Actionable insights — The $1.5 Trillion of Hidden Debt Fueling the AI Boom

5:31 1. Read the purchase commitments out of the 10-Q yourself

1. Accept that headline capex is the spent number, not the committed number. The committed number lives in the commitments-and-contingencies discussion of the 10-Q, not in the cash-flow statement or the earnings deck.
2. Open the quarterly filing and search the text for the purchase-commitment language directly — this is a text search, not a line item, because there is no standard tag for it.
3. Widen the definition to everything the company has contracted to buy: chips, memory, networking and cooling equipment, and — increasingly the big one — guaranteed power offtake.
4. Where the company breaks the total into short-term and long-term, capture the split. Where it doesn't, note that too: the absence is itself a data point.
5. Repeat for every company in the peer set and put them side by side. Score the disclosure quality as you go — which ones state a number, which ones only say they have "material upcoming payments."
6. Redo it every quarter and diff. The quarter-on-quarter change is the finding; the level alone tells you much less.

• The short-term slice as a fraction of the total — if a large share is due inside twelve months, published capex estimates for that name are too low.
• Wording that changes between quarters at the same company: it destroys comparability, and it is usually deliberate.
• A peer that discloses no number at all while its competitors do.

3:02 2. The lease-accounting tell — started leases vs footnote leases

1. Learn the one accounting rule that governs the disclosure: a lease only hits the balance sheet once it has commenced. At commencement a right-of-use asset appears with the matching liability opposite.
2. Therefore split every lease obligation into two buckets: commenced (visible as payment obligations on the balance sheet, though never labelled debt) and not yet commenced (visible only in a footnote).
3. Treat the not-yet-commenced bucket as the interesting one. It is the pipeline of liabilities that will land on the balance sheet over the next few years, and it is the part no screen will show you.
4. Apply the duck test to whatever you find: a fixed, multi-year, guaranteed payment stream you cannot exit "walks, talks and quacks a bit like debt" whatever the label.
5. Check the strength of the guarantee before deciding how debt-like it is — read whether the lessee can walk away, and at what cost.
6. Don't reinvent someone else's work: sell-side research desks sometimes do the filing sweep across the whole group. Take their aggregate and add your own orthogonal cut (he took Goldman's lease numbers and built the purchase-commitment tally beside them).

• Growth in the not-yet-commenced footnote bucket — it is the leading indicator of future balance-sheet liabilities.
• New JV structures where the operating company takes a minority equity stake and a long lease guarantee: that pairing is the signature.
• The mirror image on the counterparty's side: a developer's or neocloud's "contracted backlog" is often the same obligation counted as revenue.

8:52 3. The boom diagnostic — is it equity-financed or debt-financed?

1. Before judging whether the technology is real, ask a different question: how is the build-out being funded? This is the variable that determines what the bust looks like, and it is separable from whether the technology works.
2. Classify. Equity-financed booms (dot-com): the losses land on people who chose to take equity risk. "The stock market dropped 50% peak to trough; economically it was a nothing burger."
3. Debt-financed booms (canals, railways, 2008 housing): the losses land on levered lenders, so they transmit into the banking and credit system. "Even when the underlying premise comes true and AI transforms the world, quite often they end in tears."
4. Do not let a correct technology thesis substitute for the funding answer. The railways were genuinely transformative and bankrupted their financiers; the infrastructure survived, the capital structure did not.
5. Note the misdirection: the visible excitement (IPOs, private marks) may sit on the equity side while the actual volume sits in debt. Follow the volume.
6. Scale the comparison honestly. Size the boom against the economy of its day, not in nominal dollars.

• Record issuance years in a sector's investment-grade bond calendar — he notes last year's hyperscaler record was already broken this year.
• The debt/equity split inside big announced financings (the host's estimate on the NVIDIA MOU was ~80/20).
• Migration of funding away from public bonds toward leases, private credit and off-balance-sheet vehicles: same debt, less price discovery.

16:41 4. The safe-asset crisis detector

1. Stop scanning for risky assets. "It's not when you invest in something that's risky and it blows up in your face. That's fine. That's just risk and reward." Junk defaulting is not a crisis.
2. Scan instead for assets currently treated as money — things accepted as collateral, held to satisfy capital or mandate rules, or used as the foundation of a business model rather than as a position.
3. Walk the historical collateral ladder to see where you are: government bonds → high-grade corporates (IBM, Microsoft) → asset-backed securities in the 2000s. Each step was "super solid" until it wasn't.
4. Ask the specific killer question: does anyone's business model or entire strategy depend on this asset staying safe? If yes, a repricing becomes insolvency rather than a loss.
5. Use time since the last crisis as the accelerant. The mechanism is that "it's been so long since a previous crisis that you treat it as money" — long calm is a precondition, not a comfort.
6. Test tradability, not just credit. Many 2008 instruments were not worthless; they simply could not be sold, which is the same thing to a levered holder.

