AI PhD turned investor · robotics, AI and the commodities supercycle · guest on the Value Hive Podcast — running synthesis of appearances, with per-transcript breakdowns and a stock index.
The second-order robotics material, from the deliberately childlike question "what are the robots going to be made out of? Definitely aluminum — you're going to want them to be light." Earlier-stage than silver and not yet sized ("we're going to have to keep our eyes open for other things"); carried here for the method — decompose a projected unit volume into its bill of materials and find the constrained input.
Structural long expressed as portfolio insurance rather than a directional bet: 5% of net worth in long-dated out-of-the-money Brent calls, sized so a US WTI export ban cannot margin-call a leveraged book. Tail case "there's no way Brent stays below 500… I think Brent would go to 1000." He notes the Brent premium had doubled to ~$6 — everyone is hedging the same way, and "they don't matter until they matter, and then the gamma blows someone's head off."
The one tungsten equity he names, and he names it as an under-sizing error rather than a fresh idea: "tungsten I think is my biggest mistake. Like EQR. I should have had more." ASX-listed producer, one of very few tungsten mines outside China; the chart has already run, so this is a positive view on a name he under-owns, not a chase.
The only AI-complex equity his framework lets him own, bought small on a dip. Digital compute's inefficiency is a memory problem, and his own quote for an on-premise LLM was "like 4 million U.S. dollars and most of the cost is memory" — Jevons paradox with independents and small businesses as the incremental buyer. "Memory and compute are evolving into commodities… memory has got another leg to its cycle that will probably break people who aren't long memory."
The only named E&P, and the thesis is fiscal terms rather than geology: Saskatchewan's royalty holiday puts the first 38,000 barrels per well at a 2.5% royalty instead of the usual ~25%, which "basically guarantees that these wells will pay off their own cost as long as they're not a dud." Payback stops depending on a bullish oil price. Residual risk is policy, not rock — a Canadian barrel inside a live export-ban standoff, which the Brent calls are there to offset.
The purest expression of the robotics call and the one he says the market has not done the arithmetic on — "every Optimus Tesla robot needs an ounce of silver" against Musk's one-per-household ambition, plus EV batteries and solar. "Silver's inflection — I think actually it's front running robotics." Price call: "$100 is about to be the floor price… it may not go much higher" — so no new physical (his bullion is old $40–50 stock held as a rainy-day fund) and miners instead, for the operating leverage.
Named as the position he most regrets not sizing ("my biggest mistake"), on demand that is literal destruction — Ukraine and the Middle East "blowing shit up made of tungsten, just vaporizing it" — plus Western rearmament and Trump's stated move to "a military economy." The thesis is unchanged by the missed entry: "I feel like tungsten is still great here."
The vehicle for the trade he calls "probably the ballsiest thing I've ever done in my life": after a ~60% drawdown from the top on the Iran peace-memorandum headline and crude at ~$68, he liquidated 5% of the portfolio into oil calls, "mostly USO at the time," on the arithmetic that "equity drops less than calls will go up on a rebound." It slingshotted the book back to all-time highs. A description of a maximally aggressive margin trade, not a template.
Not a current pick — the valuation anchor he measures every miner against, and the reason his mining sleeve is under-sized despite a bullish metals view. One of the two names behind his ~2,000x personal return, bought "at 1.5 times enterprise value over cash flow": "I wish you could find an APM right now." His conclusion is about the market, not his bar — "the market is starting to… price things more correctly, at least in miners."
A realised winner named for the record with no forward view: asked what drove the run to new all-time highs, "CLMT was a driver… from like $12.00 all the way to 27 or something," held through calls and credited to someone else's idea. Useful mainly as evidence of the shape of his returns — a few leveraged, high-conviction expressions rather than a broad book.
Interested and deliberately barely positioned — his cleanest statement of an execution-risk discount. "I think if they execute, that's interesting… but I barely own any because there's still this 'if they execute'." Beylo calls it the cheapest silver miner out there; McCracken treats cheapness that is entirely conditional on future operational delivery as priced risk, not value, and is "still unsure which ones are the best to play with."
In the too-hard pile on input-cost opacity rather than demand: "the chems are super hard because their input costs are distorted… things like naphtha are exploding because it all comes out of Hormuz." The general rule — a commodity shock splits a chain into the owner of the scarce input and the buyer of it, and if you cannot tell which side a company sits on, it is a coin flip rather than an investment.
A shortage he believes in and still refuses to own — his sharpest "right thesis, wrong instrument" example. "A bunch of people made mistakes where they bought fertilizer companies. It crashed because… there's a sulfur shortage. But unless the fertilizer company had sulfur on hand… that's gonna hurt their inputs. So there's probably gonna be a fertilizer crisis and it's hard to make money off it." He could not identify a clean listed sulfur owner and passed rather than substitute a worse vehicle.
Cut ~50% (gold miners) while conceding the fundamentals are fine — a scenario trim, not a fundamental one. Two reasons: gold does badly in a genuine liquidity crunch, with retail pawning jewellery adding physical supply exactly as financial demand is liquidated; and positioning — "literally everyone on Twitter called the bottom. Everyone." Hence "I think gold pulls back a bit here." The intent is to be liquid for it: oil spikes, liquidity cracks, gold crashes, and he rotates into cheap ounces — "it'd be like a legendary play."
