TSMC raised its 2026 capital expenditure guidance to $52 billion-$56 billion, CEO C.C. Wei said, after surveying global customer demand.1 The increase spans chipmakers from Taiwan to South Korea to the Netherlands, signaling a worldwide AI hardware buildout, not a regional one.
South Korea's SK Hynix listed on the Nasdaq on July 10, 2026, tapping US capital markets to fund memory expansion for AI servers.2 Memory supply is tightening across the global chain. At Dutch lithography giant ASML, memory chips made up 51% of first-quarter net system sales, up from 42%.3 That shift toward high-bandwidth memory equipment points to more AI accelerator capacity moving through Asian, European and American supply chains over coming quarters.
Wall Street sees the moves as an early signal of broader spending. Analyst Vivek Arya expects global AI data center capex to grow 40% to 50% next year.4 That implies Asian foundries and memory makers are building capacity ahead of, not behind, orders from US cloud giants.
The pattern raises a question spanning three continents: will American hyperscalers Meta, Alphabet and Microsoft raise their own capex guidance after Taiwanese and South Korean suppliers moved first? Foundry capacity commitments have historically preceded hyperscaler spending increases, since Asian chipmakers set multi-year plans based on forward orders from US and global cloud customers.
If Meta, Alphabet and Microsoft report capex revisions exceeding historical trends in coming quarters, it would confirm that Taiwan and South Korea correctly read early demand signals. A miss would suggest the Asian capacity buildout is running ahead of confirmed Western orders.
For now, the signal points toward looser AI hardware availability worldwide, not tighter. Taiwan's TSMC and South Korea's SK Hynix are committing capital first; the response from US hyperscalers has not yet arrived. The episode underscores how tightly linked Asian manufacturing, European equipment makers and American cloud spending have become in the AI era.