• Any new asset class being accepted as loan collateral, or written into insurance capital rules, before it has a default history.
• Reported volatility that is low because nothing is marked, rather than because nothing is moving.
• Regulatory or mandate language that requires holding a category — forced buyers are what turn a repricing into a cascade.

9:42 5. Discount the press release; wait for the structure

1. Separate an announcement from a commitment. A memorandum of understanding to lend or invest is neither signed paper nor drawn money. "It's very easy to push out press releases saying we're going to lend or invest X or Y or Z into this or that."
2. Adjust for the incentive to announce: "clearly there is a lot of heat in this area now and everybody wants to be seen to be leaning into it." Late-cycle, announcement volume decouples from deployment.
3. Ask who bears the risk in the final structure, not the headline. Sophisticated managers "are going to be very careful about how they protect their own balance sheets but also the balance sheets of their investors" — which usually means less risk taken than the headline number implies.
4. Wait for the terms: seniority, collateral, recourse, draw conditions. Until those exist, the number is marketing.
5. Track the gap between announced and deployed across a cycle; the widening of that gap is itself a late-cycle signal.

• The follow-up filings — actual facility agreements, drawn balances, or the absence of either six months later.
• The word "up to" in a financing headline.
• Circular arrangements where the announcing parties are each other's suppliers and customers.

11:26 6. "Democratization" and semi-liquidity as sell-side tells

1. Treat the word democratization as a marketing signal, not a social one. "It's usually a code word for jamming something down the necks of retail investors that are not really quite ready to digest."
2. Watch the sequence: an asset class matures institutionally, returns compress, and only then is retail access "opened up." Retail arrival is a late-cycle event by construction.
3. Test the liquidity promise against the underlying asset. If the product touts liquidity as a selling point while holding illiquid assets, the promise is a mismatch waiting to be tested — "do not do it even in a semi-liquid format."
4. Demand a lockup that matches the asset's tenor: "if you invest in loans with a 5-year tenor, then you should be locked up for 5 years." Anything shorter transfers risk from early redeemers to whoever stays.
5. Assume behaviour, not sophistication: "people pull their money out when they're afraid," and wealth does not change that.
6. Reject leverage stacked on leverage — a vehicle lending to highly levered companies should ideally carry none itself; term debt matched to asset maturity is the tolerable exception.

• Launches of retail-wrapper versions of an institutional strategy; interval funds, tender-offer funds, "evergreen" vehicles.
• Redemption caps and gate language in the prospectus — the cap tells you what the manager expects.
• Investors borrowing to buy a levered vehicle ("I hope nobody's borrowing money from the brokerage and yoloing into BDCs").

26:54 7. PIK as deferred pain — separate the legitimate use from the mask

1. Know what payment-in-kind is: instead of paying cash interest, the borrower adds the interest to the principal. The loan keeps performing on paper while no cash moves.
2. Do not treat it as automatically bad. "Payment in kind is a completely viable and acceptable and important tool in many cases… the right one to use for companies growing very quickly but don't want to send cash out the door right then."
3. Distinguish the two uses by asking why it was granted: growth-stage cash preservation agreed at origination, versus a modification granted mid-life to a struggling borrower. The second is a default that hasn't been recorded.
4. Track PIK as a share of a lender's income, and its trend. A rising PIK share is a default rate being smoothed.
5. Adjust the headline default rate accordingly before comparing it to history — "the defaults have been kept probably artificially low."
6. Layer in the refinancing effect: heavy inflows into an asset class refinance maturing loans and suppress defaults independently of fundamentals. When inflows slow, the suppression stops.

• PIK income as a percentage of total investment income in BDC filings, quarter over quarter.
• Amendment activity: covenant waivers and cash-to-PIK conversions.
• Net flows into the asset class turning negative — that is when the refinancing bid disappears.

