Supply: TSMC rev surge NVDA-IRLN/SoftBank Japan + HBM crunch
Key Questions
What is causing the current AI chip supply crisis?
The crisis stems from $725-750 billion in capex and grid strain, with HBM memory sold out through 2026 and TSMC CoWoS at 95% utilization with 18-month lead times. Energy bottlenecks are also leading some data centers to cancel projects.
How long is the memory chip shortage expected to last?
Micron's CEO confirmed a structural shortage beyond 2026, with HBM3 and HBM4 remaining tight. This is driven by AI demand and alliances like SK Hynix with TSMC on MR-MUF technology.
What is the estimated cost of a Nvidia Vera Rubin rack?
A single Vera Rubin rack is estimated at $7.8 million, with over $2 million allocated to memory alone, representing a 435% increase in memory costs. This highlights the growing expense tied to AI infrastructure.
How are GPU prices being affected by supply shortages?
GPU and DRAM shortages are driving prices higher as production squeezes intensify. Nvidia has secured RAM early, giving it an advantage over competitors facing skyrocketing costs.
What role is TSMC playing in Nvidia's supply chain?
TSMC is operating at near-full capacity for advanced packaging, creating an $500 billion bottleneck. Nvidia's relationships help mitigate some risks but limit overall scalability.
Why are some hyperscalers cancelling data center projects?
Energy and grid constraints are causing potential cancellations despite $200 billion in commitments, with only 60% delivery rates. This adds uncertainty to AI expansion timelines.
How does the HBM crunch impact Nvidia's competitors and partners?
Firms like Micron and SK Hynix are seeing tight supply through 2027, with Samsung facing strike risks. This diverts memory to AI workloads and pressures non-AI GPU availability.
What is Nvidia's strategy to address supply constraints?
Nvidia secured key components like RAM well in advance and is leveraging alliances with TSMC and memory makers. However, broader infrastructure limits may cap growth potential.
AI chip crisis amid $725-750B capex/grid strain; HBM sold out '26 (MU: HBM3/4 tight; SK Hynix MR-MUF alliance with TSMC); TSMC CoWoS 95% utilization/18-month lead times. Energy/grid bottlenecks. Micron CEO confirms structural memory shortage beyond 2026. Dell surge highlights liquid cooling/supply chain advantages in AI servers.