Nvidia has committed to $279 billion in multi-year memory purchases, primarily high-bandwidth memory (HBM), a 135% quarter-over-quarter increase that underscores the company’s shift from a technology firm to a strategic commodity buyer. The scale of the arrangement—reported alongside Nvidia’s FY2027 second-quarter earnings on August 26—positions memory access as a critical bottleneck for the entire semiconductor industry.
The surge in demand has already strained supply chains and pressured gross margins. Nvidia guided fourth-quarter gross margins to 71–72%, down from prior levels due to elevated memory costs, though the company expects recovery to 72–73% by fiscal 2028. Competitors are responding to the shift. AMD shares fell 1.87% to $471.95, while Micron Technology declined 2.78% to $912.34 after a 674% one-year rally. In contrast, SK Hynix gained 2.49% in Seoul and Samsung Electronics rose 2.49%, reflecting divergent market reactions to the new competitive dynamics.
Nvidia’s strategy extends beyond securing supply. Reports indicate the company is evaluating downgrades from 12-Hi to 8-Hi HBM4 stacks for its Rubin Ultra architecture to optimize die utilization amid a constrained memory pool. This move aims to increase accelerator shipments despite tighter substrate and DRAM availability. The Jefferies Semiconductor Conference highlighted Nvidia’s broader push to lock in substrates, DRAM, and other critical inputs, signaling that supply constraints now define competitive positioning across the compute sector.
Investor positioning has also shifted. ARK Investment Management sold 37,977 AMD shares on August 26, while increasing exposure to Broadcom and Cerebras, reflecting a preference for architectural alternatives such as custom ASICs and specialized accelerators over general-purpose GPU competitors in a memory-constrained environment.
