HBM crunch & supply/geopolitics/power
Key Questions
What is TSMC's current market share in AI chip manufacturing?
TSMC holds 72.3% share of advanced AI chip production, reporting $35.5B in Q1 2026 revenue amid surging demand.
How long are HBM shortages expected to last?
HBM supply constraints are projected to continue through 2027 due to high AI infrastructure demand and limited production capacity.
What custom ASICs are gaining traction in the AI market?
Broadcom and others face backlogs for custom ASICs, with Google TPUs and Meta MTIA expanding as alternatives to Nvidia GPUs.
What is ASML's High-NA EUV rollout status?
ASML is advancing High-NA EUV machines, with first AI chips expected soon to strengthen leadership in advanced semiconductor lithography.
Why might Anthropic shift to Microsoft AI chips?
Anthropic is considering Microsoft Maia chips for inference to reduce reliance on Nvidia amid computing infrastructure competition.
How do DDR5 and HBM compare for AI servers?
HBM offers higher bandwidth for AI accelerators while DDR5 provides cost-effective capacity; both face bottlenecks in powering next-gen servers.
What advanced packaging developments support AI chips?
Whitepapers and research highlight silicon design and advanced packaging techniques as critical for scaling AI chip performance and efficiency.
What is the outlook for custom AI ASIC adoption?
Custom ASICs from Broadcom, Google, and Meta are eroding Nvidia's ~70% market share as companies seek tailored, cost-efficient alternatives.
TSMC 72.3% share, $35.5B Q1 2026; shortages to 2027; custom ASICs/Broadcom backlog; ASML High-NA rollout. New: Anthropic Microsoft Maia inference shift; DDR5 vs HBM bottlenecks; advanced packaging whitepaper; Marvell ASIC views.