← Analysis page  ·  Larry McDonald hub  ·  Research hub

Actionable insights — AI Is Hiding a $600B Credit Crisis

The repeatable analysis behind the calls: not what he owns, but how he gets there — written so each screen can be rerun later on different names.
2026-JUN-16 · Risk Takers (host Alessandro) · Larry McDonald (Bear Traps Report) · ▶ Watch · full analysis · transcript
How to read this page: each insight is a method — the observation that starts it, the steps that turn it into a position or a warning, and the signal to watch when re-running it. The boxed line shows how it played out in this appearance. Timestamps deep-link into the video.

7:23 1. The sector-weight alarm — when one sector eats the index, that is the signal

The repeatable method
  1. Pull the index's sector weights and compare each to its own history. The rule is symmetrical: an extreme weight is a warning at the top and an opportunity at the bottom.
  2. Check the precedents for the ceiling: financials reached almost 30% of the S&P in 2007–08 ("that was a big warning sign"); energy did the same thing in the 2010–14 window. Technology is now ~45% of the index "at the highest valuations ever."
  3. Translate the weight into your own money before arguing about it — "if you have say a million dollars in the stock market, $450,000 of that is in technology." Passive ownership is a concentrated bet you did not choose.
  4. Fund the diversification out of the crowded sector, into the ones whose weight has collapsed — "companies that control assets": oil and gas, metals, industrials, uranium, together only ~14–15% of the index today.
  5. Size the opportunity by the reversion, not the forecast: after a big sector move, the discarded sectors historically go to 30–40% of the index over 5–10 years.
Here: tech ~45% at record valuations vs energy/materials/industrials at 14–15% → the whole hard-asset book (TOU, OIH, copper and coal names) and the call that the S&P is flat for 5–10 years (50:23).
Watch for

19:44 2. Half-weight mean reversion — buy the sector the index defunded

The repeatable method
  1. Find a sector at roughly half its recent historical index weight (healthcare: 16% → 8% of the S&P in five years).
  2. Confirm the shrinkage is flow, not fundamentals: here the money left "because all this money is moving into the technology," while healthcare is simultaneously where the most US jobs are being created.
  3. Inside the sector, screen for companies that own something AI multiplies rather than replaces — a proprietary dataset, an installed base, a regulatory moat.
  4. Source the names from domain specialists, not the index: he ran a private ideas dinner with billionaire medical-technology family offices and took the names out of the room.
  5. Underwrite the upside as margin expansion ("massive increase in profit margins"), which is what a data moat converts into once AI is applied to it.
Here: healthcare at 8% vs 16% → ISRG (surgical data moat "so much more valuable" with AI) and BAX, both named at the medtech family-office dinner (19:21).
Watch for

51:16 3. The index-inclusion trade — front-run the add, never buy it

The repeatable method
  1. Treat an index addition as a forced-buying event, not a quality endorsement: trackers must buy at whatever price the add happens.
  2. Note where in the price cycle the add occurs. Additions happen after a stock has run — the committee is a momentum follower — so the passive buyer's entry is near the high by construction.
  3. The professional trade is to own the candidate before the announcement and sell into the forced bid: "people are figuring out what's going to go into the index. They're buying it ahead of time and then things get dumped into the index."
  4. As an investor, do the opposite of the reflex: an index add on a financially fragile business is a reason to avoid it, and a post-add drawdown is where the real entry work starts.
  5. Check the balance sheet the index committee didn't — some adds are "companies that are really financially a mess."
Here: CRWV added to the Nasdaq while "financially a mess"; LULU added to the S&P at all-time highs and now −60%; SPCX being fed into the S&P/Nasdaq/Russell over the next year.
Watch for

