Nvidia’s Rubin AI Chip Calls for 300GB DRAM, Straining Supply Through 2026
Nvidia’s new Rubin AI platform unveiled at CES 2026 requires 300GB of high-bandwidth DRAM per chip, nearly four times the H100’s 80GB, driving memory demand that has sold out advanced HBM supply through 2026. In February, investors withdrew $743 million from growth ETFs including Nvidia while allocating $15.4 billion into value strategies, hinting at rotation pressure on tech megacaps.
1. Rubin Platform Memory Requirements
At CES 2026, Nvidia introduced its Rubin AI chips demanding approximately 300GB of high-bandwidth DRAM per chip, almost four times the 80GB used by the H100, reflecting the shift toward trillion-parameter architectures and longer context windows in next-generation AI workloads.
2. HBM Supply Outlook
Advanced HBM categories are reportedly sold out through 2026, raising concerns over potential memory bottlenecks and creating pricing power for DRAM suppliers as data center operators race to secure capacity for AI infrastructure.
3. ETF Flow Impact
February fund flows saw $743 million exiting growth ETFs that include Nvidia and $15.4 billion entering value-focused funds, signaling a rotation away from technology megacaps that could weigh on Nvidia’s near-term stock performance if the trend continues.