Jensen notes that AI is now a trusted tool for experts. He expected the technology to scale but was pleased with its better reasoning. New systems use routers to check if a model needs more research. This helps stop AI from making up information. It makes tools for doctors and lawyers much more reliable.

I think the whole industry addressed one of the biggest skeptical responses of AI, which is hallucination and generating gibberish. This year the whole industry, in every field from language to vision to robotics, saw the application of reasoning and grounding make big leaps.

Jensen explains that generating these smart responses is now a profitable business. Companies are seeing high margins because the results are very valuable. People are now willing to pay for reasoning tokens that provide high quality answers.

It is really terrific to see that we are now generating tokens that are sufficiently good, so good in value that people are willing to pay good money for it.

AI has also become central to global politics. Jensen spent much of the year traveling to discuss how AI strategy impacts national security. The technology is now part of every major conversation about jobs, labor, and energy.

The traditional AI community exhibits a strange paradox. People working most intensely to advance the field are often the ones most pessimistic about its future. This doomsday narrative has influenced media coverage and public perception despite the potential benefits in healthcare, education, and productivity. History shows that technology shifts change the types of jobs available rather than reducing the total number of jobs.

When I look at the traditional AI community, even before things were scaling, there was a strong doomsday component in the people working on AI. The people who were most trying to push the field forward were often the people who are most pessimistic.

While the narrative of job loss persists, technological evolution typically shifts employment toward new areas. The focus should remain on the positive impacts AI can bring to various sectors. Whenever there is a technology shift, there is a shift in the jobs that are important, but the result is still more jobs.

AI represents a fundamental shift in how software works. Traditional programs like Excel are written by engineers and then distributed as a finished product. AI is different because it generates every single token for the first time based on current context. This constant generation requires a new type of infrastructure. Jensen refers to these facilities as AI factories because they produce the digital tokens that power applications worldwide.

I call them AI factories because it is producing tokens that will be used all over the world. The reason why it is infrastructure is because it affects every single application. It is used in every single company, industry, and country.

The transition to AI is sparking the creation of three new types of physical plants. The world needs more chip manufacturing plants, specialized supercomputer plants, and large scale AI factories. These projects are currently being built across the United States at a massive scale. This construction effort is creating a significant surge in demand for skilled labor.

I am so excited to hear that electricians are seeing their paychecks double. They are being paid to travel like we go on business trips. It is really terrific to see that these three types of plants are creating so many jobs.

This boom benefits a wide range of professionals including construction workers, plumbers, and network engineers. The sheer amount of physical hardware needed to support AI means that the near term job outlook for these trades is very strong. Electricians in particular are seeing their roles change as they travel to support these new industrial sites.

A common prediction suggested that AI would eventually eliminate the need for radiologists. While AI now powers nearly every radiology application, the number of radiologists has actually increased. This occurs because of the distinction between the tasks of a job and its actual purpose. For a radiologist, the task is to study scans, but the purpose is to diagnose disease and conduct research.

The fact that they're able to study more scans more deeply, they're able to request more scans, do a better job diagnosing disease, the hospital's more productive, they can have more patients, which allows them to make more money, which allows them to want to hire more radiologists.

Jensen notes that his own daily tasks involve a lot of typing. Automating those tasks does not make him less busy. Instead, it allows him to be more productive and take on more work. This productivity enables a company to explore more ideas and grow. Sarah highlights that there is massive latent demand for services like better healthcare. As professionals become more efficient, they can serve more people and perform more research.

If Nvidia was more productive, it doesn't result in layoffs. It results in us doing more things. The more productive we are, the more ideas we can explore, the more growth as a result, the more profitable we become, which allows us to pursue more ideas.

If the world had no more problems to solve, productivity might reduce the size of the economy. However, since there are always more challenges to address, increased efficiency actually fuels economic growth and creates more opportunities.

There is a common fear that robots and AI will eliminate jobs, but the current reality is a severe global labor shortage. Many industries, from manufacturing to trucking, are struggling to find and retain workers. This labor gap is worsened by an aging population. Robotics will allow us to cover these critical shortages in factories and logistics, where people often prefer jobs that do not require constant travel away from their families.

