Risk before reward: Q226 letter to partners

Dear partners,

Year-to-date as of June 30 2026, Stone Sentinel Capital (“SSC”) gained 7.6% while the S&P 500 Total Return Index (“S&P TR”) generated 10.2%. Figures are gross returns in USD. Year-to-date returns are not annualized.

SSC is built to generate returns in excess of the S&P TR over the long run. Performance is best judged over 3-5 years.

Hence short-term returns matter little in judging what your manager can do.

Your manager cannot guarantee excess returns but can be certain that investments are chosen on the basis of value, not popularity.

Electing unpopular investments implies that SSC is virtually certain to under-perform the index in certain periods — sometimes over quarters, other times over years.

Favor the disfavored

There are many reasons to favor unpopular businesses.

Disfavored or unknown businesses, by virtue of being unpopular, are not expected to perform well.

Low expectations imply a low bar. An unpopular stock needs little more than decent fundamentals to spur buying to move higher.

When expectations are low enough, the likelihood of fundamentals exceeding them is high.

The languished or stagnant stock price can also find support in fundamentals that are not priced fairly.

Therefore, in unpopular stocks, low expectations anchor high potential returns and under-priced fundamentals reduce downside risk.

The issue with popular stocks is the reverse.

A popular stock is more likely to offer low returns and high risk.

Its rising price means increasing expectations. Fundamentals must be so pristine that a little miss is enough to stop the price from rising.

The rising price also implies that fundamentals are priced excessively. The stock has plenty of room to fall before it finds support in the fair price of its underlying fundamentals.

Therefore, in popular stocks, high expectations drag potential returns and over-priced fundamentals heighten downside risk.

Uneconomic AI arms race

Observers of the markets should be able to see the obvious euphoria in AI-related stocks.

What is less obvious is their level of sustainable profits.

New technology by definition does not have historical precedents. Comparing it to previous evolutions of technology is incomplete at best and misleading at worst.

Spending is also not profit. Goldman Sachs projected roughly $7.6 trillion of cumulative AI capex between 2026 and 2031. The largest hyperscalers (Microsoft, Alphabet, Amazon, Meta, and Oracle) are projected to spend $1 trillion in 2027 alone.

However, it is not clear whether the expenditures have been made in anticipation of future demand or in response to demand already present.

The astronomical expenditures also demand equally high profits. A hypothetical scenario may be illustrative. Given $1 trillion of AI capex for the largest hyperscalers, a 10% return on the spending equates to $100 billion in profit. At a generous 20% profit margin, that profit requires $500 billion in revenues.

The hyperscalers generated a combined $1.67 trillion in revenues last year. Can they generate another 30% of revenues (500 billion / 1.67 trillion) from AI? It is possible but the bar is not low.

What is more realistic is a lower profit margin because of the capital-intensive nature of AI revenues. AI relies on GPUs and memory chips that are short-lived and rapidly obsolescing assets requiring continuous replacement.

If profit margin is assumed at 10%, $1 trillion of revenues would be required to justify a 10% return.

That means the hyperscalers have to generate another 60% of revenues (1 trillion / 1.67 trillion) from AI. The bar is very high.

The hyperscalers understand but ignore the economics. They defend the decision by claiming that the first to the holy grail of AGI is worth the expenditures.

They are henceforth locked in an arms race.

Spending becomes a function of what others spent.

In an arms race, no party can afford to stop because the cost of stopping and falling behind feels higher than the cost of continuing investment. Spending continues even if nobody can yet point to the demand that justifies it.

Funding strains are the most compelling evidence of the arms race. Goldman Sachs estimates that the hyperscalers would consume roughly 94% of operating cash flow on AI infrastructure before any debt financing. As a result, while the hyperscalers are historically self-funded, they are now issuing significant debt to bridge the gap.

Competition-driven spending may persist well beyond the point of economic rationality, resulting in overcapacity and losses, though the relatively short lives of GPUs (2-3 years as opposed to decades for fiber) should limit their duration.

Overcapacity may also be the higher-order consequence of breakthroughs in AI models. Better efficiency reduces server and energy needs, drastically reducing the utilization of existing servers.

There is no doubt that AI will change our world, but there ought to be plenty of doubt as to its economic consequences.

Because the extent of consequences depends on outcomes relative to expectations, not on outcomes alone.

Troubles arise when outcomes disappoint, regardless of their nature.

High expectations are embedded in popular stocks related to AI, heightening the prospects of disappointments and downside risk.

Your manager prefers to do the opposite and bear as little downside risk as possible.

Unpopular stocks are preferred because the low expectations inherent in unpopular stocks reduce the prospects of disappointments, effectively diminishing downside risk.

How to spot a good investor

Your manager weighs economics and value over popularity in judging a stock.

Intrinsic properties matter more than extrinsic signals.

The same ought to be applied in spotting a good investor.

But observers commonly over-weigh returns in judging excellence.

The problem is returns are a lagging indicator of the investor’s performance. Those who generate excellent returns in past decades can still fail.

As the typical disclaimer goes, the past may not reflect the future accurately.

Returns also tend to be volatile. So a long time frame, typically measured in years, is required to smooth the volatile periods to see what true returns are.

So depending solely on returns to spot a good investor is inefficient. You may wait for years for the track record, just for returns to fall after the track record is established.

Your manager proposes an alternative.

To focus on the intrinsic nature of returns more than the extrinsic extent of returns.

The “how” ought to matter much more than the “what”.

The “how” is a leading indicator of the investor. It better shows what an investor would do as opposed to what he has done.

