Nvidia workforce to be dominated by AI agents in a decade
Jensen Huang expects digital workers will vastly outnumber human employees at Nvidia while also revealing plans to restart mainland China operations and declaring autonomous driving a solved problem
Nvidia CEO Jensen Huang believes his company’s workforce will be dominated by artificial intelligence (AI) agents, with human employees expected to be vastly outnumbered by digital workers.
Speaking to reporters in a wide-ranging media question-and-answer session on the sidelines of GTC 2026 in San Jose this week, Huang talked up his vision for the future of enterprise work and how Nvidia is navigating the geopolitical tightrope of US-China tensions and global semiconductor supply chains, among other areas.
“In 10 years, we’ll have, hopefully, 75,000 employees – as small as possible and as big as necessary,” he said. “Those 75,000 employees will be working with seven and a half million agents. The agents will be working around the clock… We are going to solve some really incredible problems.”
Rather than putting human engineers out of work, Huang argued that the increasing adoption of agentic AI will only add to human workloads.
“It used to be that you wrote the product specification, and then the teams would go off and work on it for a month. In the next month, you’re working on something else. Life is pretty casual. Now that a month has turned into 30 minutes, you’re on a critical path all the time. AI is going to cause us to be able to do things so fast, we’re going to end up doing more.”
Defending legacy software
With the fervour over tools like Anthropic’s Claude Cowork and OpenClaw, there have been concerns that AI agents may make legacy enterprise software platforms obsolete.
Huang rejected that premise, using electronic design automation (EDA) software suppliers like Cadence and Synopsys as examples. He argued that AI agents will not “manifest transistors from zero” using probabilistic generation. Instead, they will act as power users of existing enterprise software, fundamentally shifting the traditional software business model where growth is limited by the number of human users.
“The agentic engineers are going to use the same tools we use, because when we’re done with using the tool, it needs to put it back into the structured data that I can understand,” Huang explained.
“Is SQL going to die because agents are here? No. SQL is where the ground truth of the business is going to be stored… Now, because I have agents, the number of tools that we have to license is probably going to explode, not the other way around.”
Because engineering and enterprise work requires precise, deterministic outcomes and cannot afford to be probabilistic, AI agents will be forced to rely heavily on legacy software to verify and structure their work, he said.
China’s orders and Taiwan’s reliance
For Asian markets closely watching the geopolitical tug-of-war over semiconductor dominance, Huang gave a major update on Nvidia’s business in China amid ongoing US export controls: the company is starting its China operations under the purview of the Trump administration’s trade framework.
“Is SQL going to die because agents are here? No. SQL is where the ground truth of the business is going to be stored... Now, because I have agents, the number of tools that we have to license is probably going to explode, not the other way around.”
Jensen Huang, Nvidia
“We have licences for H200,” Huang said, referring to Nvidia’s previous generation Hopper graphics chips. “We have received purchase orders from many customers in China, and we’re in the process of restarting our manufacturing… and our supply chain is getting fired up,” he revealed.
He added that the US administration wants America to retain tech leadership but recognises the economic necessity of global commerce. “President Trump’s intention is that the United States should have a leadership position and access to Nvidia’s best technology. However, he would like us to compete worldwide and not concede those markets unnecessarily.”
However, Huang poured cold water on the idea that the US can rapidly decouple its semiconductor supply chain from Asia. Responding to questions about a US Commerce Department goal to move 40% of TSMC’s chip production to American soil, Huang suggested that the target is improbable in the near term.
“TSMC is doing their best to set up fabs and the supply chain in Arizona and around the United States. But as you know, demand is growing so fast,” he said. “While new fabs and plants are being built in the United States, the overall demand around the world is increasing so greatly. I think getting to 40% will be very challenging.”
Industrial leapfrog for Asia and Europe
During the session, Huang also made a pitch to the manufacturing sectors of Asia and Europe, offering a roadmap for leading industrialised nations that missed out on the internet software boom to reclaim global technological leadership.
Huang noted that countries such as Japan and Germany – historical leaders in the physical engineering discipline of mechatronics – were left behind during the IT revolution, which came to be dominated by the US because the “ship it before it’s fixed” culture of software development clashed with the safety-first ethos of manufacturing.
But the arrival of AI agents, which can follow natural language prompts to write their own code, changes the game. “OpenClaw does not need you to program it – it just needs you to tell it what to do,” Huang said.
“This is your opportunity to let the past be the past. It doesn’t matter anymore, because, as you know, software programmers don’t need to programme anymore,” he added. “That should be the happiest news in the world for Germany and Japan. Jump directly to AI and infuse the AI technology with the genius of your mechatronics industry, and all of a sudden, you become the robotics industry.”
Self-driving cars and robotics
Defending Nvidia’s automotive business, which currently accounts for only a fraction of its total revenues, Huang said the true scale of that business is masked by the fact that global automakers are buying massive Nvidia server clusters to train their AI models in datacentres, not just buying chips for vehicles.
He also boldly declared that self-driving cars is a solved problem. “The rest of it is engineering refinement,” he said, noting that once the industry hits one trillion miles per day of autonomous driving, it will become a multitrillion-dollar business.
As for robotics, Huang said that because AI can now reason and break down complex physical environments into sub-tasks, he believes robotics will advance rapidly with reasoning systems. “Once you see proof of the technology, the refinement takes less than five years,” he said. “I think you’re going to see robots that are extremely good.”
While global memory shortages will pose a threat to the broader PC market, AI PCs are gaining momentum across Asia as companies look to cut cloud costs, boost productivity and secure sensitive information.
While firms in mature markets are using AI agents to automate routine tasks, those in emerging markets, where the cost of the technology is higher than that of human labour, are favouring revenue-generating use cases.
Read more on Artificial intelligence, automation and robotics
