AI Adoption Is Repeating a Very Old Mistake: What the History of Capitalism Tells Enterprise Leaders About Zero-Sum Thinking

Before Adam Smith dismantled it, mercantilism was the operating logic of every serious economic power in Europe. The core assumption was simple: wealth is finite, and the goal of strategy is to capture more of it than your rivals. That assumption produced trade wars, colonial extraction, and ultimately, economic stagnation in the nations that held onto it longest. Most enterprise AI strategies today are built on an identical assumption, and the consequences are likely to follow the same pattern.

The parallel is not rhetorical. When organizations treat AI capability as a fixed pool of advantage to be accumulated and defended, they make structural decisions that mirror mercantilist policy: hoarding proprietary data, restricting model access across business units, competing for AI talent as a scarcity rather than building the conditions for its reproduction. Each of these moves feels rational in isolation. In aggregate, they tend to produce the same outcome mercantilism did: internal rigidity, missed compounding returns, and eventual displacement by actors who understood the generative logic of the technology earlier.

The Fixed-Pie Assumption in AI Investment Decisions

The most visible expression of zero-sum AI thinking is in how organizations frame competitive moats. The dominant logic holds that proprietary data, fine-tuned models, and specialist AI headcount constitute durable barriers to entry. That framing borrows from intellectual property economics, where exclusivity is the mechanism of value.

The problem is that AI capability does not behave like a patent. The underlying model architectures are largely commoditized. The fine-tuning advantage erodes as foundation models improve. Data moats are real but narrower than most strategy teams assume, because the marginal value of additional proprietary data diminishes significantly once a model has learned the domain's structural patterns.

Organizations that build their AI roadmap around exclusivity as the primary mechanism are optimizing for a defensibility that is structurally time-limited. The strategic question is not how to hold the resource, but how fast you can generate new applications from it.

Talent Hoarding as a Mercantilist Policy

The hiring behavior of large enterprises in AI follows the same logic. Retaining specialists through compensation inflation, restrictive non-competes, and deliberate market thinning is a direct analogue to mercantilist labor controls. The assumption is that AI capability lives in individuals, and that restricting their mobility is equivalent to accumulating strategic resource.

This assumption is increasingly difficult to sustain. AI tooling is redistributing specialized capability across a broader base of practitioners. The value of a senior ML engineer no longer resides primarily in knowledge that others cannot access. It resides in judgment, context, and the ability to operate effectively in production environments where the easy problems have already been automated.

Organizations that continue to treat AI talent as a scarce commodity to be hoarded will find that they are paying scarcity premiums for capabilities that are no longer scarce, while underinvesting in the organizational structures that determine whether those capabilities compound over time.

Companion piece to our broader work on AI talent strategy. See The Scarcity Myth: Technical Specialization & AI for a detailed treatment of workforce planning and skill reweighting as specialist knowledge monopolies erode.

Data Moats and the Enclosure Problem

The mercantilist era had enclosures: the privatization of common land that had previously supported distributed economic activity. Enterprise data strategy has developed its own version. Business units hoard data from each other, legal teams restrict data sharing across organizational boundaries, and AI teams build models on siloed inputs that systematically underrepresent the full operational picture.

The enclosure logic feels protective. In practice, it produces models that are trained on partial signals, evaluated against incomplete ground truth, and deployed into environments they were not calibrated for. The competitive harm is internal before it is external.

The organizations that are generating the most durable value from AI are not the ones with the largest proprietary datasets. They are the ones that have built the data infrastructure and governance to move information across organizational boundaries quickly, without losing lineage or accountability.

What Positive-Sum AI Strategy Actually Requires

The transition from mercantilist to capitalist economic logic required a conceptual shift: from wealth as a stock to be captured, to wealth as a flow to be generated. The equivalent shift in AI strategy is from capability as an asset to be owned, to capability as a process to be accelerated.

In practical terms, this means evaluating AI investments not on the basis of how much proprietary advantage they accumulate, but on how much they increase the organization's rate of learning and deployment. A shared internal platform that enables ten teams to run experiments is more strategically valuable than a proprietary model that one team guards. The compounding effect of distributed experimentation consistently outperforms the one-time advantage of exclusivity.

It also means auditing organizational incentives. If business unit leaders are rewarded for AI capabilities that stay within their domain, the structural incentive is toward hoarding. Positive-sum AI strategy requires that the reward structure reflects system-level outcomes, not unit-level accumulation.

The Competitive Positioning Implication

The historical record on mercantilism is not that it produced no value. It produced significant short-term advantage for the nations and trading companies that practiced it. The problem was that it systematically prevented the conditions for compounding returns from forming, and it was eventually displaced by actors who operated on a different logic.

Enterprise AI is at an equivalent inflection point. Organizations that have built their competitive positioning on data exclusivity, talent restriction, and model hoarding are likely to find that their advantage is real but non-compounding. They will extract value from their current position without generating the conditions for the next position.

The audit question for any CTO or Chief AI Officer is direct: does your current AI strategy produce more capability over time, or does it primarily protect the capability you already have? The answer to that question determines whether you are operating a mercantilist or a generative AI strategy, and the historical evidence on which one wins over a decade is not ambiguous.