Actionable insights — Robotics, AI, and A Commodities Supercycle

1. Locate the technology on its tech tree, then ask what the next rung physically consumes (5:23)

1. Describe the current era by its dominant machine, not by its most-discussed application: the industrial revolution was the era of analog compute (steam engine → turbine → fission); the information era is the digital computer.
2. Identify the structural inefficiency the current machine accepts in exchange for its advantage. Digital computers bought generality at the cost of having to flip bits, which is why scaling them is a memory-and-power problem rather than an ideas problem.
3. Ask what the successor machine would have to be if that inefficiency were removed — and, crucially, what would have to exist first for the successor to be buildable at scale.
4. Convert that prerequisite into a bill of materials. This is the step that turns a technology view into a tradeable one.
5. Sanity-check the timing by asking whether the current rung is exhausted or still scaling. If it is still scaling, you are early enough to accumulate the inputs and too early to short the incumbents.

• Announcements of analog / in-memory / photonic compute aimed at matrix multiplication (the bottleneck he names); robot unit-volume guidance from any manufacturer; and any sign the digital scaling rung is finishing early — which would pull the whole sequence forward.

2. Size a new theme by bill-of-materials arithmetic against annual mine supply (28:53)

1. Take the most aggressive stated unit ambition from a credible operator — not your own forecast, theirs, so the number is quotable and falsifiable.
2. Find the per-unit intensity of one input (ounces, kilograms, metres per unit). One number, one material.
3. Multiply. Then divide by annual global supply of that material, not by its market cap or by anything financial.
4. If the ratio is a material fraction of annual supply, you have a physical constraint, and the price has to solve it. If it is a rounding error, drop the idea — no matter how good the narrative is.
5. Repeat the same arithmetic for the next input down the bill of materials (structure, wiring, magnets, batteries) and rank by tightness of supply.
6. Cross-check against the price: if the metal has already inflected without an obvious cause, that may be the market doing this arithmetic before you.

• Published silver intensity per humanoid unit as designs firm up; humanoid production guidance versus actual shipments; annual silver mine supply and above-ground inventories; and the same arithmetic applied to aluminium, copper and rare-earth magnets.

3. In any shortage, buy whoever owns the scarce input — never whoever has to buy it (42:51)

1. Write the shortage down as a sentence with a subject: X is scarce.
2. Draw the value chain for X and mark, for each listed company, whether it is a seller of X or a buyer of X.
3. Buyers get a margin squeeze first and a pricing benefit later, if ever — and only if they can pass costs through faster than they incur them. Assume they cannot.
4. If the only listed names are buyers, and no one owns the scarce input in investable form, pass on the trade entirely rather than substituting the nearest available vehicle.
5. The one exception worth checking: a buyer who happens to hold large physical inventory of X on the balance sheet, which converts them temporarily into a seller.

• Sulfur and naphtha price series versus fertilizer and petrochemical output prices; disclosed feedstock inventory and hedging in the notes to accounts; and whether a listed pure owner of the scarce input exists at all before you build the thesis.

4. When the frontier is crowded with undifferentiated competitors, buy the standardised input they all consume (1:05:27)

1. List the competitors at the frontier of the theme. If they are all pursuing the same architecture with the same capital access, none of them has a durable moat.
2. Ask what every one of them must buy regardless of who wins. That input is where the theme's economics are safest.
3. Test whether that input is becoming commoditised — fungible, standardised, priced on a spot-like curve. Commoditisation is bad for the seller's differentiation and good for the buyer of the cycle, because it makes the demand forecastable.
4. Prefer this leg when you cannot resource proper company-level diligence on the frontier names.
5. Keep a watch on the frontier anyway, and define in advance what price would make the equity leg buyable.

• DRAM / HBM contract pricing and lead times; announcements of on-premise model deployments by non-hyperscalers; and, on the frontier leg, robotics equities de-rating toward a defensible earnings multiple.

5. A change in fiscal terms can replace a price forecast — screen for royalty and tax changes, not just geology (36:55)

1. Monitor provincial / state / national resource-royalty announcements the way you would monitor drill results. They move project economics instantly and are usually under-covered.
2. Quantify the change on a per-well (or per-tonne) payback basis: how many units of production are now required to return the up-front capital, versus before?
3. Identify the operators whose drilling programme is physically inside the boundary of the incentive. Proximity is not enough; the acreage has to qualify.
4. Restate the investment case without a commodity-price assumption. If the payback works at the strip, the thesis no longer depends on being right about price.
5. Then list what could still break it — usually policy, not rock — and decide whether that residual risk can be hedged (see insight 6).

• Provincial budget and royalty-framework announcements in Saskatchewan, Alberta and BC; per-well IP30 rates from operators inside the qualifying zone; and how much of a company's 2026–27 drilling programme actually falls under the holiday.

6. Screen jurisdictions to exclusion, then hedge the single policy risk left standing (34:25)

1. Before valuing anything, eliminate whole jurisdictions on the grounds of expropriation, security or fiscal hostility. Write down the disqualifying reason for each.
2. Whatever survives is a concentration by construction, not by choice. Say so, because it tells you exactly which single risk your book now carries.
3. Name that residual risk as a specific, datable policy event — not "geopolitics," but "a US crude export ban."
4. Find the instrument that rises violently in precisely that scenario and would otherwise expire worthless. Cheapness is the point: it must be affordable to carry for years while being wrong.
5. Size it against the failure mode you are insuring, not against expected value. For a margin account, the correct size is "enough that the margin call cannot happen."
6. Accept the carry cost explicitly as a cost of doing business in the surviving jurisdiction.

