Supply, not demand, now rations every layer of the AI silicon stack
The foundry calls the demand-supply gap 'quite big' while committing 2026-28 capex significantly higher, the lithography monopolist's planned 2027-28 output falls short of customers already probing 2030 slots, and Nvidia expects supply bottlenecks through fiscal 2028 with memory the tightest input — physical capacity, not buyers, sets the pace of the ramp.
The same conclusion keeps arriving from across the workspace's research — 1 topics independently instantiate this theme. Filter the evidence by where it came from:
Even the vendor at the center of the stack can fill only ~70% of orders, confirming that physical capacity — power, buildings, components — not buyers, sets the pace of the ramp.
The foundry layer's sixth record quarter and 'significantly higher' 2026–28 capex commit its own balance sheet years out because physical capacity, not demand, rations the buildout.
Customers are booking lithography slots into 2029–2030, years beyond planned output — demand is being committed against physical capacity that does not exist and cannot arrive on schedule.
The '25% Rubin cut' is stale re-reporting; Rubin ships on schedule while memory stays the tightest input and the binding constraint through fiscal 2028 — supply rationing revenue, not demand destruction.