Today I wanted to share with you an excerpt from the Fall 2026 letter I wrote to IMA clients. If you find this essay educational, share it with friends, enemies, random strangers. They can also subscribe here.
Oil, Commodities, AI, and Cyclicality
We have never been big fans of oil or commodity stocks; in fact, we have owned them sparingly throughout IMA history, and for a good reason: we need to have an opinion on a commodity to value them.
My view on them has changed. Though I still don’t like the fact that I have to guess what the commodity price is going to be, I got to a semi-comfortable level with them after I married two mental models: capital cycles and antifragility.
Capital cycles
Let’s start with capital cycles. This mental model was created by Marathon Asset Management, a London-based investment firm, and popularized by my friend Edward Chancellor in his book Capital Account.
When analyzing cyclical businesses, investors usually focus on demand, not supply. But demand and its timing are difficult to analyze. Supply is often easier: you just need to count how many mines are built, oil rigs operating, or capital spent on exploration.
Also, there is a substantial lag in supply, things happen there not at the speed of electrons but at the speed of atoms, much slower. A commodity has to be found, permits approved, capital raised, equipment secured, people hired, and so on. It happens slowly, but then capacity arrives all at once.
Each company makes an individually rational decision, which, on an industry level, is irrational and results in supply exceeding demand. Eventually supply overshoots demand (this always happens, no matter how great and convincing the “this time is different” story is, it is just a matter of when).
If every company in the industry got together and tried to match production to demand, capital cycles would rarely happen (short of a sudden demand shock, the likes of a pandemic). But this is not how the world works. Even cartels like OPEC don’t work well (incentives to cheat are too great). Thus you have cycles.
Cycles are observed not just in commodities like oil, gold, natural gas, or iron ore, but also in data centers and fiber optics.
A detour: the AI capital cycle
Yes, you know where my keyboard is taking me: the AI (data center) overcapacity bubble. Okay, just a small detour. I have to get this off my fingers.
In the good old times, three years ago. You had five great companies: Google, Meta, Amazon, Apple, and Microsoft. They all dominated their enormous niches and, maybe outside of the cloud business, did not really compete that much with each other.
Then came AI. Apple decided to wait and see how this whole AI thing plays out. If you have been using Siri (part of what Apple calls “Apple Intelligence”), you are aware of this: it has the intelligence of a fruit.
The gang of four have been joined by three other trillion-dollar (for now), well-capitalized companies: SpaceX, OpenAI, and Anthropic.
Each one looks at AI as an existential risk. FOMO (fear of missing out) is replaced by FOBI (fear of being irrelevant). Each company does a rational thing: invests as if its life depends on it. In isolation, a very rational decision. As a sum, they are running a very significant risk of creating enormous overcapacity.
Also, they are suddenly fighting each other almost head to head, and when this much capital bumps against each other, future profits will come at lower margins.
Their new adventure into AI also made their businesses more capital intensive. Most of them (with the exception of Amazon) were capital light. Now the incremental income is lower, and the “capital” is higher. Thus return on capital, and thus future valuations, will likely be lower: they are not as good businesses as they once were.
They also went from net-cash balance sheets and buying back stock to being indebted entities that keep borrowing (and, in some cases, issuing stock).
What makes this analysis difficult is that nobody knows what demand is going to be and its timing, and the pricing is another big question mark.
Paul Kedrosky makes an interesting point on current demand: today AI is mostly used by early adopters. I read a number that 70% of it is used by 1% of users. Well, these 1% of users are mostly software engineers, and thus use it very differently than Joe the Lemonade Stand would use it.
When you are writing software, you write a few paragraphs, something along the lines of “create software that does this” (I am oversimplifying), and then a week later you have a few million lines of code, and a lot of tokens have been killed to create it.
A normal user throws in Anthropic’s S-1 (going-public filing) and asks it to provide key takeaways; this is a reductionist use of AI and uses a small number of tokens.
The problem with this conversation that my keyboard is having with you is that we are closer to a sci-fi world today than we have been for generations, and thus it is very difficult to know what the future will look like. Thus demand may end up greater and faster than cynics like my keyboard postulate here.
But the risk is very clear and real: trillions of dollars are being spent on the supply side, while the demand side is uncertain. This is what would keep me up at night if I owned these stocks.
Back to oil
Okay, enough with the detour. Back to our low-sci-fi oil stocks.
Over the last two decades, oil companies have been vilified, and capital has been redirected by institutional imperatives to green investments. This year, $2.2 trillion will go to renewables, grids, and other clean energy, and less than $500 billion to oil supply. Oil investment is falling for the third year in a row.
I want to have a green planet as much as the next guy who burns AI tokens by the truckload, but these green investments had questionable rates of return and have a lot of other problems, which I discussed in the past (“The Church of Climate and the Law of Unintended Consequences”).
The key is that the supply of oil is likely to be constrained for a long time, while demand has not really changed much.