In particular, how the manager survives drawdowns deserves special attention. Does he show conviction and hold existing ideas? Does he hold because the thesis is right or because of a refusal to admit the error? Does he hop to the next idea with little conviction because the high-conviction idea is in a drawdown?

Judging process is more difficult and ambiguous than judging returns. The former involves subjective metrics and analysis, while the latter involves unambiguous numbers.

But we should dive into the hard and ambiguous analysis because process typically leads and returns lag.

So question what you see the next time you witness someone buy the stock of an unprofitable rocket company at 30x sales, sell it at 200x sales, and be crowned a good investor.

In the example, returns are obviously significant. But an observer ought to ask how the investor decides to buy and hold onto the stock.

That profits are made does not mean that risk is absent.

That you survive Russian roulette does not mean that the gun is unloaded.

Some investors may even acknowledge the high downside risk, then defend the investment with the “high reward, high risk” argument.

The trouble is risk may destroy the investor before the reward can materialize.

A large enough loss requires an impossibly large gain to recover.

So considering risk and reward is incomplete.

Risk before reward ought to be the maxim to live by.

The final implication ought to be encouraging to new investors.

While the goal is to generate excellent returns, returns measure who you were.

Process indicates who you are.

You become a good investor before the returns show up.

Portfolio update

SSC is currently invested in 4 stocks that are listed in public markets in Taiwan, Malaysia, United States, and Japan.

Marex, a new long initiated earlier this year, contributed the majority of returns, while Ascentech and Protasco remain in the drawdowns discussed in the previous letter.

Ascentech has performed well as of Q127 (financial, not calendar, quarter). The core business is growing quickly. However, the market is still punishing the stock because headlines still showed year-over-year declines in revenues and profits.

The declines do not reflect the excellent state of the business because they are caused by the one-time surge in orders in the previous year from the business transformation.

Ascentech is still cheap, trading at roughly 8x 2y-forward PE and only 3x cash-adjusted.

Protasco is performing according to expectations as well. The latest quarter showed regular seasonality. The company has submitted the RFP for the new concession, which is likely to be renewed.

Protasco is the cheapest stock in the portfolio with 1.3x trailing PE and 0.3x PB (again, not a typo).

Onto the new long. Marex’s excellent track record attracted your manager’s attention. In the past decade, it increased revenues by a factor of 9 and EPS by 11.

What was more impressive is the increase in returns. ROE increased 5x from 5% a decade ago to 26% last year. Improving capital efficiency is very desirable but rare.

Yet the stock traded only at a low double-digit PE, presumably because the market perceived it as a commoditized mid-market broker.

What Marex really provides is clearing of derivative trades, in addition to standard prime brokerage services.

Clearing is a sleepy back-office function but immensely profitable at 50% PBT margins. In short, before trades are settled and completed, exchanges and affiliated member firms work together to clear the trades. While exchanges ensure the terms of trades are completed as agreed, member firms work with clients to ensure margin balances are sufficient.

But not all clearing firms can make 50% PBT margins. Where the story gets very interesting is the declining competitive intensity of the industry.

All of the largest banks are also the largest clearing firms. However, capital regulations in banking make clearing a low-return business for the banks.

Simply put, large banks have to adhere to minimum leverage ratios (capital / total exposure). The higher the exposure, the higher the capital necessary to absorb losses.

Clearing requires them to hold margin balances for clients, increasing total exposure. To avoid leverage ratios from plunging and crossing regulations, large banks are forced to hold more capital even when clearing poses very low credit risks with short durations.

The capital in cash and liquid securities has to sit on the balance sheet and forego the opportunity to be used for high-returning businesses like investment banking and lending.

So large banks only provide clearing services for their largest clients who pay fees in other areas to make up for the capital-inefficient clearing function. Small and medium clients have to look elsewhere for clearing services.

As a non-bank clearing firm, Marex requires 80-90% less capital than large banks to support clearing. The lower capital costs of clearing allow Marex to generate much higher ROE in clearing to small and medium companies rejected by large banks.

As for smaller clearing firms, they face increasing fixed costs in the form of higher technology and compliance expenses as exchanges upgrade connectivity requirements, latency standards, and risk management protocols.

Compared to smaller clearing firms, Marex has a lower fixed cost per trade for clearing because it has enough scale to be one of the top 10 clearing firms (8 out of the top 10 are large banks).

So Marex gets a natural source of clients from large competitors and wins against smaller competitors. Its margin balances — a key performance indicator — have grown 50% over the past two years, against low-to-mid-single-digit growth at large banks.

Trading clients cannot bypass clearing. If they trade with a non-clearing firm, they have to hold margin balances at both non-clearing and clearing firms. And trading with a non-clearing firm also entails higher margin costs because they are unable to utilize cross-margining and other strategies involving position offsets that clearing peers use. This dynamic means clearing firms are structurally advantaged relative to non-clearing peers.

Marex traced its roots to the bankrupt Refco and was initially capitalized by Marathon Asset Management before private equity took control. Its PE owners, largely of Lehman descent, built Marex through a series of M&A before taking it public in 2024.

The PE owners still own roughly 9% but should pare their stakes over time. CEO Ian Lowitt, former CFO of Lehman, owns a 3.6% stake. The heads of various segments at Marex have led the company as CEO at different stages of its evolution. Senior management has deep knowledge of the entire business.

Marex stock has increased by 60% year-to-date but is still cheap and disfavored at only roughly 12x forward PE. The market is still giving insufficient credit to its structurally advantaged position against large and small competitors and non-clearing peers.

Your manager welcomes questions and comments at mg@stonescap.com.

At your service,

Marcel Gozali