TSMC Risk

Original Article ↗ Hacker News Discussion ↗

Foundry Capacity, Monopsony, and Alternatives

  • Concern that TSMC’s leading-edge capacity is effectively monopolized by a few giants (Nvidia, Apple, AMD), leaving little room for large, performant RISC‑V or other alternative designs that need multiple iterations.
  • Others argue TSMC actively avoids a monopsony and that smaller players (e.g., RISC‑centric startups) can and do get capacity, though competition is intense.
  • Some point to existing or upcoming fabs in the US, EU, and South Korea (including Intel Ireland) as partial mitigations, but acknowledge they are years behind TSMC’s top nodes.

TSMC-in-Taiwan War Risk and U.S. Fabs

  • One side dismisses the idea that “AI depends on a few Taiwanese buildings” because of TSMC Arizona and other non‑Taiwan capacity; they expect US fabs could ramp within months after a crisis.
  • Others call this wildly optimistic: ~90% of TSMC capacity is still in Taiwan; US fabs lack skilled labor, local supply chains, and cutting-edge process parity, and would need 5+ years to truly substitute.
  • Debate over “scorched earth”: some assume Taiwan or the US would destroy fabs to deny China; others note Taiwanese politicians have publicly rejected this and want to preserve their “golden goose,” even under duress.
  • Consensus that a Taiwan war would not eliminate chips globally but would cause massive, multi‑year disruption in an already oversubscribed industry.

China, AI, and Level of Required Technology

  • Some commenters argue that you don’t need bleeding‑edge nodes for strong AI; older nodes plus more power and hardware are sufficient for a state actor, albeit at high cost and networking pain.
  • Others counter that frontier models trained on tens of thousands of modern GPUs are not realistically reproducible on decade‑old hardware at competitive timelines.
  • Several note China is already progressing on its own silicon and can hire ex‑TSMC talent; “hiring away” is framed as normal competition, though others stress the national‑security context makes it more sensitive.
  • Disagreement on whether China would ever strike TSMC: some think it gains nothing from destroying fabs and wants them intact; others accept the article’s premise that, in an AI‑deterrence scenario, taking TSMC “off the board” could be rational.
  • There is also pushback that cutting‑edge chips are not central to most weapon systems today, which can use legacy nodes, with compound semiconductors (GaN/SiC) being more critical.

Economic and Social Resilience

  • One thread argues that losing TSMC would mean a temporary reversion to ~2018–2022 tech, which is “not the end of the world.”
  • A rebuttal stresses that the issue isn’t just laptops and phones: automotive, logistics, and other critical sectors would be hit, with 10x prices on legacy‑class chips feeding into food and goods inflation.
  • Multiple comments discuss “pain tolerance”: some claim China’s society and leadership can endure far more economic pain than US voters, who quickly punish higher grocery or healthcare costs; others counter that US populations already accept enormous “silent” pain (healthcare, wars, opioids), and that narrative control and political systems, not intrinsic toughness, drive differences.

Big Tech, Capex, and “Wafer Wars”

  • Several note that if hyperscalers truly want more capacity, they can and likely will prepay tens of billions for future wafers, effectively co‑financing new fabs rather than just relying on TSMC’s risk appetite.
  • TSMC’s conservatism on capex is seen as rational: building a $30B+ fab that only comes online in 2029 without long‑term commitments is a huge risk in a fast‑moving market.
  • Some expect a “wafer war,” with prepayments and long‑term contracts for leading‑edge nodes, similar to what is already happening in memory, energy, and raw materials.

Intel, Samsung, and Foundry Competition

  • One camp argues that Intel’s 18A is finally competitive or even ahead and asks why US companies (e.g., Apple) don’t shift volume there, especially given geopolitical risk.
  • Pushback focuses on:
    • Intel’s history of abruptly dropping foundry ambitions.
    • Questionable claims that 18A is ahead; commenters cite density comparisons that favor TSMC and note Intel is still struggling with yields.
    • Lack of spare capacity and news of Intel’s own product shortages.
    • The non‑trivial cost and time (often 1+ years) required to port designs between foundries.
  • TSMC’s culture as a pure‑play foundry and its customer‑centric reliability are contrasted with Intel/Samsung’s potential conflicts of interest.

Export Controls, ASML, and Equipment

  • Discussion notes that although ASML is Dutch, key EUV light‑source technology originated in US‑funded programs, giving the US leverage via export‑control regimes.
  • Some question the legality or reach of such “veto power”; others point to existing US rules (EAR/FDPR‑style) and licensing arrangements as the mechanism.

Broader China–Taiwan–US Geopolitics

  • Some see US “China scare” as ideological or imperial projection; others cite explicit Chinese timelines and reunification rhetoric as real reasons for concern.
  • There is disagreement over whether China can or will invade Taiwan (vs blockade or political pressure) and how far the US and allies would go militarily.
  • A few commenters suggest markets overestimate the likelihood of catastrophic war and underestimate China’s ability to catch up in lithography if denied advanced tools, especially given past Western misjudgments about other powers’ nuclear and military timelines.