Nvidia reportedly considering reducing HBM memory specifications on Rubin Ultra GPUs due to supply constraints.

Nvidia Corp. (NVDA) is reportedly testing multiple versions of its next-generation Rubin Ultra graphics processing unit (GPU), including designs with less high-bandwidth memory (HBM) than originally planned, as it evaluates ways to navigate potential supply constraints. The move reflects broader pressure on the semiconductor industry as soaring AI demand has strained the availability of advanced memory chips.

The development has also sparked debate among Stocktwits retail traders over whether memory supplier Micron Technology (MU) could emerge as a key beneficiary if HBM shortages persist.

Over the past few weeks, Nvidia has been testing at least three versions of the Rubin Ultra GPU, some of which feature less memory than the company initially announced, according to The Information, citing people with knowledge of the trial. The company is considering the lower-memory designs in part because it may not be able to secure enough advanced memory chips to support its original configuration.

While lower memory could affect performance, AI companies running large models on the reduced-memory versions may need to deploy more GPUs than they otherwise would. CEO Jensen Huang unveiled the Rubin Ultra at Nvidia's annual developer conference in 2025, saying each GPU would include 1TB of HBM4E memory spread across 16 memory stacks.

Nvidia's reported move shows how the AI boom has stretched the chip industry's ability to meet surging data center demand. The resulting shortage of high-bandwidth memory chips has pushed up prices, increasing costs across the technology industry and forcing companies to spend more on hardware. The situation is notable because demand for Nvidia's AI GPUs has itself contributed to the memory shortage, and the company is now exploring alternative memory configurations. The Rubin Ultra samples currently under evaluation would represent a reduction in memory compared with Nvidia's Rubin chip, which is already in mass production and being rolled out to customers.

On Stocktwits, retail traders largely viewed the report as evidence that AI infrastructure demand is running up against real-world supply constraints, while debating the implications for Nvidia and memory chip suppliers. Micron entered the discussion as one of Nvidia's suppliers of high-bandwidth memory used in the chipmaker's AI GPUs. One retail trader noted that Nvidia's reported testing of lower-memory Rubin Ultra GPUs suggests "supply constraints are real" even for the AI chip leader, adding that less memory per chip could require more GPUs to run large AI models. Another trader argued that while the memory reduction could allow Nvidia to produce more GPUs, "MU is the real beneficiary here," as AI infrastructure buildouts run into physical supply constraints.

NVDA stock has gained nearly 17% year-to-date.