100-Minute Exchange: Jensen Huang Discusses Three-Year Trillion-Dollar Revenue Goal Again—Just the Beginning

On the second day of the GTC conference, NVIDIA CEO Jensen Huang gave a media interview at a hotel near the GTC event venue. He sat at the center of the stage, still dressed in a black leather jacket, with a series of NVIDIA products displayed beside him.

A day earlier, Huang showcased NVIDIA’s new product lineup at the San Jose SAP Center, including seven chips based on the Rubin architecture, five racks, new space computing modules, and a batch of open-source models. Compared to industry expectations for CPO products and new chips, NVIDIA actually released more than anticipated. Huang also raised NVIDIA’s accelerated chip revenue forecast to $1 trillion.

In this interview, Huang’s discussion was more comprehensive, covering not only the newly announced products and outlook for the AI market but also his expectations for NVIDIA’s changes over the next ten years, the transformative impact of OpenClaw, and a prediction: under AI’s influence, people will become even busier.

NVIDIA Continues Rapid Growth

As an AI infrastructure provider, NVIDIA’s business performance and revenue serve as a barometer of AI demand. “Last year at this time, I saw $500 billion in order revenue from Blackwell and Rubin for 2025-2026,” Huang revealed in his previous speech. A day earlier, he disclosed that he now expects this revenue to reach $1 trillion between 2025 and 2027.

Regarding revenue forecasts, he reiterated today, “$1 trillion only includes revenue from Blackwell and Rubin chips.” He emphasized that this figure does not include income from CPUs, Groq, storage systems, Feynman architecture products, and other diversified businesses. NVIDIA’s estimate of this $1 trillion is based on business visibility and purchase orders, and the company is confident in this outlook.

“NVIDIA is still growing rapidly,” Huang said. Last quarter, NVIDIA secured the largest order volume in its history, and growth is accelerating. The way people use computers is changing, creating a huge demand for tokens.

“Previously, people used computers to input and retrieve data—that’s the old way of using computers. In the future, computers will essentially be manufacturing machines, systems that produce tokens. Currently, there are very few computers dedicated to token production; most are used for training AI. The process of AI inference generating more tokens has just begun,” Huang explained.

To meet the increasing computational demand, manufacturing capacity must be addressed. Huang responded that many manufacturing issues need solving. For example, storage must support AI’s memory needs—short-term, working, and long-term memory all need consideration. As computing systems expand vertically and horizontally, storage becomes a critical issue. Storage system performance must be significantly improved; NVIDIA is reinventing storage systems for AI.

Specifically, data centers employ various memory technologies, including HBM and LPDDR. NVIDIA was the first to use LPDDR4 in data centers, now using LPDDR5, with LPDDR6 in the next generation. To ensure supply, NVIDIA collaborates with Samsung to manufacture Groq chips and partners with every memory supplier.

Business Goes Beyond Chips

NVIDIA’s diversified business was fully showcased at this GTC conference. Even focusing solely on data center operations, NVIDIA no longer emphasizes GPU performance alone. The Rubin architecture platform includes seven chips and incorporates the Groq 3 LPU, which was launched after NVIDIA licensed Groq technology. The networking side has also seen increasing development, with the launch of NVL72, integrating 72 Rubin GPUs and 36 Vera CPUs, along with ConnectX-9 SuperNIC and Spectrum-6 SPX racks.

Unlike the past emphasis on GPU versatility, NVIDIA now highlights different computational capabilities required by various AI workloads, responding to the new markets driven by the explosion in AI inference demand.

In the interview, Huang revealed that by 2025, NVIDIA plans to invest heavily in AI inference. With innovations like NVLink 72, performance has increased 35-fold at a fifth of the cost. He also introduced the concept that tokens from different-sized models vary. While Rubin remains a key platform for current token production needs, new market segments are emerging. As models grow larger and context lengthens, inference speed must become faster. The new chip combinations will enable computing to meet diverse requirements.

NVIDIA’s close relationships with major cloud providers have been seen as a revenue guarantee but also a risk. In the interview, Huang emphasized diversification of customers beyond products.

“Many forget that NVIDIA’s business is far more than chips. NVIDIA helps clients build AI factories worldwide; they need not only chips but also software and other solutions,” Huang said. “Forty percent of NVIDIA’s revenue comes from diversified businesses like physical AI, including autonomous vehicles. This is a significant portion, and NVIDIA cannot operate these businesses solely as a chip manufacturer. Additionally, 60% of revenue comes from cloud providers, but Huang stressed that many of these cloud services are driven by NVIDIA’s products, which they rely on to deliver their services.”

Within this 40%, physical AI—including autonomous driving and robotics—is a key component. Both were highlighted at this GTC.

A day earlier, Huang stated that the era of autonomous vehicles with OpenAI-like capabilities has arrived. In the interview, he discussed NVIDIA’s development journey in this sector. Once accounting for 0% of NVIDIA’s business but responsible for 90% of costs, the automotive segment has grown dramatically. “Ten years ago, I started working on autonomous driving with just one person on the team. Now, thousands are involved. We initially spent a lot of time and money on this, sometimes with setbacks, but we stayed committed,” he said. NVIDIA’s autonomous vehicle business includes three computers for training, data generation, and simulation, and many automakers now purchase one or more of these computers from NVIDIA.