27:16 8. The recovery-rate reality check for asset-light borrowers

1. Separate the two inputs to a credit loss estimate: probability of default and loss given default. The market argues about the first and inherits the second from history.
2. Interrogate the recovery assumption directly. High-yield convention is 70–80 cents on the dollar, "depends on where you are in the cap structure of course" — but that number was earned by a borrower population that owned things.
3. Ask the physical question about each borrower: what does the creditor actually seize? "Let's say if they're in the software industry, where there are no plants and factories and roads and trucks… if the company isn't good, it blows up and there's nothing there for you as a creditor."
4. Reweight the portfolio's blended recovery for the mix shift toward asset-light, services and software borrowers. Wigglesworth's word for the current assumptions is "fantastical."
5. Apply the same test to any new collateral type before assuming a recovery: for compute, the asset has a half-life and, in orbit, cannot even be serviced.
6. Then re-run the loss math. A default rate you can live with at 75% recovery is a different number at 20%.

• Sector mix in a credit portfolio — the software/services share versus the asset-heavy share.
• Disclosed recovery assumptions in a manager's loss modelling, and whether they are sourced from a pre-2015 borrower population.
• First actual restructurings in the current cycle: realised recoveries are the only honest calibration.

1:00:33 9. Watch the private-label rating / PE-insurer nexus

1. Start from why the rating exists at all: insurers must hold a set share of assets in investment grade, so the letter is not information, it is permission to hold. That creates demand for a favourable letter independent of credit quality.
2. Distinguish the big three from "private label" raters — smaller firms rating private credit loans that the big agencies never see.
3. Map the ownership chain for each link: who owns the insurer, who owns the rater, who originated the loan. Wigglesworth's concern is the case where they overlap — "a lot of private insurance companies that are owned by private equity… and those private equity insurance companies sometimes own also some of these private label companies."
4. Score conflict by counting the number of independent parties between the loan and the capital charge. When origination, rating and holding collapse into one ecosystem, the rating stops being an outside opinion.
5. Discount accordingly: serious insurers "take it with a pinch of salt or they know the issues, here be dragons." Treat an investment-grade label from inside the ecosystem as unrated.
6. Watch this as a slow-burn systemic item — his framing is "could at some point bear watching," not an imminent call.

• Private-equity acquisitions of life insurers and annuity books, and what those balance sheets then buy.
• The share of an insurer's investment-grade bucket carrying a rating from a non-big-three agency.
• Regulator commentary (NAIC and equivalents) on private-label ratings and capital-charge arbitrage.

50:10 10. The fallback-product test — what survives if the thesis is wrong?

1. Assume the technology thesis fails entirely, then ask what the company still earns. Not a haircut — zero.
2. Score the answer as an existing, cash-generating product with its own customers. "People are still going to be going on YouTube even if Google wastes a few hundred billion dollars on AI data centers. And that's going to save them."
3. Fail any borrower whose only product is the thesis. "With a CoreWeave, do they have that backup? Maybe crypto mining, I don't know."
4. Rank the peer group by debt carried relative to the size and durability of the fallback business — this is what separates "not great for investors" from insolvency.
5. Read the EBITDA-to-net-income gap as the tell for how much of the operating result is being consumed by depreciation and interest, the two costs a chip-buying, debt-funded business cannot escape.
6. Expect the outliers to be few: "there will be of course in any cycle a few extreme outliers that just borrowed way too much money."

• The EBITDA-vs-net-income gap widening at capex-heavy borrowers.
• Delayed-draw term loans and other undrawn facilities presented as balance-sheet strength.
• Backlog disclosures whose counterparties are the same few hyperscalers — concentration and circularity in one number.

47:59 11. The cheapest-funding-route test — when "flexibility" means optics

1. For any large borrowing by a large public company, establish the cheapest available route. For an investment-grade issuer that is almost always a plain-vanilla general-purposes corporate bond.
2. If the company chose a more expensive route — private credit, a lease structure, a JV — the difference is a price it paid for something. Identify what.
3. Reject "flexibility" as the answer unless the company can say what flexibility, at what cost. "I don't even really know what that actually means… flexibility is probably a convenient excuse."
4. The usual real answer is presentation: to obscure a business changing character "from being lean mean cash machines into being capex hungry utilities," i.e. "making them seem healthier than they really are."
5. Then apply the same scepticism to the income statement: check how much of reported net income is other income — revaluations of stakes in private companies rather than operating profit. "If you take that away, then some of those earnings look a little bit not bad, but definitely not as good" (53:51).
6. Wait for the disclosure event that settles it. For private counterparties, the S-1 is where revenue quality finally has to be shown — "a popcorn moment."

• Investment-grade issuers financing outside the public bond market with no cost explanation.
• "Other income" as a share of net income at the hyperscalers, quarter over quarter.
• The OpenAI and Anthropic S-1s: the first audited look at how much of the reported revenue is cash.