9:45 4. Float-and-unlock arithmetic — count the sellers who have no choice

The repeatable method
  1. For any mega-IPO, start with the float: SpaceX raised ~$75B on ~4% of the company, so 96% of the value is still to come to market.
  2. Read the unlock schedule against the norm — inclusion in indexes normally waits months to a year precisely to shelter passive holders from this. An accelerated schedule is a red flag, not a convenience.
  3. Identify the forced sellers, not the willing ones: a mandated fund whose SpaceX stake ballooned to 25–40% of its book "is not even legal" to leave unrebalanced once shares unlock.
  4. Anchor on the precedent drawdown: Facebook listed "with a big loud roar and then there was like a 50–60% drawdown" as its float expanded.
  5. Compare listing valuation to the company's life-cycle stage. A $30B-in-2019 business listing at $2T has already given its compounding to private holders.
Here: SPCX at ~$2T — twice Berkshire, 14× Facebook's IPO size — on a ~4% float and an unusually aggressive unlock, with Anthropic and OpenAI queued behind it.
Watch for

22:37 5. The off-balance-sheet tell — find the leverage the balance sheet doesn't show

The repeatable method
  1. When a boom is financed rather than funded from cash flow, stop reading the income statement and go looking for the debt that isn't on the balance sheet — special-purpose vehicles, joint ventures, leases, vendor financing. That is the direct Lehman echo he trades off.
  2. Size it: "over $500–600 billion of off-balance-sheet debt that the hyperscalers, the Facebooks of the world, the Microsofts" carry against the data-center build.
  3. Map the assets that debt is secured against and ask what could strand them — here, NIMBY opposition delaying construction and a competing technology (orbital data centers) taking market share from facilities that "already have financing plans."
  4. Look for the stress that has already surfaced elsewhere in the same credit chain: "a big private credit meltdown in the software space already," and CCC-rated junk making higher yields (i.e. lower prices).
  5. Separate the two distinct risks so you can hedge each: AI destroying borrowers (software companies whose revenue AI eats — "in theory, Adobe is at risk") versus AI projects failing to complete (the build-out/NIMBY problem).
Here: $500–600B off balance sheet + a software private-credit meltdown + CCC yields rising → "a Lehman-like credit crisis"; expressed via ADBE as the borrower risk and KKR/XLF as the lender risk.
Watch for

24:32 6. Trade the credit ladder from the outside in

The repeatable method
  1. Rank the financial sector by proximity to the risky lending: private-credit and private-equity firms first, brokers next, money-center banks last.
  2. Measure the spread between the rungs rather than the absolute level — a wide gap tells you the market has already priced the outer rung and not yet the inner one.
  3. Position where the repricing hasn't happened, on the thesis that stress migrates inward: "you're already seeing cracks in the foundation, but that's going to crack all the way up to the big banks."
  4. Express it at the sector level rather than name-picking a bank — options on the sector ETF, sized as a defined-risk position.
  5. Time it to the catalyst rather than holding open-ended: he dates the credit event to Q3 (September–October), coinciding with the inflation bounce.
Here: KKR −42% off the highs vs JPM +23% over ~18 months → clients "short the financials, either short puts on the XLF" (24:59).
Watch for

4:05 7. Supply-shock sequencing — the reopening is when the inflation starts

The repeatable method
  1. Don't fade a supply shock on the headline that ends it. "When you close the strait for 100 days, there's a price to pay for that" — the price arrives with a lag, after the ships resume.
  2. Trace the shock down the input chain past the obvious commodity: fertilizer and food, then semiconductors via tungsten and critical minerals that "are essentially being hoarded now" (the Japanese supply warning gives a dated checkpoint).
  3. Budget the digestion window explicitly — "probably the next three, four weeks, five weeks, six weeks in the market where you're going to have to digest that inflation bounce."
  4. Stack the seasonal amplifiers on top (summer driving season, refiner buying, the World Cup) to judge whether the bounce compounds or fades.
  5. Then trade the second-order effect, which is the rotation: an inflation shock "should trigger a big rotation out of technology into oil and gas, into metals, into companies that control hard assets" — the 2021→2022 Nasdaq-100 round trip ($20T → $12T) as the template against today's ~$41T.
Here: Hormuz reopened the morning of the interview after 100 days → inflation bounce into Q3, which also "is going to accelerate the credit crisis" (27:52).
Watch for

