2:35 1. The thesis-creep exit rule — sell when the pillar breaks, not when the price does
The repeatable method
- At entry, write the thesis as a small number of named pillars, and state which are cheapness and which are fundamental. Here: "insanely cheap" and "fundamentals would get better." Cheapness alone was never the reason.
- Define in advance what would falsify the fundamental pillar and over what horizon. For a subscriber business it is the direction of net subscriber change, not the profit line.
- Grade it every quarter against its own trend, not against consensus. Sequence beats level: 4Q improving → 1Q deteriorating → 2Q worse again is a trajectory statement.
- Apply the count you set: "two really bad quarters in a row is enough for me." Discount management's reassurance — "management stated that the worst is over… however, at this point, I'm skeptical."
- Then run the honesty test. If the only remaining reason to hold is the leg that never was the thesis, you have swapped arguments: "continuing to own the stock just because it's cheap — and it is cheap — would be thesis creep." Sell.
- Keep the door open without keeping the position: "if the fundamentals ever turn, I could come back to this stock." Exiting is not a permanent verdict on the company.
Here: CHTR, recommended in January, exited after 172,000 broadband losses in 2Q26 — worse than 1Q, "much worse than expected." "Bottom line, I give up. Simply put, I made a mistake." Contrast the same author's May 1 stance, when he added through a 120k-loss quarter because the pillars still held ("I'm being paid to wait") — the rule cuts both ways, which is what makes it a rule.
Watch for
- Your own reasons for holding drifting from the ones you wrote down; a valuation argument doing all the work; management guiding to a turn that keeps slipping a quarter; a deteriorating operating metric masked by an improving balance sheet ("the company paid down some debt, which is important" — and not enough).
3:17 2. Announce the change before you act on it — and don't let the tape talk you out of it
The repeatable method
- If you have published a recommendation, treat the disclosure of a change of mind as preceding the trade: "I recommended the stock to my viewers and I strongly believe that I should not sell until I inform my viewers of my opinion change. I will be selling the stock next week."
- Separate the decision from the price at the moment of the decision. The stock had just rallied 15% off a 52-week low — a tempting reason to defer.
- Attribute the move before you weight it. "The rally in the stock in my view has nothing to do with Charter. At least for this week, investors are reallocating out of AI related plays and that is benefiting Charter's stock price."
- Rule: a price move caused by flows into your sector because of something happening elsewhere is not evidence about your company. Do not upgrade a thesis on rotation.
Here: CHTR closed at $123, a 52-week low, then ran to $142 (+15%) by Thursday — on money leaving AI names, not on anything Charter did. The sell decision is unchanged, and pre-announced.
Watch for
- Sharp rallies in unrelated cheap names during an AI/tech drawdown; a stock rising while its operating metrics worsen; the temptation to "sell into strength later" turning into not selling.
5:06 3. Split a theme into layers with different moats — and notice when your own bear point flips into a moat
The repeatable method
- Refuse to hold one view on a whole theme. Draw the boundary explicitly: "there is a major difference between hyperscalers and LLM providers" — the ones building the data centres versus the ones making the models. Note the overlaps (Google, Microsoft are both).
- Test each layer for a moat separately. For infrastructure, ask whether the capital requirement itself excludes entrants: "the amount of money it takes to be a hyperscaler is insane. And that expenditure itself is a moat… there are only going to be a few."
- Be willing to invert your own argument when the evidence supports it. Capital intensity was the core of his bear case; here the same fact becomes the barrier to entry. A cost that only a handful of firms can bear is a barrier.
- For the layer without a moat, look for the substitution evidence directly: "enterprises are switching between models and using cheaper open-source Chinese models in order to control costs." Switching behaviour is the moat test, not benchmark scores.
- Rank within the weak layer by breadth of revenue: firms with "multiple revenue streams from established businesses which are very unlikely to simply disappear" survive a price war that kills single-product firms.
Here: moat layer — MSFT, AMZN, GOOGL, ORCL ("they have real businesses here"). No-moat layer — Anthropic, OpenAI, and the model businesses inside Google and Microsoft, which survive only because of the other revenue. The pure-play failure mode: META, "trying to play just in the LLM space which is expensive and not yet lucrative enough."
Watch for
- A new entrant funding hyperscale capex (the barrier eroding); enterprises consolidating onto one model rather than routing between several (a moat forming); price per token stabilising; a company defending a layer it doesn't actually occupy.
