Writing a thesis
You wrote down growth. The thesis was about power.
Revenue is an output. What produces it is who can raise prices on whom, and that sits in a different line of the same filing.
What people say they own it for
What they are actually betting on
“Power is the real bottleneck on AI now, not chips.”
Supplier power, and whoever owns the scarce input keeps the margin. If the bottleneck is real, the rents move to the constrained layer and away from everyone assembling on top of it.
“The platform vendors keep taking share from the point tools.”
Switching costs. Share gains from consolidation are durable only if leaving is expensive; if it is not, the same logic reverses the day a cheaper bundle appears.
“They are the obvious supplier to the hyperscalers.”
Buyer power, in the least favorable direction. Four possible customers, each large enough to design you out, is a structural position no revenue line will describe.
“Scale means they can undercut everyone on price.”
Scale economies, which pay only where cost per unit actually falls with volume. In businesses where it does not, scale buys revenue and no advantage at all.
“They have a moat. Nobody switches once embedded.”
Switching costs again, and this is the one most often asserted and least often measured. It has a number, and companies that have it tend to publish it.
Every sentence on the left is a forecast about demand. Every one on the right is a question about who captures the value that demand creates, and the two come apart constantly.
An industry can grow at thirty percent for a decade while every participant in it earns nothing, because the growth accrued to whoever held the scarce input. A thesis tracking only the demand cannot tell those two worlds apart.
The reframe
The hard part is selection, not phrasing
Anybody can put a number on a sentence. Choosing which sentence deserves the number is the whole job.
An investment thesis is the case for owning a position, written before you own it, in terms specific enough that a document you have not read yet could settle it. That much is mechanical. The difficulty is that a business throws off dozens of settleable numbers and almost all of them are consequences rather than causes.
Revenue growth is a consequence. Margin is mostly a consequence. What causes both is structural: whether the company can raise prices without losing the customer, whether its suppliers can raise prices on it, whether leaving costs the customer anything, and whether being bigger makes it cheaper to operate. Those four decide who keeps the money a market produces, and unlike a growth rate they change slowly and visibly.
Which makes them the right subject for a claim. A claim about growth tells you the weather. A claim about power tells you the climate, and you will still be holding the position in three years when the climate is what mattered.
A fund's memo carries three of these and almost nothing else, beside what the market believes instead and what the desk will do if one breaks. The whole memo, and the three documents written against it afterwards, is on the hedge fund process, for one person.
The four
Where the power sits, and where it is disclosed
Each of these is slow, structural, and reported. That combination is what makes a claim worth carrying for years.
Buyer power
How much of the price the customer sets. Concentrated buyers extract terms long before they extract price, and the terms show up first.
Shifting against you looks like. One customer growing as a share of revenue, receivable days lengthening, rebates and incentives appearing in the revenue recognition note, and pricing described as a headwind nobody chose.
Disclosed in: the customer concentration note in the 10-K, days sales outstanding, and the discounts and incentives language in revenue recognition
Supplier power
Whether somebody upstream can take the margin. When a single input is genuinely scarce, the profit pools at that layer and the assemblers below it work for wages.
Shifting against you looks like. Gross margin falling while volumes rise, prepayments and long-term supply commitments growing, and a purchase obligations table that gets longer every year.
Disclosed in: gross margin against input cost commentary, the purchase obligations table, and prepayments in the cash flow statement
Switching costs
What leaving costs the customer. This is the most asserted and least measured advantage in equity research, and it is measurable.
Shifting against you looks like. Net revenue retention drifting toward one hundred, gross retention falling at all, and a company that used to disclose both quietly reporting only the flattering one.
Disclosed in: net and gross revenue retention, disclosed by most subscription businesses and conspicuous by absence
Scale economies
Whether being bigger makes each unit cheaper. Real in semiconductors and logistics, largely absent in services, and assumed everywhere.
Shifting against you looks like. Incremental margin flat or falling as revenue grows, which is the direct evidence that the next unit costs what the last one did.
Disclosed in: the change in operating income against the change in revenue, computed across several years rather than one
None of these requires a model. They require reading four specific places in documents the company is obliged to publish, on a schedule it does not control, and noticing when one of them moves against the thing you assumed.
Side by side
One belief, two claims, and only one of them tells you anything
The belief is that a component supplier benefits from a buildout. Both claims below are specific, dated and settleable. They behave completely differently.
The output claim
Segment revenue grows at least 25% year over year through the next two prints.
- Can be met while the business decays. Volume from one customer buying ahead of its own capacity, at a discount that does not appear in this line.
- Can be missed while the business improves. A deliberate exit from low-margin work looks identical to losing it.
The power claim
No single customer exceeds 30% of segment revenue, and gross margin holds at or above 34% while that revenue grows.
- The pair is the point. Growth with margin intact means price is being held. Growth with margin falling means volume is being bought, and the buyer is the one setting terms.
- Concentration is the cleanest reading. It is buyer power stated as a number, disclosed once a year, and it moves before the pricing does.