31:11 8. The interest-expense ceiling — price what the Fed cannot do

The repeatable method
  1. Before believing a hiking cycle, price the government's carry: interest on the debt is already $1.1 trillion, versus ~$300 billion going into the 2021–22 bounce, with ~$10 trillion rolling over the next 12 months.
  2. Conclude what the policy makers can't say: "the Fed can't admit it, but they really don't have much room to hike at all."
  3. Expect the alternative tool — financial repression. Only two exits from a $40T hole: "a debt jubilee… or you inflate your way out," which means "you massage interest rates below the rate of inflation."
  4. Look for the plumbing that implements it: banks pushed out of Fed reserves into Treasuries ("the Fed and the US Treasury have a gun… pointing at the banks"), roughly $1 trillion of forced buying spread over four years.
  5. Watch for the target being redefined rather than missed — a switch to trimmed-mean inflation is "the classic thing governments do when they need to move the goal posts… a way of saying that we're walking away from the 2% inflation target" (37:37).
  6. Position accordingly: negative real rates make "companies that control assets much more valuable."
Here: $1.1T interest expense + ~$300B out of JPM's Fed reserves into Treasuries → the whole hard-asset tilt, and the reason he expects the inflation bounce to go unfought.
Watch for

40:04 9. The Bitcoin/gold ratio — rotate between hard assets on a ratio, not a price

The repeatable method
  1. Hold the two "escape hatch" assets as one sleeve and rebalance between them on their ratio, so you never have to forecast either one's price.
  2. The trigger: "we track the Bitcoin gold ratio and when it gets down into the low teens, that's where you want to be selling some gold, buying some Bitcoin." The ratio had fallen from ~40.
  3. Act in scaled trade alerts rather than a wholesale switch — trim the winner, add to the laggard.
  4. Cross-check with a liquidity read: Bitcoin "is a really good measure of liquidity." When speculative appetite is obvious elsewhere (quantum names up 6–10% on the day) but Bitcoin isn't participating, ask what's absorbing the money — here, forced selling to fund the record IPO calendar.
  5. Keep the structural leg separate from the tactical one: $17T of fiscal/monetary response plus $5–6T of capex "should be good for Bitcoin over time" regardless of the ratio trade.
Here: the ratio in the low teens → Bear Traps "recently bought some Bitcoin, for the first time" (IBIT as the retail expression), funded conceptually by trimming gold — while gold's own target is 4,600–4,700 then 6,500 next year (47:50).
Watch for

53:41 10. The "where does the revenue come from?" test

The repeatable method
  1. Take the banker/sell-side revenue projection at face value for a moment, then ask the accounting question nobody asks: whose budget does it come out of? "It's not popcorn that you just pop."
  2. Name the payers explicitly — for SpaceX going from ~$60B to $1.3–1.4T of revenue over five years, the answer is the Mag 7 plus Oracle: "this revenue's going to come from Microsoft and Meta."
  3. Check whether those payers have the money uncommitted. Here they don't: $4.5–5T of AI capex is already promised and "their cash flow is already getting raided by data centers."
  4. Conclude symmetrically — if the projection is real, it is a short case for the payers; if it isn't, it's a short case for the projection. Either way the consensus that both can win is wrong.
  5. Sanity-check the doubling math against the listing valuation (a $2T listing must reach $4T to double) before assuming index-level returns.
Here: SPCX's $1.4T revenue projection → bearish MSFT/META/ORCL free cash flow and the flat-index call, with MAGS −2–3% since October as the early confirmation.
Watch for

Methods distilled from the public YouTube video (transcript in transcript.txt) for personal study. Not investment advice. © Risk Takers for source material.