Nvidia weighs shipping Rubin with less memory; AMD confident on supply

The memory shortage facing the AI chip industry is forcing a reckoning between the two dominant players. Nvidia, which controls more than 95% of the data center GPU market, had publicly projected confidence just weeks before—hardware SVP Andrew Bell said in mid-July that the company was "in front of the memory problem." Yet testing of reduced-memory Rubin variants suggests either that confidence was misplaced or that Nvidia is preemptively adapting to a constrained supply reality. The versions under test (192GB to 256GB, versus the 1 terabyte originally announced) would represent a significant cut.

AMD's rival positioning is notably different: the company claims it has already locked up high-bandwidth memory allocations through direct relationships with all three major suppliers—Samsung Electronics, SK Hynix, and Micron Technology—and has secured major customer commitments from Microsoft, Meta, OpenAI, and Oracle. AMD's data center revenue was up 57% year over year as of Q1 2026 and now constitutes the majority of the company's total revenue, signaling real momentum. Whether AMD's supply confidence will hold and whether Helios can genuinely compete on workloads where Nvidia dominates remain open questions, but the divergence in their public posture reflects a genuine shift in competitive leverage around memory access.