Winning the AI Race Part 3: Jensen Huang, Lisa Su, James Litinsky, Chase Lochmiller

MP Materials’ Pentagon deal is an investment and guaranteed market, not a conventional subsidy. DoD is investing $400 million, taking equity and warrants, setting a commodity price floor, committing to 100% offtake for a planned 10x magnet facility, and sharing profits above a threshold 50/50.

TSMC’s Arizona fab is already matching Taiwan on AMD’s key yield metric, with a relatively modest cost penalty. Lisa Su said Arizona yields are equivalent to Taiwan’s and estimated U.S. production costs at more than 5% but less than 20% higher, roughly low double digits.

Even with U.S. reshoring, the chip supply chain would remain acutely exposed to a Taiwan disruption. Su said industry reserves would cover “months, not years,” while arguing that complete reshoring is unrealistic because lithography and other critical components will remain globally distributed.

Crusoe’s buildout shows AI infrastructure becoming physical megaproject construction at extraordinary scale. Its Abilene site targets more than 1.2 gigawatts and 400,000 NVIDIA GPUs with roughly 4,000 workers onsite daily, while Crusoe says its broader development pipeline totals about 40 gigawatts.

At NVIDIA, AI is already mandatory infrastructure for technical work rather than an experimental tool. Jensen Huang said every software engineer and chip designer uses AI, credited it with expanding the company’s capacity to pursue ideas, and predicted workers who ignore AI will lose out to peers using it.

NVIDIA’s older GPUs retain unusual economic value because software keeps improving their performance after shipment. Huang said Hopper hardware gained roughly fourfold performance through software, while estimating residual values around 75–80% after one year, 65% after two and 50% after three.

Huang views leading Chinese open models as strategically beneficial to the United States when they run on the American computing stack. He cited DeepSeek, Qwen and Kimi and argued their global adoption strengthens U.S. platform dominance because developers—including many in China—continue building on American technology.

Huang personally reviews compensation recommendations across NVIDIA’s 42,000 employees and says he always increases the company’s proposed spending. He described using machine learning to sort compensation data and confirmed maintaining discretionary equity grants, explaining his philosophy simply: take care of people and everything else follows.