Today the answer has two parts: hours and a share of each year's profits. A partner sells time, collects distributions for as long as they stay and when they retire the firm returns the capital they put in, at roughly book value and the relationship ends. That formula has held for a century. Every part of it is now under pressure and the pressure is coming from the same direction.
Kirkland & Ellis is spending $500 million to build its own legal AI platform. Thomson Reuters trained a competitive model for $40 million and Harvey trained one in two months of compute. For a firm that books more than $8 billion a year, the build is affordable. The strange part is what happens to the asset afterwards, because the way law firm ownership works, the partners helping the training are the one group with no lasting claim on it.
Ownership that expires
A Big Law equity partner's ownership has splits in two: a share of each year's profits and a capital account they paid into when they made partner. When they retire, they get the capital account back at roughly book value, often paid out over a couple of years, plus their share of that final year's earnings.
In most large firms, that is the whole exit. There is no payment for the ongoing value of the firm they spent thirty years building, because most big firms deliberately exclude goodwill from the calculation and there is no market where a partnership stake can be sold.
So the ownership is real, but it expires and cannot be sold. It is a claim on profits that lasts exactly as long as you stay.
That design made sense for most of the profession's history, because the firm's real asset, the judgment of its senior people, expired on the same schedule. A partner's judgment produced income until retirement and then left the building with them. There was nothing durable for the ownership to hold a claim on, so nobody minded that the claim itself wasn't durable.
AI breaks that symmetry. A platform trained on partner judgment keeps working after the partners who supplied it stop billing. Kirkland is spending $500 million, funded out of its partners' current distributions, to build a genuinely durable asset. And the partners paying for it hold ownership that expires. A partner who retires in 2028 will have financed the platform and will exit with a capital account returned at book value, whatever the platform is worth by then. For the first time, what the firm is worth and what a partner can take out of it are two different numbers, and the gap between them grows with every dollar spent on AI.
Hourly billing gives the AI gains to the client
The billable hour means the client buys time and when AI makes the work faster, the honest result is a smaller invoice. Every efficiency the firm creates flows to the client automatically, which means a firm adopting AI under hourly pricing is spending its own money to shrink its own revenue.
So far, firms have handled this by keeping the gains inside the invoice. Thomson Reuters measured AI compressing document review time by 60 to 70 percent, yet most in-house counsel say they have seen no savings from their outside firms. The distance between those two facts is the productivity gain, currently being absorbed rather than passed on.
Clients have noticed, and they are starting to write the gain into their contracts. Zscaler's outside counsel guidelines now state that AI-generated work product costs "shall not be passed on to the company" and Meta and UBS added similar billing provisions this year. Once enough clients do this, holding hours flat stops working as a model.
The escape is pricing the outcome instead of the time and that is the direction the market will be moving in. But outcome pricing sharpens the ownership question rather than settling it. When a fee no longer corresponds to anyone's hours, because the work was largely produced by a system trained on the firm's accumulated judgment, the money is being earned by an asset. And an asset needs owners in a way that a timesheet didn't.
Equity that survives you
There is a structure emerging that addresses this - an MSO (a management services organization). The ideas is that a firm splits in two: the lawyers keep the entity that practises law, while the technology, the data and the business operations move into a services company the law firm pays fees to. US ethics rules ban non-lawyers from owning law firms, but not from owning services companies, so outside investors can buy into the second entity. The standard reading is that this is how private equity gets into law.
Reading it from the partner's side instead - an MSO converts ownership that expires into ownership that doesn't. A stake in the services company has a valuation, survives retirement and can eventually be sold, which is everything a partnership interest is not. As long as a firm's value walked out the door with its people, that difference was academic. Once the firm's value starts accumulating in a platform, the difference becomes imporant, because the partners whose judgment trains the platform will want to hold the kind of equity that captures what the platform earns after they stop billing. The judgment is becoming valuable once in the matters they handle now and again in every future matter the system resolves because it learned from them.