The purest listed instrument for his own robotics call — and he failed it on price alone: "I was looking at the price, it was IPO, and I was like, yeah, no," despite having met the company at NeurIPS/ICLR. The generalised problem with robotics equities is "which ones, and are they already priced correctly?", which is exactly why he prefers the materials leg until a moment "basically like buying NVIDIA in 2022." Beylo adds the ownership risk of a Chinese robot in a bifurcating world.
Not a price forecast — a jurisdiction/policy veto that disqualifies otherwise-cheap North American E&Ps. A US crude export ban would strand American barrels and collapse WTI against the world price; the Canada trade war raises the odds ("let's do this WTI ban as well as products ban"), with Canadian retaliation on Alberta/Saskatchewan crude making it lose-lose — "shooting ourselves in the foot while people like China benefit." The expression is to own the barrels and buy Brent optionality against them.
In one line:The AI boom is not a technology trade, it is a materials trade — and the way to own it is the stuff robots are made of, not the companies that make them. McCracken reasons forward from computer science rather than from macro: digital computers won the 1940s argument by being general, and paid for it with brute-force inefficiency; the next efficiency step is purpose-built analog hardware; you cannot build analog hardware at scale without an era of general robotics; and robots are silver, aluminium and steel. Hence — "we're going to see an absolutely roaring commodity super cycle. I'm making that call now." The book that expresses it is concentrated, margin-financed and deliberately convex: long oil with the one policy risk (a US WTI export ban) hedged by long-dated out-of-the-money Brent calls, silver miners rather than bullion, and gold cut in half ahead of the liquidity crunch he expects his own oil call to cause.
Commodity supercycle, called explicitly — and derived, not asserted. The argument is a tech-tree argument (the information era is ending; the next rung is analog compute; that rung needs robotics; robotics needs metal), which makes it checkable at each link rather than a slogan about debasement or shortage.
Silver is the flagship, sized by bill-of-materials arithmetic. "Every Optimus Tesla robot needs an ounce of silver" against a stated one-per-household ambition, plus EV batteries and solar — and he reads the recent inflection as the market beginning to front-run robotics. Price call: $100 "about to be the floor price… it may not go much higher", which is why he wants miners (operating leverage on a plateau) rather than new physical.
Own the scarce input, never the buyer of it. The rule that kills two whole sectors for him: a sulfur shortage hurts fertilizer producers' inputs, and Hormuz-distorted naphtha makes chemical margins unforecastable — "there's probably gonna be a fertilizer crisis and it's hard to make money off it."
Jurisdiction screen to exclusion, then hedge the survivor's one policy risk. Africa, South America and the North Sea are struck out on expropriation and windfall taxes, leaving North America by construction — so 5% of net worth sits in long-dated out-of-the-money Brent calls against a Trump WTI export ban, sized so the margin call cannot happen. "Hedges cost us money… I have no choice."
Fiscal terms can replace a price forecast. His one bottom-up idea is a tax change, not a drill result: Saskatchewan's royalty holiday (first 38,000 bbl/well at 2.5% vs ~25%) "basically guarantees that these wells will pay off their own cost as long as they're not a dud."
Crowding is a sell signal; the market is an evolutionary system. Two years on P versus NP left him modelling markets as systems that punish convergence — "if everyone is doing the same thing… you get an extinction event." Applied here: gold miners cut ~50% because "literally everyone on Twitter called the bottom," and because gold historically fails in a liquidity crunch — the very crunch his $200-oil scenario would cause. The intent is to be the buyer of that crash, not a victim of it.
Valuation anchored to a number he actually paid. Andean Precious Metals at 1.5× EV/cash flow (one of the two names behind his ~2,000x) is the bar; nothing clears it today, so the mining sleeve stays under-sized and conditional names (Kuya, "if they execute") are held near zero rather than sized to the story.
He will not pay up for the theme he called. He passed on the Unitree robotics IPO purely on price and says the AI bubble "isn't anywhere near close to popping" — so he owns the input (Micron, small, on memory as a commodity) and waits for a robotics entry "basically like buying NVIDIA in 2022."
Style caveat. This is a personal, highly concentrated, margin-financed book with a documented ~60% drawdown and a 5%-of-portfolio conversion of equity into oil calls at the bottom. The methods travel; the position sizing does not.
Transcripts
One dated page per appearance — each has its full stock table, talking points, and the saved transcript. Newest first.
Gavin McCracken appears as a podcast guest rather than on a channel of his own — there is no YouTube upload of the Value Hive episodes, so discovery runs through the show's feed (Value Hive on Spotify for Creators). Newest first; a row moves up to Transcripts once processed.
Date
Title
Show / channel
Len
Video
2026-APR-10
Gavin McCracken: A Journey To 2,000x Returns — Part 1 (process, due diligence, edge, investing psychology, and a first pass at AI; the Valeura Energy / Andean Precious Metals story)