Three Basic Judgments on AI and Corporate Competitive Advantage

Oct 26-2026   

In the AI era, Jay Barney—a veteran of strategy and entrepreneurship research—has offered a basic judgment: AI in itself is unlikely to yield a sustainable competitive advantage (SCA). The claim has prompted reflection and debate worldwide. In the Chinese-language business literature, a series of works by Professor Wei Wei and his collaborators on AI, business models, and competitive advantage has likewise informed both our thinking and our practice. Their articles in Tsinghua Business Review, the commentaries and rejoinders that followed, and the exchanges they generated have been pointed and consistently illuminating, giving us a useful conceptual foundation and a practical compass for further reflection, assessment, and improvement of practice.

We have read them with real pleasure, marking passage after passage with admiration. Yet a few doubts keep nagging at us, and we would rather air them than hold them back. We offer them here for the scrutiny of readers better placed to judge, and as a cheer for this series and a brick of our own for its wall. Given the constraints of space, we privilege provocation over completeness, and aim to be brief and to the point. On the many points where the two teams already agree, we will not dwell. Our case is developed in three main respects.

Perhaps There Is No Generic, Homogeneous Version of AI

To put it provocatively: the question of whether AI can yield a competitive advantage is a false question. It is rather like asking whether resources can give a firm a competitive advantage, or whether technology can. These are the wrong questions. The right question is: what kind of resource gives a firm a competitive advantage? What kind of AI can give a firm a competitive advantage—indeed, a sustainable one? One has only to apply Barney's VRIN framework directly. Yet neither Barney and his colleagues nor most of the commentary on the subject has done so.

From the initial arguments that Barney's team advanced in the Harvard Business Review, MIT Sloan Management Review, and elsewhere, one basic premise appears to have been assumed: that AI is a general-purpose, replicable, homogeneous factor. On that premise, AI is like any other candidate foundation for sustainable advantage—if it is homogeneous, it cannot serve as one. The inference then follows naturally: since everyone can buy and use AI, AI cannot, in long-run equilibrium, confer a sustainable advantage on any particular firm. This reasoning looks especially plausible against the backdrop of China's open-source route, under which any firm may use open-source models without licensing them from the model vendor.

The crux is whether AI is in fact one homogeneous, monolithic block. Is AI like electricity—or like the earliest free telephone services, or the internet as it first appeared? If AI were that generic and that standardized, it plainly could not generate advantage. Unless, that is, one insisted, as only obsessive audiophiles do, that one can hear whether a given sound system is running on wind-generated or hydro-generated power. Put bluntly: electricity itself does not differ.

But can one plausibly posit an AI as generic and homogeneous as electricity? Compared with electricity, AI necessarily appears in more specific forms and versions—much as personal computers, or ERP systems and services, come from different vendors. Differentiation is possible, and its durability depends on the context. Hence, taking as given the differences that already exist among firms before AI enters, in incremental terms heterogeneity across AI versions will confer advantage on some firms; and if that differentiation endures, so does the advantage.

Beyond differences in the versions and forms of AI products themselves, the circumstances in which firms adopt AI, the timing, pace, scope, and scale of adoption, and the fit between people and complementary resources—all of these can create temporary or enduring differences among firms, and so make (sustained) advantage possible. That includes the sustained advantage that Barney and his colleagues describe as accumulating out of, and eventually enveloping, a run of transient advantages—a pattern already demonstrated in the early PC era of IT by firms such as Intel.

It should be added that even where AI approaches homogeneity, first movers may amplify and prolong the advantage of early adoption by pre-empting complementors. In an earlier age, a village that got electricity early could transform its schools and its roads, and so gain an advantage in human capital and other infrastructure that kept it ahead of neighbouring villages for years. In the AI era, this logic still holds—something Barney's team does not dispute.

On the practice side, more and more firms are waking up. They cannot help but reconsider what they have been doing: handing their most valuable data to the big AI vendors for free to train on, and then paying large sums to buy the vendors' products and services back, so that in the end they retain no valuable assets of their own. It is time to think about localized, context-specific AI.

What Kind of AI Can Underpin Competitive Advantage

If we treat AI as a resource, then for it to serve as a foundation for advantage there is no shortcut: it must still satisfy Barney's VRIN criteria. Concretely, it must be firm-specific and embedded in the firm's activities—in other words, AI products and services that cannot simply be bought off the shelf, together with the firm's own localized, context-specific deployment of them.