7:28 4. Find the catalyst by tracing the backlog to its counterparty — even when the counterparty is private
The repeatable method
- For any theme, ask what would have to break for a sustained selloff rather than a wobble — and insist the answer be a specific, observable thing.
- Follow the revenue backwards. The listed companies' order books are promises; find out whose promises. "So much of the hyperscaler backlogs are from these two companies. For example, of Oracle's 600 plus billion backlog, around half is from OpenAI."
- Name the monitor even if it is unlisted: "a key thing to monitor to determine a catalyst for a real sustained selloff is the health of Anthropic and OpenAI." Private-company health is trackable through funding rounds, pricing moves, secondary marks, hiring, and enterprise churn commentary.
- Understand the transmission before you need it: no-moat pressure at the model layer → those firms cut commitments → the hyperscalers' backlogs deflate → "the entire AI ecosystem could go through a correction phase."
- Re-check concentration inside every backlog you own. A large backlog and a concentrated backlog are different assets at the same headline number.
Here: ORCL is the load-bearing example — read against Gil Luria's argument four days earlier that the same backlog is "valued by the market at zero." Both statements are consistent: cheap because concentrated.
Watch for
- Model-provider price cuts or funding rounds at flat/down valuations; enterprise churn away from closed models; a hyperscaler disclosing customer concentration; contract terms described as "flexible arrangements" rather than firm commitments.
9:54 5. Read the credit market for the cycle turn — watch the spread between borrowers, not the level
The repeatable method
- Track new-issue yields inside a theme, not just equity prices. A bond deal prices the market's judgement about survival, which equity enthusiasm can mask.
- Compare two borrowers financing the same asset. When one pays a small spread and the other pays "more than, get this, 9%," lenders have stopped treating the theme as a single risk.
- Name that state: "there is something of a credit cycle here as fixed income investors are discriminating between the large companies like Google that they know can pay the money back and smaller newer companies like CoreWeave that are more risky."
- Apply it to positioning: the equity of a borrower paying that spread carries a financing risk the strong balance sheets don't, regardless of how similar the businesses look.
- Use it as an early indicator — credit typically differentiates before equity does.
Here: CRWV raising $2.6B at 9%+ "to fund additional computing capacity," against GOOGL as the reference borrower. Alongside: NASDAQ −7% from its Jan 2 high and SOXX −23% from June 22 — the equity market repricing the same theme with a much blunter instrument.
Watch for
- New-issue yields widening for second-tier names while investment-grade holds; deals pulled or upsized on covenants; sale-leaseback and off-balance-sheet lease structures growing; lease commitments rising faster than reported debt.
18:52 6. Shorting a monopolist under assault: the trigger is any sign of weakness, not a bad quarter
The repeatable method
- Establish that the valuation rests on pricing power, not on growth — a monopolist is priced for control of its own outcome.
- Decompose the EPS growth. Ask how much came from price increases and how much from expense timing: "the big EPS growth rate is largely due to FICO raising prices for years. And the EPS beat this quarter was also because of lower than expected expenses."
- Then look only at the lines a monopolist should never miss — revenue and forward guidance — because those measure volume and confidence, which price increases cannot fake.
- Treat any miss there as disproportionate evidence: "a company whose entire monopolistic business model is potentially under assault can show no signs of weakness. Missing on revenue and providing soft guidance is weakness."
- Keep the clock realistic on the displacement itself — "it is still early in that process" — so the position is sized for a slow thesis with fast repricings.
Here: FICO reported EPS 1218 vs 857 (+42%) and fell 17% in a day, because revenue of $674M (+26%) missed and guidance was soft — against the VantageScore mortgage-share thesis he has been short on for months.
Watch for
- Price-driven revenue growth with flat or falling volumes; guidance trimmed while EPS beats; expense-driven beats; the first customer publicly adopting the substitute; a regulator or GSE approving the alternative.
12:00 7. For a hypergrowth name, price the multiple decay — and state the condition it depends on
The repeatable method
- Don't judge a fast grower on the current-year multiple alone. Lay out the same price against the next two or three years of estimates and read the slope.
- Say both halves out loud: "this is and is not an expensive stock. It depends how you look at it. The 2026 estimated PE is a high 73 times, but… the 2027 and 2028 estimated PEs are only 37 times and 23 times."
- Convert the valuation question into an execution question — you are no longer asking "is it expensive?" but "will those out-year estimates arrive?"
- Attach the explicit condition to the position, so the exit is pre-specified: "if the AI story keeps going, I would expect Bloom stock to continue to perform. But again, the AI story has to keep going."
