Kirkland's AI bet is being read as a technology story, and underneath that as a proprietary data story. It is neither. It is the clearest signal yet that the most profitable firm in the world cannot buy its way out of the one problem every firm shares. The questions worth asking are not the ones being asked.

The TL;DR, for partners

$500 million is not the story. At Kirkland's size it does not hurt, and it does not make them different.

Your firm's "proprietary data" is not a moat. Every M&A practice has deal data; every employment shop has employment data. It is unique the way your grandmother's cookie recipe is unique, and it still just makes cookies.

The only real moat is your genome: how you go to market, how you hold a client, the judgment that never shows up in a document. That asset is yours, not the firm's.

Which is the whole problem. Kirkland is spending half a billion to capture the one thing it cannot buy, cannot encode, and that you would never willingly surrender. And while it tries, AI is quietly eating the leveraged base that funds your draw, not the judgment that earns it.

The pyramid is dying. The partner who owns the genome is not.

The one question nobody is asking: on what terms would you ever agree to be assimilated?

THE FULL BRIEF

Kirkland & Ellis has earmarked $500 million to build its own AI platform. The reaction has been predictable: a wave of hot takes, a few victory laps for "build, not buy," and a great deal of opining dressed up as analysis. Almost none of it asks the questions that actually matter.

I do not have all the answers here. I want to be honest about that up front, because anyone telling you they know how this resolves is selling something. But I have spent enough time inside these conversations, advising elite firms and building AI-driven service lines with an Am Law 100 firm and inhouse teams, to be confident about one thing: at this stage, asking the right questions is worth more than guessing at the answers. Let me try to ask better ones.

First, kill the premise everyone is operating on

Before we talk about the money, we have to dispose of the idea quietly underwriting almost every take on this, including the ones that sound sophisticated: that a law firm's proprietary data is a moat. It is not, and getting this wrong corrupts everything that follows.

Take my grandmother's chocolate chip cookie recipe. It is hers. Nobody else has it. And it is, technically, slightly different from every other chocolate chip cookie recipe out there. It still makes chocolate chip cookies.

Law firm data is proprietary in exactly that limited sense. A firm's matter history, its client documents, its internal work product: sure, nobody else has that exact dataset. But zoom out. Every firm doing employment law has employment data. Every M&A practice has deal data. The problems being solved inside a practice area are fundamentally the same from firm to firm, which means the data reflecting those problems is far more homogenized than people want to admit. Unique? Yes. A genuine moat? No. It is homogeny wearing the costume of differentiation.

Ok, if the data is not the defensible advantage in an LLM world, what is?

It is the genome, and the genome is not the data. A practice genome is partner specific. It is how a particular partner goes to market. It is their behavior and their posture in front of clients. It is the singular contribution they make to a client relationship and the way they cultivate and sustain it over years. That is the holy grail, and it is the precise opposite of homogenous, because every partner is one of one. The data corpus drifts toward the industry mean. The genome does not.

Hold onto that distinction, because it sets the trap sitting underneath the entire $500 million. The asset that encodes cleanly into a model is the homogenous one, the part that was never a moat. The asset that is a moat, the practice/partner genome, does not encode, because it lives in behavior, relationship, and judgment in context rather than in files. A firm can spend a fortune digitizing the convergent layer and call the output proprietary. The real differentiator stays exactly where it has always been: inside individual partners, and only there.

The economics waiting in the wings

Now the number, because the number is where everyone else stops.

$500 million sounds enormous, and in absolute terms it is. But it is committed over three to four years, which puts it at roughly $125 to $165 million a year, starting near $100 million this year and rising, against a firm that did $10.6 billion in revenue last year. Jon Ballis, Kirkland's chair, has said the firm should invest about 1 percent of revenue in new initiatives. This runs a touch above that, but only a touch, and well inside what the firm absorbs without blinking. Read this is as not a bet-the-firm gamble. It is the standing innovation budget, pointed at one thing and announced loudly. Profit per equity partner, $11.1 million, does not move because of this. The spend does not hurt.

That matters more than it appears, because a painless bet cannot be read as an act of conviction. If it costs them nothing it would appear but it’s a big number that can strike fear in competitors or serve as signal to talent. My guess Kirkland is aware of but not reporting the real cost that is hiding somewhere other than the cash. And it is.

Look at what is not in the $500 million: the partner hours. Kirkland is building this with input from roughly 250 lawyers, including 100 partners. Partner attention is the single most expensive resource in that building, and none of it appears in the headline figure, so the number is already an undercount of the true investment. But here is the part nobody says out loud. Those are the most profitable hours in the firm, and they are being spent to build a system whose purpose is to make those hours less necessary. The firm is investing its leverage in order to reduce its dependence on leverage. Yes, read that line again.

Which leads to the point almost no one is making (as it stares us in the face), and the one that actually matters: leverage is the profit.