The first part is that having robotic systems is going to allow us to cover the labor shortage gap, which is really severe and getting worse because of aging population. This is not only United States, all over the world.

New technologies also create entirely new categories of employment. Just as the rise of cars required a massive network of mechanics, a future with a billion robots will lead to the largest repair and maintenance industry on the planet. Sarah notes that while it is easy to worry about AI replacing roles like lawyers, there is a massive amount of latent demand in the economy that these tools will help satisfy.

It is important to distinguish between the tasks of a job and the ultimate purpose of a role. For example, Jensen points out that while a lawyer's task might include writing or reading contracts, their true purpose is to resolve conflict and protect the client. Automation can handle the repetitive tasks, but the core human purpose of the profession remains.

The purpose of the lawyer is to help you resolve conflict. That is more than reading a contract. It is more than writing a contract. The purpose is to protect you.

AI technology focuses on the automation of intelligence and can be visualized as a five-layer stack. Energy sits at the base, followed by chips, infrastructure, models, and finally applications. While AI is often associated with human language, its reach is much broader. It encompasses biological, chemical, and physical information. This diversity means AI is not just a chatbot but a fundamental tool for all kinds of data modalities.

The ability to understand human language and genome language and molecular language and protein language and amino acid language and physics language, all supremely well, that God AI just doesn't exist.

Open source is the essential innovation flywheel for this industry. Without it, companies in fields like healthcare or manufacturing would be suffocated. They rely on pre-trained models that they can fine-tune for specific domains. Open source already powers a large portion of global data centers and telephony. Jensen stresses that policymakers must protect open source to ensure that innovation continues across all sectors rather than being concentrated in a few monolithic entities.

The idea of a single, all-powerful God AI is an unhelpful distraction. AI is essentially the next evolution of the computer industry. Every nation and business needs computers, and soon they will all need AI. Instead of waiting for a galactic-scale intelligence that does not exist, the focus should remain on practical applications that move the world forward today.

Jensen believes the dystopian narratives surrounding AI are deeply unhelpful. While science fiction is enjoyable, using it to paint end of the world scenarios for government officials creates unnecessary fear and hinders the industry. Jensen suggests that when leaders advocate for these extremes, their intentions may not align with the best interests of society. Instead, they might be seeking regulatory capture to stifle competition from smaller startups.

I think we've done a lot of damage with very well respected people who have painted a doomer narrative, end of the world narrative, science fiction narrative. It is not helpful to people, it is not helpful to the industry, it is not helpful to society, it is not helpful to the governments.

The conversation moves to the definition of safety. Jensen argues that the primary component of safety is performance. A safe technology is one that functions exactly as it is supposed to. Just as the primary safety feature of a car is that it drives correctly rather than how someone might misuse it, an AI is safe when it is reliable and predictable.

The first part of safety of a product is that it perform as advertised. The first part of safety is performance. Suppose the first part of safety of a car isn't that some person is going to jump into the car and use it as a missile. The first part of the car is it works as advertised.

The industry has made significant strides in this area by focusing on grounding, reasoning, and research. Rather than slowing down, aggressive investment has made AI more functional and useful. Looking forward, the decreasing cost of AI will actually enhance safety. Instead of a single rogue agent, we will see millions of AI agents monitoring each other to ensure everything remains grounded and unbiased.

The cost of AI models is dropping at a staggering rate. In just one year, the cost of tokens for leading models fell over 100 times. While some argue that capital moats are rising due to high training costs, Jensen notes that the barriers to entry are shifting. What once required billions of dollars and massive supercomputers can now be replicated for much less. Jensen highlights that Nvidia improves its architecture and performance by five to ten times every single year.

In the case of AI, over the course of 10 years, it is probably 100,000 to a million times improvement. If you were to tell me that in the span of 10 years, we are going to reduce the cost of token generation by a billion times, I would not be surprised.

