At Davos 2026, the narrative around artificial intelligence (AI) shifted. The consideration points that surfaced from the dialogue sessions were clear: for AI to move beyond a "tech bubble", its benefits must diffuse into the real economy.
But for the enterprise chief information officer (CIO), this diffusion is multi-layered and complex, and even creates a structural paradox. As we move from AI augmentation to AI autonomy, where does the value go?
When intelligence is automated at scale, the most critical question isn't about speed or cost, but a firm's sovereignty.
For years, organisations have "borrowed" intelligence from hyperscalers to secure quick wins in productivity. Think finance teams using AI for anomaly detection that they can't tune or port, or developers relying on co-pilots that boost velocity while feeding proprietary patterns back to the provider, not the enterprise.
As AI begins to own decision-making and learning loops, many firms are realising they are inadvertently exporting their "tacit knowledge" — the unique operational know-how that defines their competitive edge — into the models of third-party vendors.
The shift: From governance to sovereignty
So, where is the middle ground? For CIOs, the "right balance" goes beyond managing a tech stack — it's about managing a learning loop.
As AI systems observe decisions, optimise outcomes, and retrain over time, leaders must decide where that learning accumulates. This means moving toward a model-agnostic architecture that lets enterprises leverage the best of hyperscalers while ensuring the "wisdom" generated by their data stays behind the firewall.
The middle ground isn't about building everything from scratch; it's about ensuring that every interaction with a hyperscaler leaves the company smarter, not just faster. Sovereignty is the discipline of ensuring AI-driven gains accrue on your balance sheet, not your vendor's.
In practice, this can take the form of prioritising edge and hybrid inference. For sensitive functions, such as proprietary coding patterns or financial anomaly detection, bring the AI to the data, not the data to the AI. Deploying inference engines within a governed perimeter prevents unique architectural preferences from being "telemetried" back into a hyperscaler's training set.
The four pillars of the sovereign AI stack
To achieve sovereignty, CIOs must decouple their intelligence from their infrastructure. This requires a "Hybrid-by-Design" approach across four layers:
- Data sovereignty (the learning signal): It's no longer enough to own your data at rest. You must own the learning signals. When a model resolves a complex customer issue, that "judgment" should refine your private internal models, not leak into a shared global pool.
- Model autonomy (the logic): Sovereign firms are moving away from monolithic dependencies. By using open-weight models (like Llama 3 or Mistral) hosted in Virtual Private Clouds (VPC) or on-prem "AI Factories," firms ensure they strike the right balance between leveraging third-party capabilities and managing operational risks.
- Workflow integrity: AI must be embedded into the execution layer where humans retain final authority. Sovereignty ensures that the "Why" behind a decision is anchored in company policy, not an opaque third-party algorithm.
- Economic capture: Productivity gains should compound internally. If an AI agent reduces operational costs, the firm should be wary of escalating "token taxes" that erode the savings.
Operationalising sovereignty in CX
Customer Experience (CX) is the frontline of this battle. Without sovereignty, a GenAI co-pilot is just a standardised tool that eventually flattens your brand's differentiation. Every company using the same "Black Box" model will eventually sound and act the same.
With a sovereign architecture, the AI becomes adaptive by design. You can swap models for sentiment analysis or risk classification while keeping your internal "logic layers" intact. Your interaction data doesn't just pass through, but it builds a private, competitive asset that makes your brand's CX more distinct over time.
The CIO as the architect of value
Firm sovereignty isn't just a repackaged label for well-designed architecture. It is a defence of an organisation's most valuable asset: its institutional memory.
As we scale agentic workflows, the role of the CIO is evolving from a provider of tools to an Architect of Sovereignty.
We must design systems where value compounds internally rather than dissipating across global platforms. In 2026, the winners won't be the companies with the most AI but the ones who truly own the AI they have.
Byron Fernandez is group chief information officer and executive vice-president of TDCX, a customer experience service provider.