Price the Judgment Layer: Where AI Value Actually Accrues

Frontier models are converging. As of March 2026 the top US model leads the top Chinese model by just 2.7%, and 88% of surveyed organizations have adopted AI somewhere. When the model itself becomes a commodity, the useful question for an investor is no longer who has the best model. It is where the value goes once the model is no longer scarce, and how to price it.

The model is becoming a commodity

A commodity is something where the next supplier is nearly as good as the best one, and the gap closes fast. By that test, the frontier model is becoming a commodity in front of us. The Stanford AI Index 2026 reports that, on the Chatbot Arena leaderboard as of March 2026, the leading US model is ahead of the leading Chinese model by only 2.7%, a gap that stayed in the single digits across the year while the top spot changed hands repeatedly. Capability that looked proprietary a year ago now diffuses in months.

Adoption confirms the picture. The same report puts AI adoption at 88% of surveyed organizations, using AI in at least one business function. When the best capability is a few percent ahead of the rest and almost everyone has access, the model is no longer the scarce, defensible asset. It is the input everyone shares.

Where the value actually lands

Here is the part that should reorganize how an investor thinks. The value created by this technology is enormous, but it is not landing on the companies that train the models. The AI Index estimates that US consumer surplus from generative AI grew from $112 billion to $172 billion in a single year, a rise of 54%, a figure the report says dwarfs the revenue producers actually earn from it. Value is being created at a staggering rate, and most of it is flowing straight past the producer to the user.

This is not new behavior for a general-purpose technology, it is the rule. In a classic study of American innovation, William Nordhaus found that innovators historically captured only about 2.2% of the total social value their innovations created; the other ninety-eight percent flowed to consumers and the wider economy. AI is running that same pattern at unusual speed: a thin slice to whoever trained the model, the overwhelming majority to whoever puts it to work.

The adoption to deployment gap is the opportunity

There is a tell in the same data. Adoption is at 88%, but the AI Index notes that AI agent deployment is still in the single digits across nearly all business functions: organizations have the model, but the layer that turns a model into a reliable, deployed system is barely built. Meanwhile the capital is pouring into the layer that is commoditizing fastest, with $285.9 billion of private AI investment in the US in 2025 alone. The money is upstream, at the model and the compute. The unclaimed value is one layer down, in deployment and judgment.

The distance between ‘we use AI’ and ‘AI reliably does the job’ is the whole opportunity, and it is not a model problem. It is a problem of context (the proprietary data and workflow a model has to be wired into) and of judgment (knowing what a good output even looks like in a specific business). Both are earned, not downloaded.

What this means for pricing

For an investor the implication is direct: price the judgment layer, not the model. A company whose only edge is access to a frontier model is renting a commodity that gets a few percent better somewhere else next quarter, and paying a margin to its supplier for the privilege. There is no durable price to defend there. The questions that should set the valuation are one layer down. What proprietary context does this company own that a competitor cannot simply call an API for? And what has it accumulated, deal by deal or customer by customer, about what good actually looks like in its domain?

That reframes the moat as two things that do not commoditize: proprietary context, the data, workflow and relationships a model has to be embedded in to be useful, and an accumulated record of judgment, the compounding asset of having been right, and wrong, enough times to know the difference. A model can be swapped out in an afternoon. Context and a track record of judgment cannot. As we argued in Valuation and Dilution, price follows where the durable value sits, and that is no longer the model.

The operator's version of this moat

This is also, almost exactly, the case for operator-led investing. The judgment and deployment layer is what a senior operator already is. Decades of selling into enterprises is proprietary context of the most valuable kind: which buyer, which committee, which objection kills a deal and which unlocks it. And a career of judging which teams and which deals actually work is an accumulated record of what good looks like, now pointed at investing. The model is available to everyone who can pay; the operator's context and judgment are not for sale. When intelligence is the commodity, the human judgment that decides what to do with it is the scarce asset, and it is exactly the asset our D-F-V screen is built to find.

The takeaway

When intelligence is commoditized, do not pay for the intelligence. Pay for the judgment that decides what to do with it, and the proprietary context that lets it be deployed where it matters. Producers have historically captured about two percent of the value they create; the investors and operators who win the next cycle will be the ones who learn to price, and to build, the ninety-eight.