What if a TSMC packaging bottleneck caps AI chip output?
A CoWoS packaging bottleneck is a supply shock, not a demand shock: it caps accelerator output and lifts spot GPU prices, so TSMC (the choke point) falls on output risk while the scarcity is bullish for incumbent pricing. This is the recurring 2023-24 CoWoS-constraint story that repeatedly gated Nvidia shipments. Forward angle: roots correctly show ai_capex positive — demand is intact; the move is about who captures the scarcity rent (TSMC/SK Hynix) versus who is volume-rationed.
how we built this number — every step
The class rate is measured from our dated, sourced event library (decade-normalized Poisson — the full table is public at base_rates.json). The variant’s share within its class is the analyst’s editorial call, published so you can audit it. A wider range means thinner precedent. Full recipe: methodology · scored at Reality Check.
The butterfly cascade
How this trigger trickles across markets, left → right — the root shock, its first‑order moves, then the ripple effects. Drag any node; tap a market for its real price history.
Resolution timeline — how this probability is moving
Our model's odds (gold) over time vs the crowd's (Polymarket, blue), from the past toward the 0–6 months horizon. Each dot is a real macro event that nudged the probability — green pushed it up, red pushed it down. Tap a dot for the source. The gold path is an illustrative reconstruction anchored to today's estimate — real dated events, not a live re-estimate history.
What it would mean
If this plays out, it is a mixed shock. A TSMC advanced-packaging bottleneck caps accelerator output, lifting lead times and spot GPU prices. The trigger decomposes into signed root‑shocks — AI capex ▲ · Semiconductor supply risk ▲ — which propagate through our causal graph to the markets below.
If it happens — the markets it would move
Biggest moves first. Projected moves are cascade-model priors; hist A–B% = what comparable past events actually did (measured abnormal returns), and model prior · unmeasured marks markets with no analogue backing yet. Tap any market for its price history.
| Market | Class | Projected move | |
|---|---|---|---|
| 1 | TSMC TSMon Hyperliquid 📈 chart | Equity | ▼ -1.3% hist -0.84–-0.4% |
| 2 | ASML ASMLon Hyperliquid 📈 chart | Equity | ▼ -0.6% hist -1.19–+0.14% |
| 3 | Nvidia NVDAon Hyperliquid 📈 chart | Equity | ▼ -0.3% hist -0.62–+0.76% |
| 4 | Semiconductors SMHon Hyperliquid 📈 chart | Equity | ▼ -0.3% hist -0.34–+0.23% |
| 5 | AMD AMDon Hyperliquid 📈 chart | Equity | ▼ -0.2% hist -0.16–-0.06% |
| 6 | Marvell MRVLon Hyperliquid 📈 chart | Equity | ▼ -0.2% hist -1.68–+0.55% |
| 7 | Qualcomm QCOMon Hyperliquid 📈 chart | Equity | ▼ -0.2% hist -3.8–+1.2% |
Probable recommendation
Historical precedent — what analogous events actually did
Across 40 analogous events (overlap‑weighted), as abnormal returns — market beta stripped, so it's the event's own effect, not the market backdrop. Shown at 20 days (persistent) and 5 days (immediate); ↺ fades = the two horizons disagree. Confidence = consistency × sample × significance.
| Asset | History says | Abnormal (20d · 5d) | Hit | n | Confidence | vs cascade |
|---|---|---|---|---|---|---|
| US dollar DXY | SHORT | -0.2% · 5d +-0.0% | 64% | 40 | 0.22 | · |
| Gold XAU | LONG | +1.1% · 5d -0.5% ↺ fades | 64% | 40 | 0.21 | · |
| AMD AMD | SHORT | -0.0% · 5d -2.7% | 61% | 40 | 0.17 | ✓ matches cascade |
| QCOM QCOM | SHORT | -3.4% · 5d -3.3% | 61% | 40 | 0.17 | ✓ matches cascade |
| ASML ASML | SHORT | -0.8% · 5d -2.2% | 57% | 40 | 0.11 | ✓ matches cascade |
| 10y yield DGS10 | LONG | +3bp · 5d +1bp | 57% | 40 | 0.11 | · |
| MRVL MRVL | SHORT | -1.5% · 5d -2.1% | 57% | 40 | 0.10 | ✓ matches cascade |
| High-yield credit HYG | SHORT | -0.1% · 5d +0.2% ↺ fades | 56% | 39 | 0.09 | · |
| TSM TSM | SHORT | -0.1% · 5d -1.3% | 55% | 40 | 0.08 | ✓ matches cascade |
| SMH SMH | LONG | +0.4% · 5d -0.9% ↺ fades | 52% | 40 | 0.04 | ⚠ differs |
| Volatility VIX | LONG | +2.5% · 5d -0.0% ↺ fades | 52% | 40 | 0.04 | · |
| NVDA NVDA | LONG | +1.0% · 5d -3.1% ↺ fades | 39% | 40 | 0.00 | ⚠ differs |
| Bitcoin BTC | LONG | +2.8% · 5d -1.4% ↺ fades | 45% | 36 | 0.00 | · |
Why this probability
CoWoS packaging has repeatedly bottlenecked; recurring pattern, high base rate even short-window. A base‑rate‑anchored prior, continuously scored against what actually happens — not a forecast.