BofA says $NVDA Rubin Ultra won't cheap out on memory specs — HBM demand thesis intact.
Key points:
• HBM4e/HBM5 roadmap still targeting 500GB-1TB per GPU
• Rubin Ultra performance tanks hard with lower memory configs
• 1TB likely mandatory for physical AI workloads
• HBM4 pricing up 10-20% for 2H26 deliveries
If this plays out, HBM growth stays fueled by both volume AND pricing power.
Bullish for $MU, Samsung, $SKHY — the picks and shovels play on AI infra just got stronger.
Key points:
• HBM4e/HBM5 roadmap still targeting 500GB-1TB per GPU
• Rubin Ultra performance tanks hard with lower memory configs
• 1TB likely mandatory for physical AI workloads
• HBM4 pricing up 10-20% for 2H26 deliveries
If this plays out, HBM growth stays fueled by both volume AND pricing power.
Bullish for $MU, Samsung, $SKHY — the picks and shovels play on AI infra just got stronger.