“We’re Not Far Ahead of China Overall”: Jensen Huang Sounds the Alarm on the Global AI Race
Inside the CNBC interview that redefined the AI conversation — Jensen Huang explains NVIDIA’s strategy, the race with China, and why AI’s next industrial revolution has only just begun.
NVIDIA’s Jensen Huang sat down with CNBC’s Squawk Box on October 8, 2025, and offered a rare, unfiltered look at what is powering the world’s most valuable company. The conversation ranged from OpenAI and xAI to energy strategy, China, immigration, and the economics of tokens. Every reference and quotation here is sourced to CNBC’s interview and companion clips on CNBC.com.
Huang confirmed that NVIDIA’s relationship with OpenAI has shifted from primarily cloud-channel sales to direct supply. He framed the move as a natural evolution of a partnership that began when NVIDIA delivered the first DGX system to OpenAI in 2016, as he told CNBC. What makes it consequential is the scope. He described NVIDIA’s offering as full AI infrastructure, not just GPUs, but CPUs, networking, switches, and software. That matters because the company is optimizing the entire stack to push down the cost of token generation. The tone was confident and specific, and the message to investors was straightforward. NVIDIA wants to be the single vendor that can ship a turnkey AI factory.
“I’m surprised they would give away 10%” on AMD’s OpenAI deal
Asked about AMD’s equity-sweetened arrangement with OpenAI, Huang called it imaginative and surprising, adding that he would not have expected a sizable stake to be granted before the next generation product ships. The contrast with NVIDIA’s approach is telling. NVIDIA has the option to invest alongside others in model builders but does not hinge supply on exclusivity, according to Huang’s remarks to CNBC. The strategy is to keep running faster across the system, not to win on one chip alone.
Huang’s most investor-sensitive line was about unit economics. He told CNBC that NVIDIA’s roadmap aims to slash token generation costs by an order of magnitude or more versus the prior generation. He argued that Moore’s law alone would imply incremental gains, while system-level optimization can deliver step changes. If that trajectory holds, it supports the thesis that AI demand expands as capability rises and costs fall, which in turn supports data-center capex and sustained platform share.
Huang drew a line between novelty and utility. Early models generated attention. New models, he said on CNBC, reason, research, use tools, and deliver work that enterprises will pay for. He cited Cursor, an AI coding assistant he says is now used across NVIDIA engineering, and highlighted OpenEvidence for clinical workflows. His framing was expansive. Previous tech gave humans tools. This wave is the first to give technology the capacity to use tools on its own. That distinction underpins his view that we are in the buildout phase of permanent AI infrastructure.
On CNBC, Huang praised xAI and said his only regret was not investing more. He rejected the idea that such stakes are vendor financing, calling them strategic investments in future giants that will operate on NVIDIA platforms. He used similar language for CoreWeave and other ecosystem bets. The pattern is visible. Seed the demand side with capital and software, then meet it with annual leaps in systems, networking, and compilers.
“We’re not far ahead of China overall”
Huang’s geopolitical answer was nuanced on CNBC. He said the United States leads in chips, that China leads in energy, and that infrastructure and models are close. He warned that China’s application layer moves fast because it is less regulated at the industrial level. He argued the U.S. must remain the global standard setter by ensuring the American tech stack is adopted widely, rather than ring-fencing it. In his view, winning the AI race means powering the majority of the world’s developers and applications, not only holding the most advanced chips at home.
“Data centers should generate their own power”
Addressing concerns about rising wholesale electricity prices near AI campuses, Huang told CNBC that operators should add behind-the-meter generation, from natural gas today to nuclear over time. He praised policy that enables private generation. The subtext is practical. AI factories face both compute and power constraints. NVIDIA’s thesis is that the ecosystem must scale both at once, and that energy is part of the AI stack.
On immigration, Huang was personal. He said on CNBC that NVIDIA will pay the new H-1B fee, but warned that such a cost would have barred his own path. He called immigration the foundation of the American dream and urged pragmatic refinements that keep the world’s top talent in the United States after earning American degrees. The message fit his broader competitiveness theme. Great chips need great people.
Huang closed by telling CNBC that reasoning models are so good they demand exponentially more compute, and that demand itself is compounding. He credited NVIDIA’s culture with channeling new wealth into purpose. Employees, he said, keep showing up because the work feels consequential. If token costs fall while utility rises, he believes this buildout becomes a durable industrial platform.
Conclusion
The CNBC conversation distilled NVIDIA’s edge into three ideas. Own the system, not a part. Collapse token costs every year to unlock new utility. Seed and standardize the global AI stack so developers and enterprises build on it. Huang did not claim a victory lap. He offered a roadmap. Cheaper tokens, smarter agents, faster energy and networking, broader diffusion. If those dominos fall in sequence, the company’s ambition to supply AI factories at global scale becomes more than plausible. It becomes the default setting for the next decade of computing.