He also shared his outlook on robotics development. “Today, robots still face many issues, but they are mainly engineering problems. You can see them walking around and starting to do tasks. Once existing technology improves, solving the remaining problems will take less than five years. I am very confident we will see very capable robots,” Huang said. Although challenges remain, joint motion and cognitive issues are expected to be resolved within three years. Technologies like OpenClaw will run inside robots, complemented by visual language and action models to control speech and movement.

NVIDIA’s deep involvement in the AI ecosystem also attracts attention. The company is investing to deepen relationships with AI startups. “We are investing in the next ‘Google,’ the next ‘Amazon,’” Huang said. NVIDIA provides funding to promising companies to help expand the ecosystem rapidly.

He also emphasized the importance of cash flow, stating that NVIDIA needs cash to support suppliers, partners, and ecosystem investments, while maintaining substantial free cash flow. He revealed that NVIDIA will conduct share buybacks, allocating 50% of free cash flow to investor returns—higher than last year.

“The Future Will Be Super Busy”

In Huang’s “five-layer cake” theory of AI, applications sit at the top, directly creating value and driving demand for infrastructure. The recent global popularity of OpenClaw exemplifies an AI application and aligns with Huang’s optimistic view of Agentic AI products. At this GTC, NVIDIA launched the NemoClaw software stack for OpenClaw and held a “build-a-claw” event on-site.

Why is Huang so bullish on OpenClaw? He believes this product will focus the power of building intelligent agents onto a single platform, accelerating their adoption.

“When OpenClaw appeared, I realized we finally had an open-source, agentic AI system that could serve as a standard. We could contribute to this open-source project without scattering efforts across different initiatives. If we make this project good enough, every company can start building its own agentic AI strategy,” Huang explained. Over the next 30 or 60 years, more capabilities will be added to OpenClaw, similar to contributions made to Linux and other projects, creating a global platform for collaboration.

“Imagine installing OpenClaw with just one line of code, then learning how to use it. It can then perform tasks—like designing a kitchen, learning to use tools, trying repeatedly, and eventually mastering architectural design. Later, when asked to design a living room, it will do even better,” he said.

Regarding AI safety concerns, Huang acknowledged that open-source models for building agentic AI exist, but issues like security, governance, and privacy remain significant. He dismissed the idea of scaring people with “sci-fi” AI, emphasizing that AI is needed for many tasks, such as maintaining cybersecurity—like white blood cells protecting the body.

As OpenClaw and similar AI products become more widely used, Huang made a final assertion: AI will not eliminate jobs but will make people busier.

“Honestly, I feel busier than ever. Compared to six months ago, I am busier every day. The speed of project feedback is faster, and the number of projects is increasing,” he said. NVIDIA’s growth is faster than ever, driven by more AI technology enabling quicker work and accelerating all projects.

What impact will AI have on jobs? Huang believes that everyone will be busier. “Many say AI will take away jobs, but the truth is the opposite. Computers, the internet, and mobile devices have all made us busier. AI will do the same. Tasks will be completed very quickly—within 30 minutes—just by writing instructions, defining goals, and using intelligent agents. Everything will be back in human hands,” he explained. Previously, teams spent a month writing product specs, then moved on to other tasks; now, the same work can be done in 30 minutes, keeping people engaged in executing key tasks.

“We are all busier than ever before. When was the last time you sat on your porch, drinking lemonade and watching the sunset? I don’t remember. The last time I watched a movie was almost a hundred years ago,” Huang said. He predicts this trend of busyness will continue.

This busyness will also shape NVIDIA’s future development. Last year, NVIDIA had about 42,000 employees. Huang envisions that in ten years, NVIDIA will be “super busy,” with around 75,000 employees. The company will keep its size manageable but necessary, with 7.5 million intelligent agents working around the clock.

He believes this busyness is beneficial for both company growth and societal progress. Over the next decade, NVIDIA will be able to solve extremely complex problems. Just as issues once thought impossible to solve are now being addressed, in ten years, many seemingly impossible things will become feasible. Meanwhile, tasks that currently consume vast amounts of energy, time, and costs will be reduced to a fraction—billions of times less. “People will feel like superhumans,” Huang said.

“Forty years ago, when I graduated from school, the topics we discuss now didn’t even exist in science fiction. The number of problems humanity is now trying to solve is millions of times greater than what we could imagine 40 years ago. I am very confident we will solve many more problems in the future,” Huang said. Reflecting on the key risks of the past and upcoming year, he joked that his outlook is simple: don’t get fired, don’t get bored, and stay alive.

Regarding society, he believes employment will also change. “Many companies still lack enough labor, and robots can fill this gap, driving economic growth. Many people will be employed to manage machines and help intelligent agents grow. If you compare today to 100 years ago, it’s a straight line: more jobs, economic growth. We will have rewarding work, though some jobs will disappear and new ones will emerge. AI will change everything,” Huang concluded.