Also, suffering from PTSD from the abuse by politicians and the public, (evil) Big Oil would rather be buying back stock and paying dividends than putting capital to work while taking the risk that its windfall profits will be taxed to death.
The industry now reinvests only about half of its operating cash flow and returns most of the rest to shareholders. Output from the big oil companies is expected to stay flat or grow in the low single digits through the end of the decade.
So capital has not rushed into the industry, yet. But it will at some point if oil prices stay high for a long period of time.
Antifragility
Now, it took me a while to get here, but there is a second mental model: antifragility. We have to thank Nassim Taleb for it.
Think of fragile goods or systems first. You send wine glasses by mail and pray that they don’t break. They are fragile. Adversity hurts them.
Robust systems are indifferent to adversity: rubber wine glasses.
And then you have antifragile systems: they grow stronger from adversity.
In general, cycles make cyclical businesses fragile. Management does procyclical things: it buys high (drunk on the Kool-Aid that this cycle is different) and sells low, trying to repair an overextended balance sheet when prices fall, demand declines, and the never-ending fountain of cash flows turns into a trickle.
Some companies come out of both sides of the cycle robust, which is already a victory. But then there are a few that come out stronger, they take advantage of the fragility of others by doing the opposite of what everyone else is doing.
It is incredibly difficult, it requires the right culture and shareholders and strong conviction. This goes somewhat against our genetic programming, as we seek comfort in groups of like-minded people (tribes).
The most difficult part of being a contrarian (an independent thinker who follows their research when it goes contrary to the popular opinion of the moment) is timing. Being early is often perceived as being wrong. It is, and it is not.
I remember a joke about a guy who wakes up in the morning with a giant hangover and explains to his wife how he got wasted.
“I drank a bottle of scotch. Nothing. Did not feel anything. Then I tried out a new bottle of tequila a friend gave me for Christmas. Nothing. Friends came and we had a large bottle of Smirnoff. Then everyone left and I had just a bottle of beer, and I got drunk. Can you imagine, just one bottle of beer?”
It is hard to know what is going to be that bottle of beer that will turn the drinking binge into a giant hangover. It is especially hard to be the sober one, observing everyone around you liquoring up.
Our oil stocks
But back to oil and antifragility. The oil industry is full of drunks and a few recovering alcoholics. Finding sober heads that will do the opposite of what everyone else is doing when the time comes is incredibly difficult. This is also why we have a hard time finding mining companies.
We think we own exceptional, antifragile oil companies. Most importantly, they are run by exceptional management teams; it took us a while to find them, mainly by studying management. Two of them are controlled by the same family.
Oil prices have been elevated by the war in Iran and, marginally, by Ukraine doing God’s work and finally making this war painful for the average Russian consumer by destroying Russian oil infrastructure. (I have a certain bias here.)
In August, more than 10 million barrels a day of Gulf production was shut in. Brent is about $90 today; over the past year it ranged from about $59 to $119.
We are using $80 in our models and still get rates of return of 11–14% for our oil stocks.
We have reduced our position in this industry patiently, and will continue to do so into strength. Still, as of right now, we look at the oil companies we own as core holdings.
Read next: Skin in the Game, Steam in the Ground, my visit to International Petroleum’s Blackrod oil sands project and why the Lundin family’s countercyclical capital allocation is what antifragility looks like in practice.You may also like Why We Bought Aker BP: A Technology First Oil Company with a single-basin focus on the Norwegian Continental Shelf.
Key takeaways
- Investors analyzing cyclical businesses focus on demand, which is hard. Supply is easier: count the mines, the rigs, the exploration capital. Supply moves not at the speed of electrons but at the speed of atoms — slowly, and then it all arrives at once.
- FOMO has been replaced by FOBI, fear of being irrelevant. Each of these companies invests as if its life depends on it, which in isolation is perfectly rational and in aggregate risks enormous overcapacity. They were capital-light businesses with net cash; now they are capital-intensive and borrowing. They are not as good businesses as they once were.
- The AI demand question is genuinely open. Today 70% of usage comes from 1% of users, mostly software engineers burning tokens by the truckload. The rest of the world throws in a filing and asks for key takeaways. But we are closer to a sci-fi world than we have been in generations, so demand may yet outrun the cynics.
- Capital has been shamed out of oil — $2.2 trillion into clean energy this year against less than $500 billion into oil supply, with oil investment falling for a third straight year. Big Oil, suffering PTSD from politicians and the public, would rather buy back stock than risk another windfall tax. Supply stays constrained; demand hasn’t changed much.
- The oil industry is full of drunks and a few recovering alcoholics. Cycles make cyclical businesses fragile; antifragile ones grow stronger from the adversity, by doing the opposite of everyone else at the moment it is hardest to do. The hard part is timing — being early looks exactly like being wrong, and nobody knows which bottle of beer ends the binge.





comments
0