- Prefer names where the growth is anchored in a physical bottleneck (here, power for data centres) rather than in sentiment.
Here: BE — EPS +680%, revenue past $1B for the first time (+166%), stock +90% YTD, 73× 2026 falling to 37× and 23×. The same conditional framing separates it from PWR, where the raise in full-year guidance is evidence already in hand rather than a forecast.
Watch for
- Out-year estimates being cut (the decay curve flattening); growth funded by dilution rather than orders; the anchoring bottleneck easing; a conditional thesis quietly becoming unconditional in your own notes.
10:39 8. Read the consumer with two independent networks plus one staple — the K-shape in three numbers
The repeatable method
- Use total payment volume at the card networks as the unfiltered read on spending — it isn't distorted by one retailer's execution. Check two networks so a company-specific result can't be mistaken for a macro signal.
- Then read a mass-market staple's organic revenue growth (stripping acquisitions and currency) as the volume read on the ordinary consumer.
- The spread between the two is the K. Strong card volume with zero organic staple growth means the aggregate is being carried by the top half.
- Keep the historical caveat in view — his own mailbag answer concedes inequality is not new, and the Cassidy GDP-per-capita chart is the reminder that levels have risen for everyone even as the split widens.
Here: V payment volume +10% and MA +8% ("no signs here that the consumer is slowing down") against PG at EPS −3% and 0% organic revenue growth. Same method as the Domino's same-store-sales read a week earlier.
Watch for
- Card volume decelerating toward staple growth (the K closing from the top); staples turning negative organically; cross-border vs domestic volume diverging; discounters gaining share from mid-market brands.
15:41 9. Distinguish capex that produces revenue from capex that only produces cost — and hunt the off-balance-sheet leases
The repeatable method
- Both spenders will show falling free cash flow. Do not stop there — free cash flow alone can't tell you which spend is working.
- Find the revenue line the capex is supposed to feed and check its rate of change. Accelerating means the spend is converting: "Azure saw revenue growth accelerate to 43% versus 40% in the March quarter."
- Compare expense growth to revenue growth. "Revenue was up 28% but expenses climbed 55%" is a company buying growth it isn't getting.
- Read the capex guidance for its floor, not its midpoint — narrowing a range by raising the low end is a commitment to spend more, "in other words, it raised the lower end of the range."
- Then go looking for the obligations that aren't shown as debt: "279 billion in future lease agreements, mostly related to AI, that are not yet reflected on its balance sheet. That is up, get this, 53% in just 3 months." Track the growth rate of that number, not just its size.
Here: MSFT — FCF −23% but Azure accelerating, "a very good quarter." META — FCF of $784M ("basically nothing"), R&D from $13B to $22B, weak guide, capex floor raised, $279B of off-balance-sheet leases. AMZN confirms the Microsoft pattern with AWS at 37%.
Watch for
- Cloud/AI revenue growth decelerating while capex rises; expense growth outrunning revenue growth two quarters running; lease commitments compounding faster than reported debt; capex ranges narrowed upward.
20:48 10. Being right about the technology is not a risk process — size for the drawdown you can't predict
The repeatable method
- Separate two competences that look identical in a bull market: knowing the subject, and managing the capital. "There is more to managing a hedge fund than just being smart and knowledgeable. Risk management is key."
- Treat leverage as the variable that converts a correction into a permanent loss: "the fund was on so much margin that because of the recent tech correction, a significant portion of the capital is gone."
- Set the exposure so that being wrong is survivable: "you need to be sure that you are not taking risks that way overexpose your investors. Sure, making money on the upside is great, but protecting the downside is just as important, maybe even more so."
- Discount track records built in one regime, and discount press acclaim entirely — "lauded by the press as a genius. Now maybe he is a genius" — because the returns and the risk were produced by the same leverage.
- Note the market-structure consequence: forced sellers do not sell what has fallen most, they sell what they can. Margin-call liquidations amplify the drawdown in exactly the names already under pressure.
Here: Situational Awareness, the $20B fund founded by a former OpenAI employee, "was forcibly liquidated to meet margin calls" on Thursday — the same day it was reported to be seeking fresh capital. Context: NASDAQ −7% from its high, SOXX −23% from June 22.
Watch for
- Concentrated thematic funds reported to be raising capital after losses; margin balances at highs into a correction; single-name liquidations on no news; a manager whose edge is subject expertise rather than process.