This pace far exceeds Moore's Law, which typically sees a 100 times improvement over a decade. Jensen suggests that narratives about extreme training costs are sometimes used to scare competitors out of the market. Because the cost of AI decreases by more than 10 times annually, a company that is only six months behind can remain highly competitive. Innovation is not just coming from the United States. Jensen points to the DeepSeek paper as a major contribution that benefited American startups and labs. It shows that the global research community is constantly finding more efficient ways to build frontier models.

The industry is also moving beyond just scaling for the sake of scale. While pre-training was the focus of the last few years, the focus is shifting toward post-training and specialized models. Jensen points out that it is no longer necessary to boil the entire ocean to be valuable. Companies can find success by building models that are better at specific tasks like coding or consumer accessibility.

The point of pre-training is to prepare yourself to do the real training. Now we call it post-training. It is just training. Training is where compute scaling directly translates to intelligence.

AI models are beginning to specialize in the same way humans do. Instead of trying to be good at everything, models and startups will find micro niches where they can become exceptionally good at specific tasks. Coding has emerged as a prime example of this specialization. It is likely the first category to produce a billion dollar AI native business. This shift reveals that software engineering is not just a niche, and the demand for it remains higher than ever.

At Nvidia, engineers use tools like Cursor pervasively. Jensen argues that the true purpose of a software engineer is to solve problems and discover new ones, while coding is merely one of the tasks involved. If an engineer's only value is writing code, they are at risk of being replaced. However, freeing engineers from the mechanics of coding allows them to focus on the company's many unsolved problems.

The purpose of a software engineer is to solve known problems and to find new problems to solve. Coding is one of the tasks. If the purpose is not coding, if your purpose literally is coding, somebody tells you what to do, you code it. All right, maybe you're going to get replaced by the AI. But all of our software engineers, their goal is to solve problems.

This distinction between purpose and task applies to many roles. A waiter's purpose is to ensure guests have a great experience, not just to take orders. If an AI takes the order, the waiter can focus more effectively on the overall guest experience. Similarly, the hardware supporting these tasks must remain flexible. Nvidia chooses programmable architectures over fixed ones because AI algorithms change faster than transistor technology. While dedicated chips might perform one job well, the rapid evolution from transformers to hybrid models requires a flexible foundation.

Moore's law is largely over. Transistor benefit is only 10 percent, maybe a couple of years, and yet we would like to have hundreds of X every year. The benefit is actually all in algorithms. And an architecture that enables any algorithm is likely going to be the best one.

Programmability also ensures a large install base. When software engineers optimize an algorithm, they want it to run on as many computers as possible. By protecting architectural compatibility, Nvidia ensures that everything from older models to the newest innovations can run across their entire stack. This approach drives costs down, making advanced models like Mixture of Experts more accessible and affordable for new applications.

Several industries are approaching a ChatGPT moment. This shift is driven by multi-modality, long context, and breakthroughs in synthetic data generation. One of the most promising areas is digital biology. While understanding individual proteins is already advanced, the next step involves multi-protein understanding and representation. Jensen notes that a new model called La Prutina helps with learning and generating complex protein structures.

The ChatGPT moment for digital biology. That moment is coming. And by digital biology I see proteins in the system. I think we are good at protein understanding now. Multi-protein understanding is coming online.

This evolution extends to molecule design and chemical generation. Because biological data is often sparse compared to human language, synthetic data will be essential. The major breakthrough will happen when researchers can train foundation models specifically for proteins and cells. Once these models exist, the data flywheel will accelerate generative capabilities in the field.

The real breakthrough is going to be when we can train a world foundation model, a foundation model for proteins, a foundation model for cells. I am very excited about both of those things. Once we have a foundation model, our understanding capability, our generative capability, that data flywheel is really going to take off.

Reasoning represents a major breakthrough for autonomous systems. While language models have improved significantly, the same logic is now being applied to cars. Instead of just relying on perception and planning, cars are becoming reasoning machines. This allows them to handle unfamiliar situations by breaking them down into known scenarios and constructing a logical path forward. Jensen notes that these cars will be thinking constantly to navigate through circumstances they have never encountered before.

