Intro
Gavin McCracken talks to Brandon Beylo about what he saw at an AI mathematics conference, why he thinks digital compute is the last leg of the information era, and how a robotics build-out turns into demand for silver, aluminum and tungsten. The second half is a tour of his own book: Brent calls held against a WTI export ban, oil shut in around the Strait of Hormuz, and gold miners he has been cutting.
Guest: Gavin McCracken, mathematician and AI researcher, and a private investor in mining and energy who publishes on Substack
Host: Brandon Beylo
Published: 28 August 2026 on Value Hive Podcast
Episode page | 1 hr 14 min
Key Takeaways
A roaring commodity supercycle is the call, and robotics is what starts it
"we're going to see an absolutely roaring commodity super cycle. I'm making that call now"
The mechanism is AI helping people design analog computers, which have to be physically built
Silver is front-running robotics, and he sees $100 as the floor rather than the target
"every Optimus Tesla robot needs an ounce of silver"
"I think a hundred dollars is about to be the floor price. It may not go much higher"
The AI trade is nowhere near popping, because memory demand keeps compounding
"that's why I think this bubble won't pop" — data centers and robotics both pull on commodities
China, not OPEC or the White House, sets the oil price into the US midterms
"Trump is actually in a position now where China controls the oil price"
A US ban on WTI exports is the tail risk he is paid to hold
"I have no choice, but to have basically 5% of my net worth locked in Brent calls"
If Canada retaliates, "there's no way Brent stays below 500"
$200 oil is the thing that ends the everything-rally, which is why he cut gold miners
"I cut gold by about 50%, my gold miners" — gold historically does badly in a liquidity crunch
Two metals markets, a China price and an ex-China price, are an opportunity rather than a problem
"two markets is absolutely going to give us so many opportunities"
Tungsten was the miss he regrets most, and the military-economy bid is still there
"tungsten, I think, is my biggest mistake"
LLMs are extremely good at breaking things and have proved nothing he found impressive
"AI has really just been proving things false"
Cheap AI made information cheap, not thinking
"they make information cheap. I don't think they make thinking cheap"
Nobody has built anything genuinely impressive since the nuclear reactor
"nuclear power plants are still the most impressive thing humans have ever built"
Everyone is winning at once, and evolutionary systems do not allow that for long
"not everyone's allowed to win, but everyone is winning right now"
The Conference Where Mathematicians Started Worrying About Their Jobs
McCracken had just returned from a Boston conference on topology, algebra and geometry in data science — a heavily mathematical AI conference — and said the room was uneasy
On what set it off: "multiple major conjectures have been solved now by AI" — Gavin McCracken
He described tweets from OpenAI employees posting counterexamples to conjectures "that's like, 200 years old or a hundred years old"
His own reaction was not concern: he thinks mathematicians will have to innovate, but sees the shift as an improvement rather than a threat
Beylo asked, in his words, a dumb question — how do you innovate in mathematics — and McCracken answered from his own path
He taught himself the subject and rediscovered theorems by accident: he found results in his master's work he thought were new, and learned afterward they were already established
On the method he uses instead: most mathematicians are "mostly just memorizing, like, all the results and like kind of doing like combinatorial search when they try and prove something"
He works the other way — break the problem apart and ask what the problem even is, which he says nobody in math or computer science really asks
Grothendieck's Motto and Why the Counterexamples Are What AI Is Good At
The line he says he lived by before he knew whose it was: Alexander Grothendieck's "we don't read books, we write them" — relayed by McCracken, not his own phrase
On what a large language model is actually doing: it takes an enormous amount of compute and searches for the answer
He would not call it intelligent search, but "it's not as bad as random search"
On the counterexample he found himself: it had so much going on that a human would have needed a year of work and would have had to hold all of it in working memory
He called that kind of construction "a waste of a human's life" — the human is not thinking, just building
The Civilization Tech Tree and the End of the Information Era
The framing for his whole investment view: "I actually think that we're at the end of the information era" — Gavin McCracken
He laid it out as eras in a strategy game: ancient, classical, Renaissance, then the Industrial Revolution
His name for the Industrial Revolution is the era of analog compute — the steam engine driving railways, the car, the airplane turbine, ending in nuclear fission as the ultimate turbine, gigawatt-scale and running on cheap uranium
The information era is the digital computer, and the trade-off is the whole story: analog machines like a sundial let the universe flow energy to compute for you, so they are fast and efficient but only do one thing
