Actionable insights — The AI Crash is Coming (And It's Worse than 2008)

11:17 1. The second-derivative canary — you don't need it to stop, only to slow

1. Stop asking whether the capex cycle ends. Hyper-growth valuations are priced off the rate of change of growth, so the question is only whether growth is decelerating — "you don't need it to stop, you just need it to slow."
2. Find the purest listed proxy for the spend, as far up the supply chain as possible. Semiconductors sit at the front of the AI build-out, so they turn before revenue, before capex guidance, and long before GDP.
3. Prefer an index whose composition is the theme. South Korea's index is ~50% two AI-memory stocks — a cleaner read than a diversified US index where the signal is diluted.
4. Date the peak precisely and treat the first rally after it as the test, not the all-clear: a failed rally into a new low is the confirmation, a reclaimed high is the refutation.
5. Don't call the top on the first sell-off. "I'm not calling a top, but I'm watching the structure" — the position is a monitored hypothesis until the lower low prints.

• The Korean index breaking its post-June low; semis making a lower low while fundamentals still read strong — "growth stock investing 101," the valuation breaks before the earnings do.

5:24 2. Discount the backlog — screen a capex boom for double ordering

1. In any shortage-driven build-out, treat the reported backlog as an upper bound, not a fact. When buyers are allocated a fraction of what they request, the rational response is to inflate the next order — "I need my million, you tell me I can only get 500, so next time I say I need 2 million."
2. Look for direct evidence of the behaviour rather than inferring it: buyer surveys, channel checks, allocation policies. A survey result in the tens of percent is enough to void the backlog as a demand signal.
3. Cross-check with physical evidence of over-ordering — inventory sitting in warehouses for facilities that don't exist yet ("shadow inventory") means units were booked as demand but are producing nothing.
4. Remember the sequencing: double ordering is not an early-cycle phenomenon. "That's how these cycles end" — its appearance dates the cycle, not just the order book.

• Allocation/lead-time normalisation (the moment supply catches up, phantom orders cancel en masse); inventory days rising at the buyers while the supplier's backlog still grows.

7:15 3. MOU-versus-contract — discount announcements to what is legally binding

1. For every headline financing or offtake number, read the instrument. A memorandum of understanding is a statement of intent, not a commitment — "it's not a real commitment until the contracts are written," and "it can just vaporize overnight."
2. Keep a ledger of announced-but-unfunded deals and their age. A large announcement that has not converted into signed, funded contracts after many months is evidence the financing was never there.
3. Fade the price reaction. Markets rip on the headline number; the analytical edge is knowing the number is contingent while the multiple re-rates as though it were cash.
4. Apply the same test outside markets — a geopolitical "deal" announced as an MOU deserves the same discount as a corporate one.

• Definitive-agreement announcements (or the absence of them) 3–6 months after any headline MOU; conversion rate of announced AI financings into funded facilities.

2:05 4. Watch what the players do — behaviour signposts before the data

1. Rank the tells by how much they cost the insider to send. Asking for government financing is expensive to admit, so it is high-information: you request a public backstop only when "the engines of financing were starting to crack."
2. Track insider departures at the theme's flagship private company — people leaving before a liquidity event "suggests that things aren't going so well internally."
3. Log defections among credible establishment voices — when a large allocator publicly says the revenue is circular ("coming from investors, not end customers"), the consensus narrative has begun to break.
4. Read the price of new debt, not the size of it: each successive bond deal from the complex pricing at a wider spread is the credit market voting before the equity market does.
5. Treat these as a sequence, not a single trigger — "salvo number one," then the next, then the next. The trade is sized as the signposts accumulate.

• New-issue spreads on AI-complex bonds vs the prior deal; further senior departures; any move toward explicit public-sector support for AI capex.

13:42 5. The for-sale-vs-sold gap — measure a frozen market, then wait for price

1. Gauge housing with the gap between homes for sale and homes sold, not with the price index. A record-wide gap means the market is frozen: sellers won't cut, buyers won't pay — "there's a buyer's strike."
2. Quantify the strike: run the affordability math to a percentage overvaluation (here ~30%). A frozen market only clears one way — "the only way you clear that is through price."
3. Check who is on each side. If ~60% of listings are older boomers and the buyers are millennials starting families, the mismatch is structural, not cyclical, and the clearing takes longer.
4. Date the cycle from permits, not prices — this rollover's genesis was the 2022 peak in new permits. "By the time people are aware there's a real estate problem," most of the damage is done.
5. Then invert it: falling prices are the green shoots, not the disaster. Once prices readjust, that's the signal to get more bullish on the economy.

• The for-sale/sold gap narrowing (the clearing has begun) and blue-city median prices rolling; construction-employment losses as the transmission into recession.

27:58 6. Equity yield below the risk-free rate — the asset-allocation trigger

1. Compare the yield on stocks against the yield on government bonds. Historically equities pay more than the risk-free rate as compensation for risk; when that inverts, the relationship "usually doesn't last long in history."
2. Use the inversion to derive the forward return rather than to time an entry: it is the arithmetic behind "0% over 10 years" on the index — you are paying up for cash flows you can buy more cheaply in bonds.
3. Identify the catalyst that forces the switch: a growth scare. It makes bonds cheaper "on a cash flow basis" and simultaneously undermines the equity story.
4. Position before the flow, because the reallocation is not gradual — "once the flows begin, it happens quick."

• The equity-yield-vs-Treasury spread re-normalising; the first prints that turn "growth" into "growth scare" in the bond market's two inputs (growth and inflation expectations).

25:51 7. "The solution to high yields is high yields" — fade the extrapolation

1. When a price is rising because of a supply/demand squeeze, model the self-correction rather than extrapolating the trend: high yields attract capital but choke the economy; high commodity prices bring on more supply.
2. Find the historical analogue with the same structure. 2007–08: yields rose on oil-shock inflation (oil doubled to ~$149 on the "insatiable China demand" story) until demand destruction hit and everything rolled over as growth slowed.
3. Separate the two drivers of the bond market — growth and inflation expectations — and ask which one the current move is pricing. A move driven by an energy shock is a growth tax, so it eventually resolves lower, not higher.
4. Take the other side of the consensus extrapolation ("we're going to choke on these high yields") once the recession call is made, and accept that the position can go against you first.

• Demand destruction in the shock commodity; the point where rising yields visibly slow credit growth and hiring rather than just repricing it.

29:41 8. Raise cash, don't short — express the bear case by capacity

1. Separate the direction call from the vehicle. Being right on a bubble does not make shorting it profitable — "don't try to short the market, that's a waste of time, because timing is hard on these things."
2. Match the expression to who you are. Retail ("Joe saver"): raise cash so you can buy cheap assets on the other side. Institutional, willing to take risk: own the long end of the curve.
3. Benchmark the cash level against people whose job is capital preservation, not participation — Buffett at 40% cash "waiting for a fat pitch," Tepper announcing 40% in June, Paul Tudor Jones declining to recommend the S&P at these valuations.
4. Accept the asymmetry of being early: cash costs you the tail of the rally; being fully invested costs you 40–50%. The purpose of cash is optionality on lower prices, not a market-timing bet.

• Reported cash levels at the large discretionary allocators; your own ability to buy after a 40% drawdown as the real test of the position.