The judgment
Four things a checklist will not settle
Every one of these arrives eventually, and deciding the rule while nothing is happening is the only cheap moment to do it.
The claim was 30%. The filing says 28.4%.
Nothing about a threshold tells you what a near miss means, and the moment it happens you will have a reason ready for why this one does not count. Decide the rule before the number exists: either the threshold is a line and 28.4 breaks it, or you write a band and a consecutive-periods rule into the claim itself.
The second is usually more honest. Most structural claims are about direction and persistence rather than a single print, so "below 30% in two consecutive filings" states what you actually believe. A single-period threshold you intend to forgive is a threshold you have already forgiven.
The variable that matters is not disclosed.
This is the common case, not the exception. Pricing power rarely appears as a line, and a company under pressure discloses less rather than more. You have two options and one of them is worse than it looks.
Pick a proxy and write down that it is a proxy, along with what would make you stop trusting it. Gross margin stands in for pricing power until a mix shift breaks the link. The worse option is switching the claim to something that IS disclosed and quietly forgetting it was never the thing you cared about, which is how a thesis ends up tracking a number nobody would have chosen.
There is a third answer worth taking seriously: if the deciding variable is undisclosed and has no honest proxy, that is information about the position rather than about your process.
It broke, but the market already re-rated.
A broken claim and an attractive price are separate facts and both can be true. The question is not whether you were wrong, which is settled, but whether the price now assumes something worse than what the filing showed.
That requires a new case rather than a defense of the old one, and it deserves the same treatment: a claim, a number, a date. Written as a fresh thesis it is a legitimate second position. Written as a reason to keep holding, it is the first substitution in the sequence the drift page describes.
The claim is falsifiable and economically irrelevant.
The most common failure among people who take this seriously. A claim can be perfectly specific, settled quarterly, and make no difference to the outcome: unit shipments in a segment that contributes four percent of profit, a metric that moves with the cycle rather than with the company.
The test is a counterfactual. If this claim is missed and everything else holds, does the case change? If the honest answer is no, the claim is instrumentation rather than thesis, and tracking it is how attention gets spent on the wrong quarter.
Questions
Common questions
How do you write an investment thesis?
Decide what the position actually depends on, which is usually a question of bargaining power rather than growth: whether the company can hold price, whether a supplier can take the margin, whether leaving costs the customer anything, whether scale lowers unit cost. Write that as a claim with a threshold and a date tied to a disclosure that will settle it, and write what you will do if it breaks. Two or three claims is enough. The phrasing is mechanical; the choice of subject is the work.
What should an investment thesis actually track?
Causes rather than consequences. Revenue growth and margin are outputs of a structural position, so a claim about them can be met while the business decays and missed while it improves. Claims about customer concentration, retention, incremental margin and input cost pass-through describe the position itself, change slowly, and are disclosed on a schedule the company does not control.
How many claims should a thesis have?
Two or three, each testing a different force. A set of claims that are three phrasings of one number is one claim wearing three hats and breaks all at once or not at all. A growth claim paired with a margin claim is informative precisely because they can come apart, and the case where they diverge is the one that tells you who is setting price.
What do you do when a claim is ambiguously missed?
Decide the rule before the number exists. Either the threshold is a hard line, or the claim itself carries a band and a consecutive-periods condition. Most structural claims are about persistence rather than a single print, so "below the threshold in two consecutive filings" usually states the real belief. A threshold you intend to forgive on the day is one you have already forgiven.
What if the thing that matters is not disclosed?
Choose a proxy and record that it is one, with the condition that would make you stop trusting it. Gross margin substitutes for pricing power until a mix shift breaks the link. The failure mode is switching the claim to whatever happens to be disclosed and forgetting it was never the thing you cared about. If the deciding variable has no honest proxy at all, that is a fact about the position worth weighing.
Is a broken thesis still a sell if the stock already fell?
Those are two separate facts. The claim is settled and the price has moved, and what remains is whether today's price assumes something worse than the filing showed. Answering yes requires a new case written to the same standard, a claim with a number and a date. A new case is a legitimate position; the same argument offered as a reason to keep holding is the first substitution in a chain.
The product
Where Solon sits
Selection is yours. The reading is the part that does not fit in a week.
Nothing above needs software. Choosing which force your position actually depends on is judgment, and it is the part worth your afternoon. What does not fit in an afternoon is the rest of it: a concentration note once a year, a margin line four times a year, a retention disclosure when they feel like giving one, across a dozen positions, forever.
Solon takes the claims you settled on and reads filings, prints, headlines and rival moves against them each night, quoting the line it read. When one is pressed it says so, with the plan you wrote in advance beside it. What happens the morning that lands is on when your investment thesis breaks, and what to demand of any tool doing this job, including this one, is on investment thesis tracker.
The practice this belongs to is investment thesis monitoring, and what happens to a claim nobody checks again is thesis drift: the same position held for four different reasons in eleven months.
You know which claim to write. The reading is the part that does not fit.
A concentration note once a year across a dozen positions is the work Solon does after you choose.
Already invited? Create your account