A Kirkland partner does not earn $11 million from their own billable hours. They earn it from the spread, the margin on leveraged associate hours, the high-volume base-of-pyramid work: fund formation, compliance reviews, due diligence, portfolio-company support, billed at $500 to $700 an hour by people who cost the firm far less. That pyramid is the profit engine. And AI compresses exactly that base. Not the bet-the-company negotiation. Not the novel regulatory question. The base.

Now follow where the compression is coming from. In November 2025, Blackstone, plausibly the largest single buyer of elite legal services in the world, invested $50 million in Norm AI and helped launch Norm Law, an AI-native firm aimed squarely at the institutional, high-volume work Kirkland considers its own. Norm Law has no incumbent profit to protect. AI does the production, a thin layer of elite lawyers does the judgment, and there is no associate class whose billings must be defended. It can match a cost structure Kirkland cannot, because Kirkland's profitability is the structure being undercut.

This gives us a sharper picture into something uncomfortable. Kirkland's profit engine and the work being siphoned off are the same work, and the entity doing the siphoning is being built and funded by Kirkland's most important client.

Watch one number over the next few years. Blackstone's disclosed fees to Kirkland were $101.3 million in 2024 and $87.8 million in 2025. One year is noise; deal flow moves around. But it is a number worth watching, because the most sophisticated buyer of legal services on earth is now running the same play on its legal spend that it runs on everything else. It is, in effect, applying leveraged-buyout logic to its own outside counsel: fund the challenger, pressure the incumbent, and capture the surplus from both ends.

Defensive, not offensive

The dominant read of this move is that Kirkland, being Kirkland, is going on offense. I think that is exactly backwards, and the error is a halo effect: because they are number one, we assume any move comes from strength and ambition.

But the largest, most leveraged, highest-volume firm in the world also has the most exposed flank. The size of the firm is the size of the target. This is the market leader defending its largest vulnerable surface, not a challenger attacking an opening.

The precedent is their own. A decade ago Kirkland built CTRAN, a proprietary database of its corporate transactions that let the firm know what was market on deal terms better than anyone, because its own deal flow effectively is the private M&A market. They never sold it. They used it to defend and deepen the core franchise, and it is arguably part of why they became the largest firm in the world.

But notice what CTRAN actually captured: deal terms. What is market. That is precisely the convergent, homogenous layer from the cookie problem. It was a real advantage in its day, the first-mover edge of having systematized what was market before anyone else, and part of how they grew. But a first-mover edge is not a durable moat, and this is exactly the kind of edge AI commoditizes fastest, because what is market is by definition a shared fact the entire market is converging on. CTRAN never captured the genome, and it never could. The $500 million platform is CTRAN at scale, and the quiet risk is that scale only deepens the firm's grip on the layer that was never the moat to begin with.

I will concede the obvious objection. By owning the technology outright, Kirkland buys a call option on offense, the freedom to commercialize later if it chooses. That optionality is real. Some folks have pointed this out. But firms fail at being software companies. And the primary driver is defense anyway. The honest framing is defensive now, with a call option on offense, and the firm that confuses the two will misread why it is spending the money.

Why announce it at all?

Here is a question almost nobody is asking. Why announce this?

The spend is too small to require PR cover. The capability does not differentiate them; Fried Frank, Linklaters, A&O Shearman, and Dentons are all building versions of this. Proprietary legal AI is becoming table stakes, not an edge. The announcement is doing work the spend itself is not.

Part of it is competitive signaling, and this is the one genuinely offensive use of a defensive build: we can commit half a billion dollars without flinching, can you? For rivals on thinner margins, the same spend would hit distributions. The announcement invites them into a race many of them structurally cannot run.

But the most revealing reading inverts the conventional wisdom. Everyone assumes a $500 million AI program is a recruiting magnet, a way to say come build the biggest thing in legal AI. Look at it from the chair of the lawyer being recruited and the offer reads differently: come let us assimilate your genome. And remember what the genome is. It is not the firm's homogenous data, which the firm already has. It is the partner's own, the singular thing that constitutes their individual market value. The pitch, decoded, is an invitation to pour the one asset that is actually yours into a model the firm owns, after which your individual indispensability falls. You would be funding, with your own leverage, the thing that reduces it.

There is a second twist that makes it worse for the firm, not better for the partner. The genome cannot be extracted cleanly from documents anyway, which means the firm needs the partner's active, willing participation to capture it at all. That is precisely what a rational partner withholds. This means the recruiting signal cuts the wrong way twice: it threatens the senior people it is meant to attract, and it depends on a cooperation those same people have every incentive to refuse.

Now set that beside the competing offer. When Mike Schmidtberger, who chaired a $4 billion firm, joins an AI-native firm with a handful of lawyers, that is a conviction trade, and part of what makes it one is equity. Norm Law can offer upside. A partnership structurally cannot. Meaning Kirkland's announcement, read honestly, is a weaker recruiting pitch than it looks. It advertises assimilation to exactly the senior people who now have somewhere better, and more ownable, to go.

Not an event. An admission.

It is a rational defensive move. It is also aimed at the wrong layer, the capturable data that was never the thing worth defending, which is why it is not actually an answer. This is where I would push every firm leader to reframe what they are looking at.