Digital compute buys generality and pays for it by flipping bits, which is why the machines need so much memory — his aside was that this is why "Micron stocks going up"
On where that leaves AI now: the only way to scale it is sideways, more GPUs stacked in parallel, which drags commodity and silicon demand along with it
On the bubble question: with trillion-dollar capex plans being discussed, "I still don't think it's anywhere near close to popping"
But he called it the last part of the tech tree for this era, and said what is being built is closer to virtual intelligence than intelligence
What Human Intelligence Has That Pattern Recognition Doesn't
On the definition of artificial superintelligence: if the test is a system that self-improves, humans already qualify — "every day I am better than I was yesterday"
On where patterns run out: in trading and investing, a situation can rhyme with something historical and still not be one you can go all in on
Beylo's own tangent on complexity: driving is a genuinely complex task that the brain automates until the driver barely notices doing it, which he offered as evidence of how good humans are at compressing complexity into patterns
McCracken's qualification: the automation holds until something out of distribution happens — "if something shocking happens, it's like your brain can really kick on"
He raised near-death experiences behind the wheel, where people report time slowing down
The Call: Robotics First, Then Analog Computers, Then Commodities
The prediction, made on the record: "we're going to see an absolutely roaring commodity super cycle. I'm making that call now" — Gavin McCracken
The mechanism is not AI demand directly: "It's going to be because people are going to use AI to help them build analog computers"
Why robotics has to come first: "you need to have general like workers that help you do things that would just be too overwhelming to do as humans"
He was explicit that he does not mean humanoids — tiny machines that position a cable would count
On how hard the physical build actually is: "the most complicated analog computer we ever built was the nuclear reactor"
The specific analog target he named: matrix multiplication, which he says is one of the bottlenecks in compute today and something you can imagine an analog circuit doing
Skip Connections, Transformers, and How Little Anyone Knows
Beylo asked why matrix multiplication is the bottleneck, and McCracken walked through it for the audience: a deep neural network is a sequence of matrices, and "matrix multiplication is perfectly parallelizable", so more GPUs make each layer faster
On the limit that used to exist: stacking layers made the gradient of the error go to zero, so the weights stopped updating
On the fix: a paper introduced skip connections, which let information jump five layers at a time so the gradient survives
"It's a great idea, but it's like so dumb that no one really thought of it"
On what came after: the Attention Is All You Need paper got transformers going, and after that "You just add as many layers as you can, use a ridiculous amount of compute, and train it well, and you get GPT"
His verdict on the state of the field: "We honestly don't know what we're doing" — the core idea is incredibly simple and most research is about training faster and better rather than about understanding it
Why AI Breaks Things Rather Than Builds Them
On what the models have actually delivered in mathematics: "AI has really just been proving things false"
Every major conjecture he has seen fall was disproved by a counterexample, never proved true in a way that impressed him
On the same skill showing up in security: he cited reporting that Claude Mythos was very good at breaking cryptography, finding a loophole and taking root-level access, and said GPT models do it too
"they seem to be incredibly good at finding where things break. At least for now"
Beylo tied it back to the intelligence question — the models can spot where a pattern fails but cannot generate their own conjecture — and McCracken agreed flatly
On what mathematicians are still for: deciding which problems are worth asking and which objects are worth constructing
The next stop after math is biology: he pointed to the recent move in mRNA-linked stock, where "People had calls go up by 400,000%", as AI picking things out of a very complex system
Where the commodity demand finally arrives: when AI is used to make robotics better and to build new physical things
Nothing Impressive Has Been Built Since the Reactor
His claim, delivered as an opinion he admits may be arrogant: "nuclear power plants are still the most impressive thing humans have ever built", and the know-how dates to the 1960s
On where the last great wave came from: mathematicians of the two world wars and the Cold War, working under life-or-death pressure
The example he chose is the Fourier transform, which he called the most influential and important algorithm in his opinion
The theory was roughly two centuries old, but nobody knew how to compute it quickly
Two people connected to John von Neumann worked out the fast version during the Cold War, on machines he described as calculators that take forever
