3 Insights on Semis
- Complexity and lack of resilience are two different vectors. Semis exhibit both whereas oil only exhibits one. On semis:
- Geography - high concentration of fabs in Taiwan/SK, wafer design in Japan, EUV in Netherlands, raw materials in China
- Capital/Time/Scale - extremely high capital expenditure ($10-20B for state-of-the-art fab), years not only to build new but to retool existing, scaling more chip output usually requires new fabs.
- Human Capital/Expertise - limited number of experts in say, lithography
- Product - high technical lockin towards proprietary systems, different interface standards which makes it very difficult to change suppliers
Compare oil which has a complex, multi-stage supply chain but much more flexibility (Substitutability between different sources of oil/blending, alternate transportation routes between geographical stages, multiple producers across different political orientations, ability to on-shore refining capability)
AI: “In semiconductors, the serial dependency is extreme: virtually every electronic product in the world depends on a handful of advanced semiconductor fabs and equipment suppliers. If one link in this chain fails (say a critical chemical isn’t delivered, or a lithography machine breaks without replacement), there is often no short-term workaround. This differs from, for example, the oil supply chain, where a disruption at one refinery can often be compensated by ramping others or drawing down reserves. Oil flows through a more flexible network (multiple wells, pipelines, alternative shipping routes), whereas chips flow through specific narrow chokepoints.”
- Resilience is a public good. Foundries are able to produce high margins and pass on shocks to customers in the form of higher prices, therefore there is less incentive for contingency planning, shock adjustment. Because demand can be so cyclical, an actor bears a lot of risk for ensuring resilience (like building extra capacity means risk of losses in a downturn), so they won’t do it unless compensated. It is only through government where this happens.
AI: If a company decided to build an “extra” fab purely for backup, that would drag down its financials and its stock price, unless there’s subsidy or guaranteed revenue. Thus, we saw no company built a duplicate of TSMC’s capacity in a safer location until governments (US, EU) started literally paying for it. The framework explicitly asks “Who underinvests in redundancy and why?” – in semis, everyone underinvested in redundancy. Why: because redundancy costs are immediate and benefits are hypothetical and shared by all. TSMC might say, why should we build capacity in the US at our own cost to benefit U.S. supply security? They rightly waited until subsidies made it ROI-neutral.
- Follow up on the previous → corrections towards resilience have typically taken the form of government subsidies rather than market-driven mechanisms. We currently see CHIPS act, proposals by DoD to maintain a strategic reserve of radition-hardened chips, but:
- No “cartel” association of suppliers to coordinate prices or supplier diversification
- No financial hedging instruments to facilitate swaps, futures, etc. for hedging against price volatility
- No demand index to provide greater insight/preparation for cyclical changes
- No companies commercializing a “chips reserve,” holding chips for a warehousing fee and delivering them to firms with an “insurance contract” of sorts during shortages
These could be ideas for “resilience as a service”
Dustin’s:
- Value chain over supply chain
- Lagging in financial and logistical infrastructure (biggest difference from oil)
- Automotive industry is very strong candidate (hit very hard)
Missing and underdeveloped layers (financial products!) commodity trading risk hedging, insurance mechanisms strategic buffer capital formation structure market intelligence geopolitical risk mitigation state market interface
Counterfactual if oil were prevented today, how would industry be designed?
Ideas: Commodity exchange and hedging platform Independent semiconductor trading house insurance and risk solutions pay premiums in, pay out when it’s called. Get pricing right (high enough to cover payout, low enough that people will buy) syndication platform - building fabs (Hadrian comp) master lease - lease from a fininite amount of time everything associated in the building master lease from a fab real supply chain intelligence and coordination platform (suggest alternatives, marketplace, Waze of supply chain) All of these rely on the same data asset!! HW model the likelihood/utility of future fluctuations related to the value chain likelihood that shit goes wrong, how expensive it is