The media is covering the $500 million as an event, as though Kirkland did a big new thing. It is not an event. It is an admission. The number is novel; the predicament it responds to is universal. Every firm faces the same vise: vendors threatening to own your knowledge from above, AI-native firms compressing your base from below, and clients funding both. Kirkland is simply the first to write a check large enough that the predicament became visible.

And that is exactly why it matters more than the dollar figure. When a struggling mid-market firm has no AI answer, you can blame resources. When Kirkland has no answer, you have ruled resources out. They have more money, more data (for whatever that is worth, given what we established at the outset), a ten-year head start on proprietary systems, and a marquee client willing to co-build than anyone alive, and they are still buying optionality rather than executing a resolution. That is the tell. The binding constraint was never technology, money, data, or client access.

We can now name what it actually is, because the first section gave us the language for it. The only durable moat is the genome, and the genome is individual, behavioral, relational, and resident in the partner. Which means it is the partner's moat, not the firm's, and that is exactly the firm's tragedy: its one defensible asset is not institutionally ownable. You cannot buy it. You cannot encode it from files. And the person who holds it will not hand it over, because surrendering it is surrendering their own power inside the partnership. Meanwhile the layer you can capture, the homogenous data, is not worth defending. So, the most profitable firm in the world spends $500 million and arrives at the wall every firm arrives at: the defensible asset does not convert, and the convertible asset is not defensible. No checkbook moves that wall.

This is the clearest existential signal we have had, precisely because it comes from the firm with the fewest excuses.

How leaders should actually read this

Ok, what is my advice? Honestly, it is not a five-point action plan, and I would distrust anyone who hands you one. But there are readings I am confident are right.

First, do not mistake their spend for a solution you can copy. The worst outcome of this announcement is fifty managing partners deciding they need their own nine-figure program to keep up. Kirkland can run this at 1 percent of revenue without touching PEP. For almost everyone else, the same spend hits distributions, which means taking real pain to buy the same unsolved problem. Resist the arms-race reflex. Matching the visible move is the trap.

Second, change the question you are asking. The reflexive question is what proprietary data do we have, and the answer, as we have seen, is the same homogenous corpus everyone else has. The real question is sharper and more uncomfortable: what is each partner's genome, can the firm keep it, and what happens the day it walks out the door. That question costs nothing to ask and most firms are not asking it.

Third, and this is the frontier I would stake the most on: the binding constraint is governance, not technology. The firm that figures out how to make partners want to contribute, through compensation redesign, some equity-like instrument, a credible promise that institutionalizing a partner's value grows their position rather than erodes it, has solved the part $500 million cannot touch. That problem is organizational, not computational. In the serious engagements I am involved in, it is the question lurking beneath every technology conversation, and the firms making real progress are the ones treating it as the actual work.

Because here is why the most profitable firm in the world still has no answer, and why no checkbook produces one. The only asset worth defending is held by the very people whose consent you need to institutionalize it, and institutionalizing it reduces their individual power. A rational partner contributes their B material and guards the genome that keeps them indispensable. Which means a firm can spend half a billion dollars building a model of how it practices, trained on everything except how its best people actually practice.

What survives

None of this means law firms are doomed. That would be its own kind of hot take. The genome work is not going anywhere, because the genome is the thing AI cannot reach. Clients will still pay for a particular partner's judgment in context, for the relationship that partner has built, for the board room gravitas, for the name on the bet-the-company opinion. The firms that consciously concentrate there, smaller, higher-margin, fewer associates, have a real future.

What is dying is not the law firm. It is the high-leverage pyramid as the default way to make money in law. The $500 million is not a moat the rest of the industry cannot cross. It is a flare, lighting up a problem none of them, Kirkland included, has solved.

I do not know how this resolves. But I am increasingly sure the firms that survive the next decade will be the ones that understood early that the work was never going to be done by a checkbook, and that the only thing worth protecting was never the data. It was the people who hold the genome, and the terms on which they will, or will not, agree to build the machine.

Those are the questions. I would rather get them right than pretend to have the answers.

Josh.

Late last year, a Klimt sold for the highest price ever paid for modern art at auction.

An outlier sure, but it wasn't a fluke. U.S. auction sales grew 23.1% in 2025. The $1-5mm segment even grew 40.8% YoY.

Meanwhile, Apollo’s chief economist Torsten Slok said to expect ‘zero in return in the S&P 500 over the coming decade.’

So, that's where it gets interesting.

Each environment is unique, but after dot-com, post war and contemporary art grew about 24% annually for a decade. After 2008, about 11% for 12 years.

It’s also had near-zero correlation with the S&P 500 since ‘95.*

Now, Masterworks lets you invest in shares of artworks featuring legends like Banksy, Basquiat, and Picasso.

Shares can sell quickly, but my subscribers can skip the waitlist:

Skip the waitlist

^{*Investing involves risk. Past performance is not indicative of future returns. See important Reg A disclosures at masterworks.com/cd.}

To read previous editions, click here.

DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.