HBM/power shortages & DC chokepoints
Key Questions
What is the main bottleneck in AI data center supply?
HBM memory is the primary bottleneck, with SK Hynix dominating through its MR-MUF packaging alliance with NVIDIA and TSMC.
How is Micron performing in the HBM market?
Micron's HBM is sold out through 2026 and offers a 30% power advantage, positioning it as a key alternative supplier.
What role does power grid infrastructure play in AI scaling?
Power grid delays are emerging as a bigger limiter than chips, constraining data center expansion amid rising AI energy demand.
Why are GPU prices inflated above MSRP?
Supply constraints have pushed GPU prices 20-90% above MSRP, creating a secondary market premium for AI accelerators.
How fast is AI energy demand growing?
AI energy demand is expanding 4-5x yearly, with US data centers projected to grow at 29% CAGR through 2032.
What packaging technology gives SK Hynix its edge?
SK Hynix's MR-MUF packaging technology underpins its alliance with NVIDIA and TSMC, supporting high-margin HBM production.
How are cooling and power issues affecting NVIDIA?
Cooling and power constraints are causing some AI engineers to explore alternatives, eroding NVIDIA's perceived dominance in certain deployments.
What is the long-term impact of the memory wall on AI?
The memory wall continues to limit GenAI progress, making vertically stacked HBM a critical solution for overcoming bandwidth limits.
HBM primary bottleneck with SK Hynix dominance via MR-MUF packaging alliance with NVIDIA/TSMC; Micron HBM sold out through 2026 with 30% power advantage. Power grid delays bigger limiter. GPU prices inflated 20-90% above MSRP amid constraints. AI energy demand growing 4-5x yearly; US data centers 29% CAGR through 2032.