On the eve of NVIDIA's earnings report, scheduled for after the U.S. market close on May 20, CEO Jensen Huang and Dell Technologies CEO Michael Dell shared a stage for a Bloomberg TV interview.
Speaking at the Dell World conference in Las Vegas on May 19, the two executives provided insights on the deployment of AI agents and supply chain constraints. Dell stated that AI has moved from testing and evaluation into production deployment, with enterprises seeing efficiency gains of 10x, 20x, or even 100x after redesigning workflows, far exceeding mere percentage improvements. Both leaders identified memory as the primary bottleneck currently constraining growth, followed by advanced process node semiconductors, noting that overall demand in the semiconductor supply chain continues to outpace supply growth.
When asked about his recent trip to China, Huang remarked, "The demand for AI in China is extremely robust, just like it is here. Agentic AI is also making tremendous progress there. I believe over time, the market will gradually open up... I was there representing the United States, which was an honor. I was there to support President Trump, and that was the real focus of my trip."
**AI Moves from 'Testing' to 'Working': The Enterprise as the Next Battleground** Michael Dell provided a concrete figure: over the past year, Dell has added 1,000 new customers for AI servers, bringing the total to 5,000. He characterized this shift as a move "from testing and evaluation into production deployment," citing real-world implementations by global leaders like Eli Lilly and Samsung. "These are not demos on a screen; these are the largest companies in the world deploying in the real world," he said.
Dell emphasized the magnitude of efficiency gains, stating enterprises are achieving "not 10%, 20%, or 30% improvements, but 10x, 20x, even 100x improvements." He believes the enterprise market holds the real, massive opportunity and that this wave is "just the beginning."
Jensen Huang offered a technical explanation. He noted that early AI largely ran in the cloud, but enterprise data—secure, proprietary, and containing business-specific knowledge—resides on-premises. AI agents must be deployed where the data is. "Intelligence has to be generated where the context is. Wherever the context is, wherever the action is, that's where you need to generate the intelligence," Huang stated.
He defined the current era as the "Agentic AI" phase, distinct from early generative AI. "Generating content is important, but actually getting work done is an incredibly valuable thing."
**The Biggest Bottlenecks: Memory and Advanced Process Chips** When asked directly about the biggest supply constraints, Michael Dell responded, "Without question, memory is a challenge. Advanced node semiconductors are also still under pressure. Overall, the semiconductor supply chain is continuing to ramp, but demand is growing faster than supply."
Huang supplemented this by detailing NVIDIA's approach—planning its supply chain two to three years in advance, aligning HBM (High Bandwidth Memory), advanced packaging, and platforms like Grace Blackwell. However, he conceded, "It's just that demand far exceeds the overall capacity of the planet. It's an industry-wide issue."
The root cause lies in a structural shift in AI workloads. Huang described a scenario of exponential scaling: in the future, billions of people will be supported by hundreds of billions of AI agents. "Humans use tools occasionally, but agents use tools all the time, everywhere, very frequently," he said, likening AI agents to "digital workers" that will require corresponding compute, memory, and infrastructure.
To address this demand, Huang revealed he personally outlined the future need to Micron CEO Sanjay Mehrotra years ago and had similar early discussions with SK hynix leadership. "We have multi-decade, deep relationships with these partners... They also see that we're winning in the marketplace, so they're more inclined to work deeply with us," he said. Yet, he acknowledged the difficulty of prediction: "If you're trying to forecast 2027 demand in 2023, it's quite difficult. Building fabs also takes a long time."
**Huang on China: Robust AI Demand, Market Will Gradually Open** Fresh from the trip to China with former President Trump, Huang was pressed on the outcome. His response was measured: "The demand for AI in China is extremely robust, just like it is here. Agentic AI is also making tremendous progress there. I believe over time, the market will gradually open up."
When pressed on whether selling chips to Chinese tech firms was discussed, Huang shifted focus: "I was there representing the United States, which was an honor. I was there to support President Trump, and that was the real focus of my trip. President Trump had some meetings, and I look forward to them making the appropriate decisions."
Michael Dell stated that Dell has business in China and complies with all relevant restrictions but expressed hope for "more economic cooperation between the U.S. and China, which ultimately leads to better outcomes and prosperity for both, and for the world."
On the Taiwan Semiconductor Manufacturing Company supply chain, Huang called Taiwan "absolutely a central hub for global technology manufacturing and technology R&D" while emphasizing NVIDIA's push for U.S. manufacturing. "We are building factories at scale in the United States—chip factories, packaging factories, computer factories, and of course, AI factories," he said, adding that supply chain diversification "is possible and should be a goal for everybody."
**Huang: AI Buildout is a Decade-Long Cycle, 'We Are at the Very Beginning'** This was one of the key signals from Huang, directly relevant to market expectations for the AI infrastructure buildout cycle. "We are at the very beginning of AI infrastructure buildout—the literal beginning. We will be building this out for the next decade and longer," he stated.
He provided a longer-term narrative: after the current AI infrastructure phase, the next stage is "physical AI," where AI moves from the digital to the physical world. That era will "for the first time in human history, truly enable the $90 trillion physical industries of the world," and "that era hasn't even begun yet."
He also acknowledged the limitations of supply ramp-up speed: "The supply chain is doubling every year, maybe even quadrupling, but to keep up with the buildout over the next decade is an enormous challenge."
Comments