First, distinctive data. This is the basis for localized AI learning. For such data to be valuable, rare, inimitable, and non-substitutable—whether we call it, following Barney's original statement of the resource-based view, "the heterogeneity of strategic factors," or, following what his team now emphasizes, "residual heterogeneity"—is an indispensable property of any source of sustainable advantage.

Second, localized and context-specific AI learning conducted on the firm's distinctive data, and the firm-specific algorithms that result. These algorithms and models help extract distinctive insight from data—even from generic data—and thereby promote the generation, accumulation, and real-time application and adjustment of distinctive data. They also encompass the design of the interfaces through which the firm's functions coordinate, and the norms and processes of cross-firm collaboration.

Third, in settings where AI appears as tool, partner, and agent, the firm's capacity for creative, continuous learning, together with the results of its repeated iteration and replacement. Any application of technology in business should be a race among firms—particularly in how technology and people are combined. In the end, the firms that learn well, and that keep learning and adjusting, are the ones that win.

In sum, distinctive data for AI learning, distinctive firm-specific algorithms, and a capacity for rapid AI learning, adaptation, and adjustment suited to the particular firm—these are the sources of sustainable competitive advantage in the AI era. None of them is a generic, homogeneous version. All are bound up with the firm's own distinctive practice: firm-specific AI practice and routine.

Human–Machine Interaction Is Indispensable to Sustainable Advantage

Sustainable competitive advantage exists because people exist. Everyone says AI is a black box, beyond explanation. In truth, people are far harder to explain. Machine behaviour and performance are, most of the time, predictable, hallucinations notwithstanding. Certain kinds of human imagination and extreme behaviour remain beyond anything machines can match—and this matters enormously for how firms cope with uncertainty and extend the frontier of productivity.

Building on the third point of the preceding section: how people and machines learn together, how they learn quickly and effectively, and how they learn better than rivals—this is the ultimate source of a genuinely new form of sustainable advantage, one that lies in the incremental space of the AI era. It is neither the residual heterogeneity of the analog age nor the homogeneous asset of the AI age, but a distinctive capability that is firm-specific to the new era.

Suppose future firms are run by AI agents, with human influence reduced to a minimum or to zero—where the one-person company or the people-free company becomes the norm. Then Barney's team will have been absolutely right: AI will confer no sustainable advantage. No company will hold an advantage because of AI. You can buy AI; so can I. My AI firm can buy your AI firm; no people need get in the way between AI firms, since technical standards and interfaces can settle everything. That is the world in which AI is (most nearly) homogeneous. People have emotions. Groups have cultures. Coordinating people carries costs, and those costs can rise high enough to make coordination and integration impossible. Heterogeneity among machines, by contrast, can usually be bought out and settled—machines are easier to coordinate and integrate than people. This too is consistent with the tenor of Barney's team's original argument, which plays down the heterogeneity of AI and stresses that AI cannot yield sustainable advantage.

And here lies the paradox. Perhaps no firm has a competitive advantage, and only you do—because the end point of acquisition is natural monopoly. String together every village's power line in the country and you are State Grid: you hold every advantage, but you have no competition. Yet State Grid still needs people. An AI-agent company does not even need those. Anything that is technology can simply be bought and sold. In the end, coal-fired, hydro, and even wind and nuclear power are all just electricity. All AI agents are taught by the same teacher. All one-person companies end up with a single boss—all of them subsidiaries of one parent company.

Finally, we thank both teams once again for a feast of ideas. Within the broader conversation on AI, their exchange on business models is equally instructive—above all Wei Wei's team's systematic thinking on business models and on value creation and capture in the AI era, and their engagement with, and comparison to, Barney's theory. Quite so: the core and substance of a business model is the transaction model. How it shapes value creation and value capture in the AI era surely deserves deeper investigation and testing. In the AI era, different transaction models—the differing degrees of AI adoption across platforms, for instance, and how each platform's AI agents coordinate with merchants on that platform, with all potential participants, and with other platforms, whether partners or rivals—are all important dimensions for examining the possibility and sustainability of particular competitive advantages. We look forward to further, even more illuminating academic dialogue, and to further innovation in practice.