical imaging systems must be regulated like medical instruments; autonomous vehicles should also be required to pass a driving test to obtain a "license" like human drivers. He agrees with swift action but opposes “breaking things.” “The benefit of quicker action is that better technology is safer.”
In response to Geoffrey Hinton and other "AI apocalypse theorists" claiming that AI has a 20-30% chance of ending humanity, Huang replied that they are wrong for “ignoring that there are so many excellent people in the world working hard to prevent these things from happening.” Many are dedicated to making AI smarter, but ten times as many are working to ensure it is safe, controllable, does not generate hallucinations, and produces useful work. He warned that if fear leads to panic, it would actually harm ourselves.
AI and Employment: Separation of Purpose and Task
Huang offered a profound rebuttal to concerns about AI replacing jobs: people confuse the "purpose" and "task" of work. He cited radiologists as an example: ten years ago, renowned computer scientists predicted that AI would first replace radiologists because computer vision had surpassed humans in analyzing scan images. Today, AI has indeed become 100% integrated into radiology, but the work of radiologists has not disappeared. On the contrary, they are processing more scans, receiving more patients, and making more precise diagnoses; radiology has become one of the largest profit centers in hospitals, requiring hiring more radiologists.
“Your purpose in life is not to sit in a dark room staring at a workstation studying scans,” Huang said, “your purpose is to collaborate with doctors to help treat patients, diagnose diseases, and help people recover. Studying scans is just one task you execute.” AI automates tasks but frees humans to pursue a higher purpose in work.
He illustrated with his own work: “I spend 100% of my time typing and speaking. Both typing and speaking have been completely automated, and the level of automation is beyond human capability. I should have lost my job long ago. However, you and I have noticed that we are working harder than ever.” This reveals the goal of capitalism: to make us more efficient, give us more free time, and enable us to find new and better ways to apply our intellect.
Huang acknowledged that there are bound to be "mismatches" during the transition period. For instance, now if a university graduate does not have proficient AI skills, their competitiveness will be far inferior to those proficient in AI. Some jobs purely based on tasks (like answering restaurant reservation calls) may disappear, but employees can shift to serving customers in the store. He emphasized that not every job will disappear; many new jobs will be created, some jobs will be eliminated, but every job will be affected. The first wave of job growth created by AI has already become evident: construction of data centers, chip factories, AI factories, and the frantic hiring at AI companies. Last year, venture capital invested over hundreds of billions in AI startups, the largest scale of investment in human history, all flowing into jobs. Meanwhile, the demand for software engineers is not decreasing but increasing.
“People need to think about these issues with more life experience and wisdom rather than merely from a technical perspective,” Huang summarized.
Belief, Responsibility, and an Infinite Future
At the end of the conversation, the host mentioned a potential wealth tax proposal in California, estimating that this could cost Huang about $8 billion. Huang's response reflects his consistent pragmatism and belief: he certainly prefers low tax rates, but he doesn’t mind paying taxes. “We love this country... that’s our way of giving back.” He and his wife have never struggled with it. They chose California for its schools, companies, and culture, rather than for low tax rates. He humorously expressed that he just hopes the government uses the taxes he pays to fix the pothole on Highway 101.
On the controversy surrounding Anthropic's collaboration with the US Department of Defense, Huang expressed a clear stance: he believes that if technology created by American companies is used for national defense, to protect the country and families, in ways that comply with the Constitution and laws, he has no objections. He believes that a CEO is not an elected official, and when the country goes to war, he does not wish to receive a call asking whether his technology should be used. “I would respect their judgment... but one thing we will not do is hinder America from defending our families.” He believes this is the operation method of a democratic country.
Finally, when asked what surprising things he has been thinking about recently, Huang's optimism was palpable. He communicates with professors and scientists every day and sees that AI is fundamentally changing research. New ideas that required months of exploration in the past can now be completed in a day with the help of AI. “The essence of a scientist’s work is to discover, explore, and push the boundaries of human knowledge.” The breakthroughs in fields such as energy science, climate science, biology, healthcare, drug discovery, and physical sciences are astonishing.
“If you could see everything I see every day, you would be filled with passion and excitement for the future, realizing that regardless of your past ambitions, the only thing you should say to yourself now is: your ambitions are still not high enough.” Huang stated that his fundamental transformation is that now, if someone tells him something can be done, he multiplies that expectation by 100 in his mind. AI is about to fundamentally change every domain of science and industry, and the future is incredibly vast.
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。