So these cars are going to be thinking all the time and when they come up to a circumstance they have never encountered before, they can break it down into circumstances they have encountered before and construct a reasoning system for how to navigate through it.

Jensen identifies four distinct eras in the development of self-driving technology. The first era involved smart sensors and human-engineered algorithms that acted like digital rails. This was followed by the use of deep learning in specific modules, then end-to-end models, and finally end-to-end models with reasoning. If the industry had started just three years ago with current technology, it might have reached the same level of progress much faster.

There is also great optimism for robotics because the foundational AI technology is now ready. While humanoid robots face mechanical challenges such as weight and safety, the underlying AI can be applied to many different physical forms. This concept of multi-embodiment means a single general-purpose AI could control an excavator, a tractor, or a humanoid robot, much like a human can switch between different tools.

Everything that moves will be robotic. And everything that moves is a very large space. It is not all humanoid. Every AI will be multi-embodiment, meaning just like a human can sit in a car or pick up a chopstick, these systems will be general purpose.

The market for these technologies will likely see a surge in verticalization over the next five years. While general-purpose models might reach 90 or 99 percent accuracy, industrial applications require near-perfect reliability. Specialized companies will take core technology and refine it for specific sectors. This creates a significant opportunity for startups to build vertical solutions rather than competing only on the research and development scale of general-purpose giants.

People building technology often talk about professions like surgery or accounting without having lived those experiences. This lack of personal history can lead to a misunderstanding of what those jobs actually require. Jensen argues that developers must have more empathy for the depth and complexity of work they have not done themselves. It is easy to overlook the nuances of a role when you only see it from the outside.

Oftentimes the technology addresses the task, it doesn't address the purpose.

A major gap exists between completing a task and fulfilling a purpose. While AI might be able to handle specific technical actions, it often fails to grasp the broader intent behind a profession. True innovation requires understanding the purpose of the work rather than just automating the individual tasks within it.

Energy is the foundation of the current industrial revolution. Without a significant increase in energy capacity, progress in AI and industrial growth would stall. This growth is essential for national prosperity and the ability to address social issues. While the transition to sustainable sources is ongoing, every form of energy is needed to meet demand, including natural gas, nuclear, wind, and solar. For the next decade, natural gas is likely the only practical way to support the massive power requirements of AI clusters.

Without energy, there can be no new industry. Without energy, you can't have industrial growth. Without industrial growth, the nation can't be more prosperous. Without being more prosperous, we can't take care of domestic issues or social issues.

The demand for AI infrastructure is actually serving as a major driver for climate innovation. Because developers see clear demand, they are investing in massive battery companies, solar concentrators, and small modular reactors. This market demand is a powerful force that moves sustainable energy forward more effectively than policy alone. Sarah points out that even if these technologies take time to scale, the AI infrastructure problem is currently the biggest catalyst for green energy development.

There is also a necessary balance between caution and optimism. While some warnings about AI are sensible, excessive pessimism can be counterproductive. It can scare away the investments needed to make the technology safer and more useful. Technology is often the best path to safety. Modern cars are safer than those from 50 years ago because of technological advancement, and AI will follow a similar path as it becomes more functional and secure through continued innovation.

I think we're scaring people from making the investments in AI that makes it safer, more functional, more productive and more useful to society. It takes technology to be safe and technology to be secure. I'm delighted to see that the advancement of technology is still accelerating.

Jensen expresses optimism regarding the future relationship between the United States and China. He believes that the idea of decoupling is naive and lacks common sense. While it is healthy for both nations to seek a level of independence to avoid overly emotional dependencies, their economies remain deeply intertwined. A productive and constructive relationship between these two nations is the most important factor for the next century.

The idea of decoupling is naive. It is not based on any common sense. The more deeply you look into it, the more the two countries are actually highly coupled. Both countries ought to invest in their own independence. When you depend too much on someone, the relationship becomes too emotional.

Jensen argues that the current approach to export controls is grounded in national security and economic prosperity. A strong military depends on a strong economy. By allowing American technology companies to compete and win globally, the United States generates the wealth and tax revenue needed to maintain its global leadership. Jensen points out that while the great firewall blocked American internet companies, other parts of the US technology stack benefited immensely from China's growth.

