NVIDIA CorporationNvidia may cut HBM on Rubin Ultra due to memory shortage, potentially reducing performance and competitiveness.
Nvidia is reportedly testing versions of its next-generation Rubin Ultra GPU with significantly less high-bandwidth memory than originally promised, with some configurations as low as 192GB to 256GB versus the 1 terabyte Jensen Huang initially announced. The move comes amid a severe industry-wide memory shortage, though Nvidia still controls more than 95% of the data center GPU market and has struck a $500 billion partnership with SK Hynix's parent company to co-develop future memory technology. Rival AMD says it has already secured the HBM it needs through ties with Samsung Electronics, SK Hynix, and Micron Technology, and its Helios AI system ships to customers including Microsoft, Meta, OpenAI, and Oracle later this year. AMD's data center revenue grew 57% year over year as of Q1 2026 and now makes up the majority of its total revenue, but Helios reportedly costs $5 million to $5.5 million per system versus $3.5 million to $4 million for Nvidia's Vera Rubin. Nvidia was the most widely held chip stock among hedge funds in Q1 2026 with 275 holders, compared to 134 for AMD.
NVIDIA CorporationNvidia may cut HBM on Rubin Ultra due to memory shortage, potentially reducing performance and competitiveness.
Advanced Micro Devices IncAMD secured HBM supply and its Helios AI system ships to major customers, with data center revenue growing 57%.
Samsung Electronics Co LtdSamsung supplies HBM to AMD, benefiting from memory shortage and AMD's secured supply.
Apple Inc.
Meta Platforms Inc.
Microsoft Corporation
Oracle Corporation
SK Hynix IncSK Hynix's parent partners with Nvidia on $500B memory co-development, and memory shortage boosts demand.
Micron Technology IncMemory shortage and Nvidia's reduced HBM usage may increase demand for Micron's HBM, and AMD's ties with Micron secure supply.