Huang names Land, Power, Shell as AI's bottleneck

Huang's August 17 note reframed Nvidia's earnings-call messaging as a named scarce asset. In the most recent calls, Nvidia on August 26 and Broadcom on September 2 both put Land, Power, and Shell at the center of the AI data-center bottleneck. Huang went further by publishing the concept and attaching it to a concrete site.

His note split Nvidia's customers in two. The largest cloud providers and investment-grade enterprises, he wrote, have the balance sheets, expertise, and long-term contracts to secure Land, Power, and Shell on their own, and that model will remain most of Nvidia's business. Frontier AI labs are the exception: strong demand and growing revenue, but balance sheets and credit profiles that lag their growth.

The physical market, as the article frames it, is not one pipeline but three inventories of buildings: energized now, ready to energize in the next few quarters, and planned for 2027–29. The PORTS-Pike site illustrates the economics Huang described: the Land, Power, and Shell commitment locks a long-lived site, while the compute inside can be upgraded every generation. Whether that upgrade cycle holds, and whether Nvidia extends the arrangement toward the roughly 8 gigawatts of site potential, appears to be the test of how far this model travels.