As enterprises move beyond AI experimentation, a different question begins to surface. Who should actually own AI?

The answer is rarely straightforward. A centralized model gives the enterprise greater control over technology, governance, security, and investment. A federated model puts more responsibility with business functions, allowing teams closer to the work to identify and develop use cases faster. Both models have clear advantages. Both can also create significant problems when taken too far. The debate, therefore, should not be about choosing centralization or federation as an absolute. The more important question is which decisions need enterprise wide control and which decisions should remain close to the business.

‍

‍

The case for centralization

A centralized AI function can create consistency at a time when enterprises are experimenting with rapidly changing technologies. Instead of every business unit evaluating its own tools, the organization can establish common standards for technology selection, security, data access, governance, and risk management. Centralization can also reduce duplication. If finance, procurement, operations, and customer service are independently solving similar problems, a central team can identify opportunities to reuse capabilities rather than building the same foundation repeatedly. There is also a strong governance argument. Enterprise AI introduces questions around data privacy, security, model performance, access, compliance, and accountability. A central function can establish common principles and provide oversight across use cases.

The problem begins when centralization extends beyond what actually needs to be centralized. A central AI team can quickly become a bottleneck if every use case, technology decision, and implementation request has to pass through it. Business teams may have to wait for resource from a team that does not have enough understanding of their processes or priorities. AI then becomes something the enterprise controls rather than something the business can actively use.

‍

‍

The case for federation

A federated model addresses this problem by putting AI closer to the functions where business problems exist. A procurement team understands its processes, suppliers, decision points, and operational constraints. A finance team understands the information needed for financial decisions. An operations team understands the conditions that influence its workflows. Giving these teams more responsibility can make AI adoption faster and more relevant. Federation also encourages experimentation. Business teams do not have to translate every opportunity into a central technology roadmap before testing it. They can identify problems, explore solutions, and learn from their own processes.

But federation introduces another risk. When every function makes its own decisions, the enterprise can end up with fragmented technology, inconsistent standards, duplicated investments, and different approaches to data and governance. What works for one function may not connect easily with another. The organization gains speed locally while losing coherence globally. A federated model can therefore solve the bottleneck created by excessive centralization while creating fragmentation of its own.

‍

‍

The real choice is about decision rights

The most effective enterprise AI operating model is unlikely to be completely centralized or completely federated. Instead, organizations need to determine where decision rights should sit. Some areas naturally require enterprise oversight. Security, data protection, risk management, architecture standards, and governance should not vary dramatically from one business unit to another. Shared platforms, reusable capabilities, and common technology standards can also benefit from central coordination.

Other decisions are better made closer to the business. Identifying valuable use cases, understanding process requirements, defining business outcomes, and determining how employees will use a solution require functional knowledge that a central team may not possess. This creates a more useful distinction. The question is not who owns AI. It is who owns which decisions about AI. A central function can establish the guardrails while business teams determine where AI should create value within those guardrails.

‍

‍

Governance should enable adoption, not slow it

One of the risks of centralization is treating governance as a gate that every AI initiative must pass before meaningful experimentation can begin. Enterprise governance is necessary, but its role should be to make responsible adoption easier rather than create unnecessary friction. This requires different levels of oversight for different types of use cases. A low risk internal application does not necessarily require the same review process as an AI capability influencing financial decisions, customer outcomes, or sensitive information.

The same principle applies to technology standards. Enterprises benefit from common foundations, but those foundations should allow business teams to build on them without repeatedly solving the same technical and governance problems. The objective is controlled freedom. Business teams should have enough autonomy to move quickly, while the enterprise retains enough control to manage risks and maintain coherence.

‍

‍

What should the operating model look like?

The answer will vary by enterprise maturity, regulatory environment, technology landscape, and business structure. But one principle remains consistent: the operating model should reflect how value is created. If the central team owns everything, business adoption can become slow. If every business unit owns everything, the organization can become fragmented. A more balanced model gives the enterprise responsibility for shared foundations and guardrails while giving business functions responsibility for identifying and prioritizing opportunities.

This also requires strong coordination between the center and the functions. The central AI organization should not operate as a project factory that receives requests from the business. It should provide capabilities, standards, expertise, and reusable foundations that help business teams move faster. Business functions, in turn, should not treat AI as an isolated technology initiative. They need to connect use cases to measurable business outcomes and participate in governance and adoption.

‍

‍

The operating model will determine how fast AI scales

The centralized versus federated debate is ultimately a question about organizational design. Enterprises that centralize too much risk creating dependency on a small group of specialists. Enterprises that federate everything risk creating a collection of disconnected AI environments that become difficult to govern and integrate. The more sustainable model is one where central control exists where consistency matters and business autonomy exists where context matters.

AI at enterprise scale requires both. The central organization provides the foundation, guardrails, and shared capabilities. The business provides the problems, context, priorities, and ownership. The real measure of an AI operating model is therefore not whether it is centralized or federated. It is whether it allows the enterprise to move quickly without losing control, encourages business ownership without creating fragmentation, and makes successful AI capabilities easier to scale across the organization.

‍