The application was nuclear detection: working out whether Russia had set off a bomb requires multiplying polynomials, which naively costs n squared operations
With a fast Fourier transform it drops to n log n, effectively linear — "That's like saying you can multiply two things as fast as you can read them"
On war as the engine of invention: "war has inspired most of humanity's great inventions", with fertilizer coming out of the gases of the first and second world wars
His diagnosis of the drought: "people stopped asking what problems are"
The dominant approach now is to take everything you already know and aim it at the problem, which is exactly what a model does — "LLMs literally, they take the fact that you've trained them with the entire internet and they're going to just throw ideas at things"
That, he said, is why so much compute is needed: it is looping and retrying, and a gigawatt can run for a week to solve one thing
Cheap Calories, Cheap Thinking, and the Counterpoint
Beylo brought a tweet he had read arguing that when calories became cheap people became unhealthy, and that AI has now made thinking cheap in the same way
McCracken took the opposite side: he does not accept that most people were thinking in the first place
"Thinking is this other thing. It's like you deconstruct something. You have to truly understand it, reverse engineer it, know how to build it yourself"
His hope is that the models stay accessible to everyone, because they remove the need to read everything as a lookup table — his example was asking for a crash course on the Fourier transform, then asking for it again at high-school level
The distinction the section turns on: "they make information cheap. I don't think they make thinking cheap. I think thinking is different to information"
Learning to Think, and the Market as the Hardest System Ever Built
Beylo asked what someone without the training does to build the skill, comparing it to friends who can rebuild an engine, and suggested it is like exercise
McCracken's answer was breadth: "you have to do a whole bunch of different tasks", and he credits doing many different things for whatever he is good at now
On the market specifically: "I think that is actually the hardest system humanity has ever made"
Soccer and football are closed systems with a limited set of things that can happen, whereas "the market will keep creating new things forever"
Everyone front-runs the last pattern, and the front-running is what stops it repeating
Beylo's illustration was positioning in bonds: everyone is maximum short, so everyone wants to be long, so in theory you should be short again
McCracken's reply: there are levels to it, and you can dig yourself into some deep holes
Silver Is Front-Running Robotics
The first sign he pointed to: "silver's inflection is like, I think actually it's front running robotics"
Beylo noted the tape: "silver's pretty strong, like crossed over 70 today, which I was kind of surprised"
The demand math he does in his head: solar is known, but "every Optimus Tesla robot needs an ounce of silver"
When Elon Musk talks about one in every household on the planet, the ounce count is the question that follows
EV batteries need silver too
How he owns it: "I still, I'm just sitting on miners that produce it"
The other robotics metal he named was aluminum: "what are the robots going to be made out of? Definitely aluminum", because the machines need to be light
On why building demand broadens from here: information being cheap means people can find out how to build things quickly and then go and do it
3D Printing, and Why That Bubble Collapsed the First Time
Beylo raised 3D printing as a way to play a hardware constraint, noting the sector was obliterated after the SPAC boom, naming "Velo3D, 3D Systems" and pointing to bankruptcies in the space
McCracken's response was a question rather than an endorsement: anyone interested should first ask why the bubble collapsed the first time, and he said he does not know the answer
He has friends who 3D print and says it looks incredibly useful
He guessed the raw material is plastic of some kind and did not claim to know
Beylo added that some printers can now print metal, which McCracken took as the point — raw materials are the thing
Two Metals Markets and Resource Nationalization
Beylo laid out the bifurcation: the US and its allies on one side, China and its allies on the other, with rare earths already trading at "there's two prices: there's the China price, and then there's the ex-China price"
McCracken said that is exactly what he wants as a mining investor: a less efficient market is one you can take advantage of
"two markets is absolutely going to give us so many opportunities", especially if you can model which commodities are about to inflect
On the direction of travel: "I think resource nationalization is where we're going", a view he says he has held for a year and a half
His self-criticism: his portfolio is probably not reflecting it enough and he should own more miners of certain minerals
The reason he gives: "raw materials, commodities are going to be the thing that matter"
The WTI Export Ban That Sits On His Shoulder