Accounting has already run this process at scale. When Baker Tilly took $1 billion from private equity in 2024, a large part of the money went to buying out partner retirements, which is to say, to letting partners finally cash out value their partnership structure had trapped. Half of the top 25 US accounting firms have now taken private equity money and the wave is best understood as a series of liquidity events.
What other industries can tell us about monetizing the knowledge itself
Everything above is about firms building internally. The second version of the question is what happens when the AI is built outside, by companies that need the profession's knowledge to train it and here other industries are further down the road.
In publishing, AI companies have committed close to $3 billion to publishers for training data, with News Corp signing for up to $250 million over five years and Reddit collecting around $60 million a year from Google. Photo libraries, academic publishers and, as of this year, the music industry all followed the same arc: initial free scraping, then conflict, then priced licensing. Professional services firms have not started down that arc.
In other hourly-billed industries, the knowledge is leaving anyway, just unpriced. Bloomberg found around 150 former McKinsey, BCG and Bain consultants hired individually by AI companies to train models on consulting work, paid by the hour. The expertise is already being sold - it is being sold retail, by individuals, as labor, with the firms capturing nothing.
What the model builders actually need from firms is the firm's niche knowledge layer: the playbooks, the precedent banks and, above all, the accumulated sense of how to act for a particular kind of client in a particular kind of moment - call it "judgment" or "taste". Which points to push in a negotiation and which to concede, what this regulator will accept, how a specific board needs bad news delivered. None of it was ever written down anywhere public, which means the frontier models cannot scrape it and do not have it. And it is exactly what they will pay for, because generic legal capability is already commoditized and the value that remains sits in the niches where this judgement was built.
Right now, firms are handing this layer over for free. The elite firms signing on as launch partners for the big AI platforms contribute exactly this kind of workflow knowledge in exchange for early access and publicity. They accept that trade because nobody knows what the knowledge is worth: no deal has ever put a price on it. The first firm to sign a real licensing deal, with revenue share, creates the number every other firm can point to across the table. Publishing had that moment when News Corp signed. Legal hasn't had it yet.
The regulators' move
The rule that non-lawyers cannot own law firms exists so that a lawyer's judgment answers to the client and nothing else and the MSO structure does not make that concern disappear.
What has changed is that the demand is no longer hypothetical. Holland & Knight's team, which advises on these deals, says it has closed around 25 MSO transactions with roughly 100 more in progress, and that it has spoken with half of the Am Law 100. Yet every completed deal so far is mid-market. No major firm has done one, and the reason is plain: the first one to move carries the entire regulatory risk by itself, in front of every bar authority in the country, with no precedent to hide behind. And regulators' will most probably speed up their reaction if such a large deal happens.
That is why the interesting question for the next two years is not whether firms want this, the pipeline already answers that. It is how regulators respond when a major firm finally forces the issue. They can accommodate the structure, constrain it, or fight it and whichever they choose sets the pace for everything described above.
The next two years
Three things follow from this argument.
Partner compensation starts shifting toward equity tied to the firm's long-term performance and those packages grow as the AI platforms do. Not because firms become generous, but because a paycheck rewards this year's hours while the thing being built runs on decades of accumulated judgment and the people holding that judgment will price it accordingly. Top firms will start extending durable, appreciating stakes to the partners they most need to keep.
A major firm signs the first knowledge licensing deal with a model builder within two years, with revenue share and gives legal knowledge its first market price. Until that happens, every firm contributing its know-how to an AI platform is doing so without knowing what it gave up.
And a major firm completes an MSO, forcing the regulatory answer. The mid-market has built the playbook and absorbed the early scrutiny. One of the top 20 US law firms will eventually decide the trapped value is worth more than the first-mover risk.
So, what will a partner get paid for in 2030? Less and less for hours, because clients are already refusing to fund them. More and more for judgment: applied across many clients at once by systems trained on it.