China's Internet growth has been a boon for Intel and AMD selling CPUs, Micron selling DRAMS, SK Hynix and Samsung selling DRAMS. It is the second largest Internet market for American technology industry. Maybe it wasn't helpful to the Googles of the world, but don't exclude every layer of the stack.

He suggests looking at the entire technology stack rather than focusing on one layer. China is also the largest contributor to open source software globally. Many American startups rely on these open source contributions to innovate and grow. Viewing the relationship through a wide lens shows that China's development has created significant prosperity for America beyond just the internet service layer.

The core shift in the technology industry is the transition from general purpose computing to accelerated computing. This change is driven by the fact that CPUs can no longer handle the growing demands of modern data processing and machine learning. Even if chatbots did not exist today, the move to a new computing model would still make Nvidia a massive company. The foundation of computing itself is changing.

If chatbots did not exist today, Nvidia would be a multi-hundred billion dollar company. The reason for that is because the foundation of computing is shifting to accelerated computing.

The debate over an AI bubble often focuses too narrowly on the revenue of a few companies. However, these companies are primarily limited by their capacity to generate tokens. Just as a chip manufacturer needs wafers, AI companies need factory capacity to grow. When they get more capacity, their revenue tends to scale accordingly. Beyond chatbots, AI is transforming autonomous vehicles, financial services, and digital biology. Many quantitative traders are replacing human mathematicians with supercomputers to find predictive features in the markets.

A massive shift is occurring in global research and development. Around two percent of the world's GDP is spent on research, totaling about two trillion dollars. That investment is moving away from traditional methods toward AI-powered discovery. For example, large drug companies are moving their R and D from wet labs to supercomputers. Every startup and researcher is currently struggling to find enough computing capacity. This global demand suggests that the current growth is part of a deep structural change rather than a superficial bubble. Some critics point to slow enterprise adoption, but these observations often ignore the time required for organizational change and planning cycles.

There is a fundamental shift in how people think about that two trillion dollars of R and D. It used to be for the old way of doing things. It is now going to be the AI way of doing things. That is going to be powered by a whole bunch of infrastructure.

Enterprise adoption of new technology is often slow, but startups are making significant progress in conservative industries. In fields like healthcare, professionals are so focused on accuracy that they are quick to use tools that provide grounded research. This highlights a broader shift in user behavior. People do not actually want to conduct research or search through documents. They want answers.

Nobody wants to do research, they want answers. Nobody wants to do search, they want answers. The goal is to get answers. The goal is to get smarter. And these AIs allow us to help us do all that.

Jensen describes how he no longer reads through individual archive papers in the traditional way. The volume of new biomedical and technical knowledge is now an impossible ask for a human to track alone. He now loads interesting papers into ChatGPT to make it learn and summarize the content. He then interacts with the AI to extract the necessary insights.

Viewing AI as a multi-layer cake rather than just a chatbot helps simplify the narrative. This framework moves the focus from a single winning company to leadership across all domains. This perspective also emphasizes the need for physical infrastructure. To have AI, you need factories. To run those factories, you need energy. Without these layers, the entire system cannot function. A better understanding of this system makes the discussion about AI more pragmatic and balanced.

Advancing technology quickly is one of the best ways to ensure safety. Nobody would choose to drive a car from the first decade of automotive history because modern safety features like ABS and lane keeping are vastly superior. Jensen believes that industry progress in AI follows this same path. He feels proud that the industry made significant strides this year, proving that scaling laws remain intact. More compute consistently leads to more intelligence.

We want to keep people safe, but one of the best ways to keep people safe is advancing a technology quickly. And I think the industry is doing that, and I'm very proud of the industry for doing that.

Innovation in one sector now spreads across all other sectors with incredible speed. This rapid diffusion of intelligence suggests that the next five years will be an extraordinary period for AI development. While current progress is impressive, the upcoming year promises even more significant breakthroughs as these technologies continue to mature.