Beylo said the idea has changed how he looks at US producers: he finds a cheap exploration and production name, sees that it is all WTI, and hears McCracken in his ear telling him a ban is coming
How McCracken is positioned: he uses margin, so "I have no choice, but to have basically 5% of my net worth locked in Brent calls"
They are long-dated and out of the money, so a ban would send Brent up rather than margin-call him
What he thinks a ban would actually do: eastern Canada takes its refined product from the United States, while crude flows south from Alberta and Saskatchewan to be refined there
A US ban on crude and products would crater eastern Canada, and Canada can retaliate by stopping its own exports
"we're all just shooting ourselves in the foot" while China benefits
The price he thinks that scenario forces: "there's no way Brent stays below 500. Honestly, I think Brent would go to a thousand", with the Strait of Hormuz already disrupted and Iran able to close it instantly
Why he is stuck hedging rather than avoiding: "all the good E&Ps are North American"
He rules out Africa outright on expropriation risk, is wary of South America, and dismissed the North Sea over "80% windfall taxes and communist governments will pass"
Saskatchewan's Royalty Holiday
The detail he had just written up on his Substack: Saskatchewan is giving a royalty holiday on new wells
The terms as he described them: "their first 38,000 barrels from each well are only at a 2.5% royalty. And that's a huge deal, because, it's usually like 25%"
The effect is that a well pays off its own cost as long as it is not a dud
He named ROK Resources as the operator drilling in southeast Saskatchewan, and credited Beylo with putting the name in front of him
The offsetting risk is the same political one: both Washington and Ottawa could ban exports
The Brent-WTI Spread and the Options Nobody Wants to Own
Beylo asked why Brent is not already higher if the risk is real, and put the spread at "I think the spread's pretty big. I just checked at six bucks"
McCracken thought it was smaller than he expected, having assumed around ten
On what it means: "Brent definitely is at a higher than normal premium. Usually it's like $3", so the premium has doubled
His explanation for why it is not wider: everyone is hedging the same way he is, with out-of-the-money options
They do not matter until they matter, and then the gamma takes someone's head off and whoever sold them gets fired
Tungsten Was the Trade He Missed
His biggest regret in the book: "tungsten, I think, is my biggest mistake"
Beylo agreed, saying "I'm so pissed I didn't buy EQR or Almonty" and that the chart is painful to look at
The demand case as McCracken framed it: tungsten is being vaporized in Russia and Ukraine and now in the Middle East, and there are reports of missile shortages
With Trump talking about moving to a military economy, "the demand is absolutely there"
Beylo brought a book he had just read, Tungsten in Peace and War, which he recommended for learning the tungsten market and which weaves in Bernard Baruch as a mining entrepreneur of that era
His read is that a military economy runs the same playbook again
On why tungsten is the cleanest way in: he said an investor friend who does defense investing told him pure-play missile exposure is very hard to find, since everything else comes bundled with ancillary businesses
McCracken's version of the same point: raw materials let you guess at a demand spike or a supply shock in a way equities do not
Hormuz, the Tanker War and the Media Blackout
His running joke against himself: last time everything was a Hormuz trade, and the update is that nothing has happened to the price
He counted two more tankers in the previous 48 hours and said the market is still waiting to see whether oil cares
Beylo's affectionate trading rule for his guest: "I'm like, dude, I just need to short oil when Gavin starts posting screenshots as P&L", and buy when the all-caps tweets start
McCracken owned the timing: he was posting because he was about to break out and did, describing a "sudden 50% on my net worth" to new all-time highs
Calumet calls were a big part of it, and "I got Toilet King to thank for that one", a call he did not act on from $12 to $27
On the chemicals companies: "The chems are super hard because their input costs are distorted", with naphtha exploding because it comes out of Hormuz
"it's the same reason I avoided fertilizer" — investors bought fertilizer names on a sulfur shortage, but unless the company held sulfur the shortage hits its inputs
Beylo ran through the fertilizer names anyway — Millennial Potash as the small cap that has done well, plus Nutrien, CF and SQM — and raised El Niño
McCracken has done a little work on it: if "Panama Canal has problems" and tankers have to go around Latin America, tanker rates go up
On his arithmetic oil should be 200, and he says it will not go there for what he calls magic reasons
He noted "Kazakhstan is once a month shutting in production" and closing its CPC terminal, and thinks Ukraine hit one of the loading docks
The anecdote he used to argue the story is being suppressed: he asked his parents whether they knew Ukraine was hitting Russian tankers and Iran was hitting roughly one a day, and they had heard neither
