Top Mark Capital · 2026 Q2

Partner Letter

There Will Be Blood.

Dear Partners,

In the post large language model world, it is very easy to produce content. I’ve witnessed it personally and I’m sure you have as well. We value you and your time, so we are explicitly making our letters as brief as possible while also communicating everything that is important. Your feedback is greatly appreciated.

The cover image and the topic of this letter are inspired by an infamous scene from the 2007 film There Will Be Blood.¹ But first, Anthropic’s quarter. While we do not yet have full Q2 data, the company reportedly added nearly the entire enterprise value of Illumina in revenue during the first two months.

Anthropic x OpenAI Run Rate Revenue

Annualized run-rate revenue, estimated, not GAAP. Source: Epoch AI, AI Companies Revenue Reports.²

Meanwhile, models are converging, if not commoditizing, so tech companies do what tech companies do: they differentiate. Differentiation for an AI lab means integrating models into the application layer, and Anthropic’s revenue is the reward for being first to take that leap.

Claude Code³ was a sharper version of an application its own partner Cursor had built on Anthropic’s models. Claude Cowork followed. Then Claude Design, pointed at Figma, Canva, and Adobe. Three days before Claude Design shipped, Anthropic’s chief product officer resigned from Figma’s board; Figma’s chief executive said Anthropic had not been consistently candid.⁴ That scene raised eyebrows among Anthropic’s customers.

What came next stood the hair up on the backs of their necks. With its mid-year Fable 5 model release, Anthropic began retaining usage data for thirty days even on plans that had promised zero data retention, on safety grounds, without a binding guarantee against training on it later.⁵

Near the end of the film that inspired our cover art, the oilman Daniel Plainview tells a rival that the oil beneath the man’s land is already gone. He explains drainage: if you have a milkshake, and I have a milkshake, and I have a straw, and my straw reaches across the room, then I drink your milkshake. I drink it up. He sank a well at the edge of his own land and drained the oil from beneath his neighbor.

The AI labs sit where Plainview sat. Absent a zero-data-retention policy, the straw takes up every request an enterprise sends a model, and far more when the work runs through an application-layer tool.

So the useful question for a concentrated, long-term investor is not whether a company “has a moat.” It is what the barrier is made of, and whether the straw weakens it or strengthens it.

Charlie Munger’s favorite mental model was invert, always invert. Invert for what gets stronger: which moats does the straw improve?

Quickly, a definition. I do not care much for the word moat, but a liquid fortress is apt given this straw. By moat we mean the collection of Powers that lets a business sustain economic returns above its cost of capital, over the long run. Helmer’s discipline is that a Power needs a benefit, which improves cash flow, and a barrier, which is why a well-funded rival does not simply copy you.⁶ The benefit shows up in the numbers; the barrier is the half most forget. So the better question is never whether a business has a moat. It is: what is the barrier made of?

That leads to a thought experiment. Write down the secret sauce: the processes, the patents, the org charts, the playbooks. Hand the file to the fiercest competitor. What was lost? Call it the drainage test. The straw drains knowledge. A barrier made of nothing but undisclosed knowledge is already gone. It was a secret, and a secret was never a moat.

The straw improves a barrier when better models raise the value of something only the owner has: private feedback, earned judgment, a process that generates ground truth a rival cannot rent.

For an investor, the scarce input used to be throughput: how much you could read and hold. That pool is draining. What becomes more valuable as intelligence commoditizes is taste: the felt sense of what is good before you can fully explain why. Taste cannot be prompted into existence. It is built through reps and through being wrong, inside a circle of competence you have earned, not one you have declared.

Financial firms have always dealt in intelligence, so they know intelligence alone is not a moat, it’s a prerequisite. Ken Griffin has argued that ‘calling the quarter’ is being competed away by alternative data and AI, and that what survives is vision about which companies change society over years, not quarters.⁷ That is taste. We agree.

One last inversion: what Power do the AI labs hold? Weights depreciate in months. The leaders climb the same curve and open weights are just months behind. The climb into applications isn’t the reward for seeing the future first. It is the only place a lab can go to manufacture a barrier. The reason Plainview is drinking your milkshake is that he doesn’t have one.

The labs know this, which is why the fight over distillation is so bitter: a rival can point its own straw at a frontier model’s interface, train on the answers, and walk off with much of the capability for a fraction of the original bill. They call it theft, but the “learning is transformative” defense used to train on the open internet is essentially the same argument. The drainer discovers he too can be drained.

We are at a fork in the road. One path will concentrate the value captured by artificial intelligence amongst a chosen few, while the other will democratize its value across society. We hope the latter wins out.

Our largest position by a wide margin, held across both TMCP and TMHP, is deliberately not an AI story. It is a specialty operator whose edge is its own and that the market is not chasing. That is much of the attraction. We added to that position in the teeth of the first-quarter drawdown⁸ because the price moved and the theses did not.

What would tell us we are genuinely wrong is not the price. It is the owner economics of these businesses breaking against our expectations. That is what we will report to you, and it is what to hold us to: not the quarterly mark, which tells you what the crowd is chasing this month, but whether the barriers we thought we owned still look like barriers a year from now under the drainage test. We may be wrong about how fast the drainage occurs. We do not want to be wrong about what kind of barrier we thought we owned. Put all your eggs in one basket, as Andrew Carnegie said, and then watch that basket.⁹

Our results reflect a deliberate process. Some investments may take five years or more, others may appreciate sooner. A rare few may be held indefinitely. Along the way, there will be adventures. Detours and damage alike. The question is never whether they will happen, but whether we bought right and whether we have the patience to continue on. We invest for outcomes measured in years, not quarters. In the near term, our results are as likely to be poor as strong. Over time, we believe our results will prove satisfactory.

If you share our temperament and long-term perspective, we invite you to consider joining us as a Partner.

Mike & Jason

This letter is published by Top Mark Capital as an advertisement and a solicitation of interest in a private offering conducted under Rule 506(c) of Regulation D, intended only for accredited investors who are also qualified clients (a higher standard that applies because the partnerships charge a performance fee). Top Mark Capital Partners (TMCP) and Top Mark Health Partners (TMHP) are offered solely through definitive offering documents and only after the firm takes reasonable steps to verify accredited status; no subscription will be accepted prior to completion of that verification. This letter reflects the authors’ views as of the date of publication, is not personalized investment advice, and is not an offer or sale in any jurisdiction where such offer or sale would be unlawful. Any companies mentioned are discussed to illustrate our process and market views; they are not recommendations to buy or sell any security, and the firm and its partners may hold positions in them. Registration with the SEC does not imply a certain level of skill or training. See topmarkcapital.com/terms.

Top Mark Capital Management, 416 13th St, 2nd Floor, San Diego, CA 92101 · (646) 470-3016