• Any US export-restriction or products-ban language out of Washington, and Canadian retaliation signalling on Alberta/Saskatchewan barrels; the Brent–WTI spread versus its ~$3 norm (it was $6 on the day); and implied vol on long-dated Brent calls, since the hedge gets expensive exactly as it gets popular.

7. Treat unanimity as a sell signal — count who is left to buy (51:08)

1. Model the market as an evolutionary system: strategies that work attract imitators, imitators compete away the return, and the crowd then suffers a common shock.
2. For any position, ask who is on the other side and why. If you cannot construct their argument, the trade is crowded.
3. Look specifically for the moment the public consensus forms — everybody calling the same bottom, the same breakout, the same seasonal. That timestamp is the risk marker.
4. Act by reducing size rather than reversing, so a continued move does not force you to fight your own thesis.
5. Keep the freed capital earmarked for the same asset at a lower price. The point of the trim is to become the marginal buyer later.
6. Distinguish this from a fundamental change: your view of the asset may be unchanged: only the price of consensus has moved.

• Timeline consensus on a single level or narrative; positioning data (COT, fund flows, options skew) confirming it; and whether you can still articulate the bear case for your own long.

8. Ask what your own thesis working would do to the rest of your book (55:57)

1. Take your highest-conviction position and follow its success through to the second-order macro consequence, not just the P&L on that line.
2. Ask which of your other holdings that consequence damages. Correlations that look diversifying in normal times often collapse in exactly the scenario you are betting on.
3. Reduce the position that would be hurt before the event, even if its fundamentals are fine, and say out loud that you are trimming on scenario grounds rather than on the merits.
4. Pre-commit what you would buy with the proceeds if the scenario arrives, so the drawdown becomes a shopping list instead of a shock.
5. Accept the opportunity cost of being early, and state the condition under which you would re-add.

• Credit spreads and funding stress as the confirmation that a crude spike is turning into a liquidity event; gold-miner valuations on a pullback (his stated buy trigger); and whether gold and oil finally decorrelate.

9. At the point of maximum despair, convert equity into convexity — but only against a datable, falsifiable belief (58:48)

1. Identify a drawdown caused by a specific headline you believe is wrong — not by deteriorating fundamentals. The distinction is the entire trade.
2. State what would have to be true for the headline to hold, and why you think it will not. Here: a ceasefire memorandum, against a counterparty escalating its demands.
3. Run the arithmetic of the swap: on a rebound, equity recovers roughly in line with the underlying; out-of-the-money calls can multiply. Selling the former to buy the latter concentrates the recovery.
4. Cap the conversion at a fixed, survivable share of the portfolio and choose expiries long enough that the thesis has time to resolve.
5. Accept that this raises, not lowers, portfolio risk. It is a deliberate concentration into one outcome.

• Headline-driven versus fundamental drawdowns (does the physical data confirm the story?); implied volatility at the moment of despair, which determines whether convexity is cheap; and a hard cap on the share of the book converted.

10. Anchor "cheap" to a number you have actually paid, and size conditional stories near zero (1:08:21)

1. Keep an explicit numerical anchor from a past position where the price was unambiguously wrong — an enterprise-value-to-cash-flow multiple, not a story.
2. Measure every new candidate against that anchor. If nothing clears it, the correct conclusion is that the sector has re-rated, not that your bar is too high.
3. Separate cheapness from conditional cheapness. A developer that is cheap only if it executes is priced for the probability that it does not.
4. Size conditional names at a level where being wrong is irrelevant, and say why: "if they execute" is a sentence that belongs in the position-size decision, not the thesis.
5. Keep dry powder aimed at the one place the anchor might still be met — and be explicit about what event would create it.
6. Test your own theme with its purest listed instrument, and be willing to fail it on price alone.

• EV/cash-flow multiples across producing silver and gold miners versus the 1.5x anchor; execution milestones (permits, financing, commissioning, first steady-state quarter) that convert a conditional name into a real one; and the gold-miner pullback he has earmarked as the moment the anchor might be met again.

11. Where the consensus depends on bad data, the alpha is in the measurement, not the opinion (1:01:35)

1. For any market where everyone quotes the same third-party statistic, ask how that statistic is actually produced and how reliable the underlying collection is.
2. If the answer is "estimated," identify the single physical quantity that would settle the question — here, the number of tankers actually transiting.
3. Ask what it would cost to measure it directly. Independent measurement of a widely-guessed number is a durable edge in a way that a better opinion is not.
4. Cross-check the official series against observable events. A persistent gap between what is happening and what is being reported is itself information.
5. Treat suppression of coverage as a data point about incentives, not as proof of a conspiracy: ask who benefits from the number staying unclear.

• Divergence between reported tanker transits/AIS data and confirmed strike counts; global floating-storage and onshore inventory draws as the check on the "4 million bbl/d shut in" claim; and whether crude finally responds to physical disruption (his own stated test: "next time we talk, I'll let you know how it's gone").