Who Owns the Technology Harness?

In Brief

• AI models may provide reasoning, but the technology harness turns that reasoning into useful, secure, and authorized business outcomes.
• As multiple models and chips start to optimize, value may shift toward the layer that routes workloads, applies context and permissions, and owns the customer relationship.
• For investors, the key question is not simply who builds the best AI model, but who controls the harness, the customer relationship, and the economics.

In our last Strategist’s Corner, Who Owns the Outcome?, we argued that AI could take over the screen and much of the workflow without capturing the underlying economics. An agent still needs context, rules, permissions, auditability, and authority to turn intent into action. The next question is one layer higher: if the AI model is the engine, who builds and owns the car?

An Engine Is Not a Car

Consider an automobile. The engine supplies power, but few people buy one to leave in the driveway (except a gearhead friend of mine). The transmission, brakes, steering, and controls turn that power into safe, reliable transportation. The technology harness plays the same role for AI (Exhibit 1): it selects the model, connects it to tools, memory, and data, applies security and permissions, and turns an answer into an action. Customers are not paying for horsepower or tokens alone; they are paying for a dependable business outcome.

When the Engine Becomes Good Enough

When one engine is materially better than every alternative, tight integration matters. The same is true for AI models. Superior capability can justify premium pricing and support a more integrated product.

But the basis of competition changes when several models become good enough for a growing number of tasks. Customers may accept slightly less capability in exchange for lower cost, faster response times, stronger privacy, or greater control.

At that point, the model can begin to separate from the harness. If an application can utilize any one of several capable engines without materially weakening the product, model quality will no longer determine who owns the customer relationship or captures the economics.

Our technology analyst team’s work points in that direction. The model ecosystem is fragmenting on a token basis, even if revenue remains concentrated. The chip ecosystem is broadening as well. Meanwhile, lower token prices appear to be encouraging more usage, consistent with Jevon’s paradox, which we explore in AI Risks Extending Beyond Software. Together, these trends can increase the value of the harness.

Competition Below, Control Above

A multi-model harness can assign the strongest model to a difficult task, a lower-cost model to a routine one, and a private model when data sensitivity matters. That same harness could also route workloads across different chips and computing environments, while the customer continues to experience a single product on the front end. 

More competition below the harness can lower costs, improve choice, and broaden usage. But, as we’ve previously argued, value creation is not the same as value capture.

If the harness is merely a thin router, competition may pass a lot of the savings to the customer. Durable economics require control over something that remains scarce, such as proprietary context, identity, distribution, user history, business rules, permissions, or transaction authority.

The strongest harness can therefore choose among competing models and chips while remaining difficult for the customer to replace.

Who Can Own the Harness?

Frontier model companies own the reasoning layer and may move closer to the end user to avoid becoming commodity suppliers. Hyperscalers own compute, storage, security, billing, and access to multiple models. Application vendors own workflows, company-specific context, and, in some cases, transaction authority.

Each is trying to control more of the car. In other words, competition is rising. 

Conclusion

Who gets paid for the car ride? This is the central question. The first stage of AI rewarded the companies that were building, or helping to build, the fastest models. The next stage may reward those that turn those models into products customers can use and trust.

History does not suggest that incumbents always lose. It tells us that profit pools move. Exceptional returns attract capital and new entrants. Customers gain alternatives, pricing comes under pressure, and incumbents spend more to defend their positions.

Leaders do not have to disappear for the economics to change. They may simply capture less of the next dollar or earn a lower return on it. Railways and telecommunications transformed the economy, but the capital and competition they attracted ultimately created additional capacity and diluted returns.

That is why identifying AI winners is a security-selection problem, not simply an AI-allocation question. A cap-weighted benchmark owns today’s leaders at today’s weights. It cannot distinguish in advance between a company whose harness is strengthening its customer relationship and one whose economics are being competed away.

The AI model may provide the horsepower. The more important and difficult question for investors is who owns the harness, the customer relationship, and the economics.

Keep in mind that all investments carry a certain amount of risk, including the possible loss of the principal amount invested.

The views expressed are those of the author(s) and are subject to change at any time. These views are for informational purposes only and should not be relied upon as a recommendation to purchase any security or as a solicitation or investment advice. No forecasts can be guaranteed. Past performance is no guarantee of future results.