"it's very obvious there's a media blackout to manipulate the oil price"
China Controls the Oil Price Into the Midterms
His theory of Chinese behavior: they like making money and had no plan the first time Trump won because they expected Hillary Clinton
"I have a feeling they're going to try and make him lose midterms"
The reason he thinks they can: "the only reason oil is not 200 is China"
"They cut their imports by five million barrels a day", and rerouting plus strategic reserve releases covered the gap
"China was the single largest SPR player", and unwinding it means picking up the phone and buying every tanker on earth
The consequence he draws: oil spikes to 200, gas and diesel follow immediately, and Trump loses the midterms
"Trump is actually in a position now where China controls the oil price", with China aligned with Iran and Russia
Beylo's political read went further than the oil: a bad enough midterm takes out the reputations of everyone who staked them on the second term
He said the question investors then have to ask is "what would the world look like under an AOC presidency"
Sleepwalking Into Another 2022
Beylo's macro worry is a rerun of 2022 with Hormuz and El Niño layered on top
He recalled a Macro Ops trade — "We loaded up on wheat futures right before Russia started bombing" — and called the timing total luck
His view is that wheat and agriculture have not priced any re-escalation
McCracken sees the same price action as 2022 without the same prices: Ukraine is hitting Russian cargo and Russia is hitting Ukrainian cargo
"Brent went to 138" in 2022 on people assuming exactly what is now happening, and it did not start happening until a couple of months ago
His summary of the whole tape: "this is the most disconnected from reality market, like definitely in my life, but I feel like it might be ever"
"it feels like the price of literally everything is wrong"
Beylo's counterweight: "price is truth and you got to trade the market that's in front of you, not the one that you wish it was"
He relayed a phrase he attributed to someone he named only as Calvin — "Calvin likes to call it the great peace of our time"
The banter that closed it: Beylo compared the global conflict to a middle-school couple who will not say they are together, with everyone feeling it is World War Three and nobody willing to say so first
The Thread Between an AI PhD and Junior Mining
Beylo asked how the two halves of McCracken's life connect
His answer was computer science rather than AI: "I worked on P versus NP and I realized that's like the class of evolutionary problems", and the stock market is one of them
The idea he credits for his exits: "this concept of the self avoiding random walk"
He said he sold out of oil entirely in December 2023 ahead of the 2024 bear market
On who might know more: "the people who actually might know more would be like RenTec and Jane Street", because quants have been trying to extract alpha from an evolving system for years
Why even they get caught: "they end up competing with each other" and converge on the same strategy
A mini bubble forms, pops, and everyone loses 5 to 10 percent, which he framed as an extinction event caused by a lack of diversity
Everyone Is Winning, Which Is the Problem
Beylo on the absence of real bear markets: the technical 20 percent decline no longer means what it did, because every one has been a V bottom
A real bear market to him is five to ten years of no returns and two to three years of losses
"the market has like a TikTok brain" — it can only do things in rapid succession
McCracken agreed and dated the last one: "2008's the last time", and even that recovered quickly by his reckoning
His conclusion: "I think people need to be broken", because "not everyone's allowed to win, but everyone is winning right now"
"you have so many people who are up, more than 10x trailing five years", much of it in crypto
The cause he assigns is the debt bubble: cheap access to liquidity means everything goes up and everyone gets more indebted
The trigger he watches for is oil: "if oil goes to 200, there's a massive inflation that hits absolutely everything"
"oil is the blood of the modern economy", so that is what a liquidity crunch would look like
"This might be why they're so panicked about it" and why he thinks the tanker attacks are not being covered
Why He Has Been Cutting Gold Miners
The position change: "I cut gold by about 50%, my gold miners"
The reasoning: "I don't think gold would do well in a liquidity crunch. Right, historically it hasn't"
Beylo offered the sovereign version — countries selling gold to fund a war, as with Turkey
McCracken's version is retail: if the price of everything doubles, "people are gonna pawn their jewelry", especially at the current gold price
On the case for staying invested anyway: there is absolutely an end to cheap debt, but nobody knows when, so "Play the casino the best you can, even if you know it's going to blow up"
Beylo's own conflict: he thinks gold is overextended and Macro Ops trend work has the dollar flashing sell signals, but the fundamentals still look good
McCracken's defense of running margin: "if everyone's devaluing their currency, then debt is a nice thing to have"
