AI summary by TwinMind
The Coming Semiconductor Bust
Asianometry
11m 51s
Semiconductor Industry Volatility & Current Context
- Industry highly volatile with boom-bust cycles occurring at frightening speed
- Normal times still unpredictable: shipments can soar 70% one year, crash 40% next
- Currently working through major chip shortage, questioning when it turns to glut
- Everyone racing to build/expand fabs, potentially sooner than expected
Demand Dynamics & Forecasting Challenges
- End users don't directly buy microchips in large quantities
- Chips purchased by vendors for consumer goods (PCs, smartphones)
- Short product lifecycles - always new products threatening obsolescence
- Missing annual launch cycles can have dire company consequences (AMD mid-2000s example)
- Foundry customers want products made/shipped immediately
- Double booking orders common in uncertain times
- TSMC Chairman Mark Leo cites as major reason shortage not as bad as appears
- Driven by pandemic and China-America trade tensions
- Trade tensions bigger concern with no end in sight
Macro Economic Impact
- Electronics are consumer items - fewer purchases during tough times
- Three recent semiconductor downturns attributed to oil shocks: 1975, 1982, 1991
- Fourth downturn from Asian financial crisis 1998
- March 2020: stock market down 25%, experts predicted economic collapse
- Companies cutting expenses, hunkering down
- Stimulus packages and money printing caused faster-than-anticipated Q3 2020 recovery
Supply-Side Factors More Significant Than Demand
- 2005 study found semiconductor down cycles triggered more by oversupply than under-demand
- Surges in semiconductor inventory levels and fab capacity more significant than demand decreases
- Industry nature/economics force companies to overproduce until collapse
Fab Economics & Moore's/Rock's Laws
- Fabs cost enormous amounts to build
- Moore's Law: transistor count doubles every 2 years
- Rock's Law (Moore's Second Law): semiconductor tool costs double every 4 years
- Tools represent 75% of total fab cost
- 2000: leading edge fab cost $1 billion (2x from 1995)
- 2005: $3 billion
- TSMC N5 fab in Tainan: ~$17 billion
- TSMC N3 process fab in Tainan: estimated $19.5 billion
- Research ongoing to slow Rock's Law while accelerating Moore's
Market Dynamics & Pricing Collapse
- Foundries need billion-dollar fabs to recoup investments
- Fabs lose value almost immediately after construction
- Fixed construction costs high, marginal chip production cost essentially nil
- Incentive to maximize production to amortize fixed costs
- Supply exceeds demand → price collapse → reduced profits
- Foundries still sell every possible chip for survival
- Samsung and TSMC prioritize market share over profitability
- Refuse to mothball capacity to prevent competitor gains
- Massive economies of scale allow survival pricing below competitors
- Results in fierce down cycles, bankruptcies, buyouts, supply rebalancing
Government Subsidies & Policy
US Initiatives
- CHIPS Act and American Foundries Act in development
- Establish R&D fund and 20-40% tax credit for semiconductor production equipment
- Tax credit effects: shorter term (3-5 years, one fab generation)
- R&D fund preferred for longer-term effects, public-private partnerships (like EUV creation)
- Mixed feelings about venture - understands national sovereignty value
- Concern about commercial viability after subsidies end
- Subsidies create market distortions (intentionally)
- Question whether subsidized fabs can stand independently afterward
Global Subsidy Competition
- China: $20 billion semiconductor plan
- US: $50 billion in Biden stimulus plan
- EU: €145 billion for 2nm cutting-edge fab
Current Global Fab Buildout
Major Projects
- TSMC Arizona fab: ongoing for 1 year
- Originally 20,000 wafer capacity, whispers of upscaling
- Final capacity: 25,000-100,000 wafer starts/month
- Samsung Austin fab: major investment buildout, tens of billions planned
- Global Foundries: $1.4 billion expansion, moving up IPO
- SMIC: $2.35 billion Shenzhen government-funded project
- Expected 40,000 wafers/month capacity by 2022
Capacity Analysis
- Hundreds of thousands of wafers coming to market over next 1-2 years
- TSMC Arizona alone at 50,000 wafers/month = theoretical 3.5 billion dies/month
- US Department of Defense buys only $1.9 billion semiconductors/year
- Questions whether demand creation matches capacity expansion
- Doubt about sustained demand from returning students/workers, PC enthusiasts, crypto miners
Market Outlook Concerns
- Chip shortage widely publicized (Wall Street Journal, The Verge)
- Semiconductors currently "super hot"
- Curiosity about boom duration and inevitable bust timing
- Watching buildout with interest given historical boom-bust patterns
