When AI strategy becomes infrastructure: what CDOs must own in 2026

A Fortune 500 retailer runs seventeen distinct AI initiatives across pricing, demand forecasting, customer segmentationcustomer segmentationDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.View full definition →, and supply chain optimization. Each was approved by a different business unit. Each uses a different vendor. None of them share a data layer. When the CFO asks how much lift the company is actually getting from its AI investment, nobody can answer with confidence. This is not an edge case. It describes the majority of large enterprises in 2026.

The CDO's position in this moment is simultaneously more powerful and more precarious than it was two years ago. The pressure to show AI results is intense. The budget is often real. But the organizational architecture to capture value at scale remains fragmented in most companies.

The structural shift happening across the enterprise

The first wave of enterprise AI (roughly 2022 to 2024) was dominated by experimentation. Teams stood up proof-of-concepts, pilots ran in controlled environments, and executives watched demos. The governing question was feasibility: can this work?

That phase is largely over. By 2026, the governing question has shifted to durability: can this scale, can it be governed, and can the business actually depend on it? That is a fundamentally different problem, and it lands squarely on the CDO's desk.

Several structural patterns have become visible across industries. First, the proliferation of large language modellarge language modelA Large Language Model is an AI system trained on vast text data to predict and generate language, enabling tasks like writing, summarizing, and answering questions.View full definition → integrations into business workflows has created a new category of data risk. When a sales team uses a vendor-embedded AI assistant (Salesforce Einstein, Microsoft Copilot embedded in Dynamics) to draft proposals or summarize customer history, that activity generates data flows that most organizations have not mapped. According to Gartner research from 2025, fewer than 30% of enterprises had documented their shadow AI usage as of that year. The number has improved, but the gap remains significant.

Second, the economics of model deployment have shifted enough that build-versus-buy decisions need revisiting. The cost of running fine-tuned open-weight models (Meta's Llama family, Mistral) has dropped sharply since 2023. For organizations with proprietary data assets, the argument for retaining that data inside a self-hosted environment, rather than sending it to a third-party APIAPIApplication Programming Interface: a standardised interface that lets applications communicate and exchange data without knowing each other's internal workings.View full definition →, has become more compelling on both cost and governance grounds.

Third, boards have started asking about AI governance in a way they were not two years ago. The EU AI Act's phased enforcement timeline, combined with early enforcement actions in financial services, has moved this from a compliance team conversation to a board-level one.

What this means for the CDO

The CDO who treats AI strategy as a technology procurement function will lose relevance fast. The ones building durable influence are doing something different: they are defining AI strategy as an extension of data strategy, which means owning the infrastructure layer that every AI initiative depends on.

Concretely, this involves four things that are harder than they sound.

Unified data semantics. AI models, whether they are predictive, generative, or a hybrid, inherit the inconsistencies in your data. If "customer" means something different in your CRMCRMCustomer Relationship Management: software and strategy to manage and analyse customer interactions throughout their lifecycle.View full definition → than in your data warehousedata warehouseA central repository that consolidates data from many source systems into a structured, query-optimized store designed for analytics, reporting, and business intelligence.View full definition →, your churn prediction model and your LLMLLMA Large Language Model is an AI system trained on vast text data to predict and generate language, enabling tasks like writing, summarizing, and answering questions.View full definition →-generated outreach will produce outputs that contradict each other. Resolving semantic conflicts is unglamorous work, but it is the prerequisite for AI that a business can actually rely on.

Vendor accountability frameworks. Most enterprises are running AI from a mix of providers: hyperscalers (Azure OpenAI, Google Vertex, AWS Bedrock), specialized vendors, and internal builds. Each comes with different data retention policies, model update cadences, and audit capabilities. The CDO needs a structured framework for evaluating these, not just a legal checklist. When Anthropic or OpenAI updates an underlying model, does your output validation process catch behavioral drift? Most organizations do not have one.

Redefining what "data productdata productA data asset managed like a product, with an owner, defined users, guaranteed quality, and measurable business value.View full definition →" means in an AI context. The data product concept has been in circulation for several years, but it requires a meaningful update for AI use cases. A data product that feeds a dashboard needs freshness and accuracy. A data product that feeds an AI agent also needs documented lineage, access controls that account for model behavior, and clear ownership of the outputs the model generates. That is a more complex specification, and it requires the CDO to work much more closely with AI engineering teams than has historically been the case.

Governance that moves at the speed of deployment. This is the hardest one. Traditional data governancedata governanceData governance is the set of policies, roles, and processes that ensure data is accurate, secure, well-defined, and used responsibly across an organization.View full definition → processes were built for a world where new data pipelines took weeks or months to deploy. In an environment where a business analyst can spin up a custom GPT or connect a no-code AI tool to a business system in an afternoon, the governance function has to shift from gate-keeping to embedded standards. That means policy-as-code where possible, and clear escalation paths where it is not.

Practical priorities for CDOs navigating this

• Audit the AI vendor landscape inside your organization before the next budget cycle, not as a cost exercise but as a data flowdata flowAn automated sequence of steps that moves data from source to destination: ingestion, transformation, validation, and loading, so it arrives clean and ready to use.View full definition → mapping exercise. You cannot govern what you have not located.
• Require that every new AI initiative names a data owner, not just a business sponsor. The business sponsor cares about the outcome. The data owner is accountable for the inputs and the lineage.
• Revisit your data contracts with hyperscalers. Several major providers updated their terms of service in 2024 and 2025 in ways that affect how customer data can be used for model training. These changes were not always announced prominently (Microsoft, Google, and Amazon each made revisions worth reviewing with legal counsel).
• Push for AI literacy at the governance layer, not just the practitioner layer. A data governance committee that does not understand the difference between a fine-tuned model and a retrieval-augmented generation system will make poor tradeoff decisions about risk and investment.
• Build a small internal capability for model evaluation and red-teaming. This does not require a large team, but it does require dedicated attention. Relying entirely on vendor-provided benchmarks (note: all major AI vendors publish performance claims that serve their commercial interests and should be cross-referenced with independent assessments) leaves the organization exposed.

The CDO who can credibly answer the question "what is the data architecture that makes our AI portfolio reliable and auditable?" holds a position that no other executive currently owns. That is the role worth building toward in 2026.