It has worked well for him, and he still asks "what is going to be the straw that breaks the camel's back"
The 60% Drawdown and the Trade That Undid It
The drawdown he took on the peace-deal headlines: "I think my net worth pulled back from the top like 60 percent", on what he corrected to a memorandum of understanding rather than a peace deal
Beylo placed it by the tape — "crude dipped like 65 or 64" — and McCracken put the low at "Yeah, it hit 68, I think"
What he did at the bottom: "I started selling my equity for calls", on the logic that equity falls less than calls rise on a rebound
He liquidated 5 percent of the portfolio and "put all 5% in calls. On oil, mostly USO at the time"
He had flagged it publicly in an angry tweet at the time
His own verdict: "that's probably the ballsiest thing I ever done in my life", and it slingshot him back to new highs
Why he was convinced: he did not believe the memorandum would hold, with Iran raising its demands and getting more vocal
Why He Would Not Sail Through Hormuz, and Nobody Knows How Many Do
On the state of the strait: it is mostly open, but he said you would have to be a madman to take a full tanker through it
"I'd quit my job before I do that", because there is no pattern to who gets hit and Iran hits a random tanker when it needs to
"I'd be terrified to go through there at certain times of the day"
Beylo asked whether there is a price that would get him to do it, and the answer was no — if you die you cannot make more money
The tangent that followed: "Maybe part of Warren Buffett's compounding has just been how long he's lived"
On who is actually being hit: Chinese tankers, everyone's tankers, and "lots of Greek ones, mostly Greek ones", which he attributes to Greek owners being the ones willing to run the risk
He noted that "Centurini's guys didn't get hit"
The scheme he keeps thinking about: going out and setting up a sonar to count transiting tankers himself
"that would be the highest alpha there is", and "if I do it in a canoe or something, the odds anyone blows me up are low"
The flaw is that dropping a sonar could be mistaken for laying a mine, so his fix is a drone boat that drops it instead
Why nobody has done it: "The amount of alpha there, if you know exactly how many tankers are transiting right now" — you would know whether the market is being manipulated
The historical rhyme he reached for: "It's the George Soros thing with the great irony being that his prodigy Scott Bessent is now the one manipulating the market", since Soros made his money betting against manipulators
On the data itself: "I'm actually amazed how bad, the tanker data we have is"
"I'm still convinced at least 4 million barrels a day are shut in globally", which is why he is comfortable holding oil
Memory, Micron and a Local LLM
A position he has started: "I actually bought some Micron on the dip", deliberately small
The reason is personal as much as analytical: he wants his own local LLM, and "I got to drop like 4 million US dollars. And most of the cost is memory"
If he wants one, he assumes independents and small businesses will want one too — "that's a huge amount of demand coming from independents or small businesses"
He raised Jevons paradox by name and said there are scientific advances that would ease the compute bottleneck but not the memory bottleneck
One of them is something he wants to work on and is too busy with stocks to do
Beylo's parallel trade: "that was kind of my whole thesis for like going long like Dell Computer, HPE" on the same build-your-own demand
Where this lands for the AI trade overall: "that's why I think this bubble won't pop", with data centers and robotics both driving commodities
Why he prefers commodities to the frontier: it is "super hard to bet on tech" when Gemini, GPT and Claude are all doing the same thing
"I think like memory and compute are evolving into commodities now"
His forecast: "I think memory has got another leg to its cycle" that will break people who are not long it
Gold, Silver and Where the Cheap Ounces Are
Beylo's read on sentiment: there was so much grave dancing on semiconductors, memory and AI during the pullback that he took it as a sign the trade was not finished
He thinks gold is set up for the same reason in reverse — "Everyone, everyone was like, Oh, it hit 3,900" and called the bottom, which made him unwilling to buy
He was public about it, the price broke the wedge anyway, and Macro Ops got long
McCracken's short answer: "I think gold pulls back a bit here"
On silver he is more constructive: "There's a lot of tech demand coming that I think people don't have covered, like robotic developers"
His two-year price view: "I think a hundred dollars is about to be the floor price. It may not go much higher" — it may simply go there and sit
Why he will not buy physical at this price: what he owns is "that's my old stuff I bought when it was 40 to $50" and functions as a rainy-day fund
For new money he wants silver miners instead, without being sure which are best
On Kuya: "we talked before about Kuya", which he thinks is the cheapest silver miner out there if it executes, and which he barely owns because of the if
What he is hunting for: another miner "trading at 1.5 times enterprise value over cash flow", which he says has become harder as miners get priced more correctly
"Still lots of cheap gold miners", and he wants a dip and a small liquidity crunch to rotate into them
"It seems like both gold and oil want to go up at the same time", which he finds baffling
The M&A Wave the Gold Majors Have Not Started
Beylo's upside risk on gold is waiting for a pullback that never arrives, with mergers as the catalyst
"so many juniors trade at, like, 0.3 or lower times NAV" while the majors generate enormous cash flow at this gold price
"It's going to cost them so much more to find them organically", and the wave has not happened yet
McCracken's explanation for the delay is scar tissue: there has been much more capital discipline, "which is good to see", because companies bought everything in the last cycle and got crushed
Beylo's conclusion: the discipline is what leaves cheap ounces available to a retail investor
The Book He Wants Next: Robotics Plus Miners
Both agreed to record again on robotics, with McCracken saying he has thoughts he has not laid out yet
On the timing: there is going to be a moment to rotate into robotics that is "basically like buying Nvidia in 2022", assuming the companies can get the commodities
The construction he wants: "the best book coming up is going to be some mix of robotics with miners", owning both the materials suppliers and the robotics names capital is flowing into
Beylo's objection is valuation: almost always you have to plug your nose and buy irrespective of value, because the good story and the cheap price do not arrive together
He called Nvidia the exception, since "Nvidia is kind of traded around, like, 10 to 15 times next 12 months earnings" for the whole ride
Nothing in robotics is screaming cheap
The example that made the point: "that Chinese IPO that just happened" — Unitree, whose name McCracken had forgotten despite having met the company at a conference — priced where he wanted no part of it
Beylo's other objection was political: "I'm also just sketched about buying Chinese robotics companies" given the bifurcation, and he would not want a Chinese robot in his house
McCracken's closing joke about that risk: "Or they just turn off and you lose a pile of money. Right. It'll be one or the other"
McCracken's bottom line is that the AI build-out is not the trade — the physical world it forces into existence is, and he would rather own the silver, aluminum, tungsten and oil that a robotics era has to consume than try to pick which model wins.
Products, Companies & Tools Mentioned
Silver miners, and silver itself (The metal he says is already front-running robotics — an ounce per Optimus robot, plus solar and EV batteries; he owns the producers rather than adding physical here)
Brent and WTI crude (The core position — long-dated out-of-the-money Brent calls as insurance against a US ban on WTI exports, with the Brent premium already at double its usual $3)
ROK Resources (Drilling in southeast Saskatchewan, where the royalty holiday cuts the rate on the first 38,000 barrels per well to 2.5% from about 25%)
Micron (Bought on the dip in small size; the memory bottleneck is the reason he thinks the cycle has another leg, and the reason a personal local LLM costs what it does)
EQR and Almonty (The tungsten names both men wish they had bought, on missile demand and Trump's military-economy talk)
Calumet (The calls that drove his rebound to new highs, a name he credits to a pseudonymous account and did not buy from $12 to $27)
Kuya (The silver miner he calls the cheapest out there if it executes, which is why he barely owns any)
Unitree (The Chinese robotics IPO he passed on at the offering price, having met the company at a conference)
Tesla and Optimus (The robot behind his silver arithmetic, via Musk's ambition to put one in every household)
Nvidia (The template for the robotics rotation he expects, and the exception that traded at 10 to 15 times forward earnings through its whole run)
Gemini, GPT and Claude, including Claude Mythos (Why he will not bet on the frontier — they are all doing the same thing — and the models he says are unusually good at finding security gaps)
Dell Computer and HPE (Beylo's own build-your-own-infrastructure trade, on the same demand McCracken sees for memory)
Velo3D and 3D Systems (The 3D printing wreckage Beylo raised as a possible hardware play; McCracken would want to know why the first bubble collapsed)
Millennial Potash, Nutrien, CF and SQM (The fertilizer and potash names Beylo ran through before McCracken explained why he avoids the sector)
USO (Where the 5% of portfolio he converted from equity into calls actually went)
RenTec and Jane Street (The quants he says have come closest to modeling the market as an evolving system, and who still crowd into the same trades)
Books & Resources Mentioned
Tungsten in Peace and War – Ronald H. Limbaugh (Beylo's recommendation for learning the tungsten market, set in wartime and weaving in Bernard Baruch)
Attention Is All You Need (The paper that got transformers going, and with skip connections the whole recipe behind GPT in McCracken's telling)
Gavin McCracken's Substack (Where he wrote up the Saskatchewan royalty holiday)
Get the latest market chatter and takes as they happen:
X | Threads | Instagram | YouTube | TikTok | Facebook