Latest / MBA Training Data / When AI strategy becomes infrastructure: what CDOs must own in 2026
Transcript
- 0:00Welcome to the Leaders' Insights podcast. Today's episode, when AI strategy becomes
- 0:10infrastructure, what CDOs must own in 2026. Most CDOs I speak to in 2026 are still running
- 0:17AI as a project portfolio. Your argument is that's already the wrong frame. Why?
- 0:24Because projects have end dates and budgets. Infrastructure doesn't. When JPMorgan Chase
- 0:29talks about AI being embedded in 300-plus use cases across the bank, they're not managing
- 0:34a project list. They're operating a system. The CDO, who's still approving individual
- 0:39AI initiatives one by one, has already lost the plot.
- 0:43So what does it mean in practice to treat AI as infrastructure rather than a set of initiatives?
- 0:49It means the data layer comes first, every time. You can't retrofit a foundation. What I see
- 0:55failing repeatedly is organizations that bought into a model provider, whether that's OpenAI,
- 1:00Mistral, Google, without first answering whether their data is clean, governed, and accessible
- 1:07at the speed AI actually requires. Inference is cheap now. Bad data is still catastrophically
- 1:14expensive. Where does the CDO's decision authority actually sit in 2026? Because the CTO wants the
- 1:20stack, the CFO wants the ROI case, and the business units want autonomy.
- 1:25The CDO has to own three things non-negotiably. The data contracts between systems, the governance
- 1:31layer that determines what AI can act on autonomously versus what requires human sign-off,
- 1:37and the metadata strategy. Everything else can be federated. But if you seed those three,
- 1:43you end up with what I'd call shadow AI, business units running models on unverified data,
- 1:49and nobody knows until something breaks publicly. Walmart has been pretty vocal about their
- 1:54federated AI model. Does that hold up as an example of getting this balance right?
- 1:58Partially. Walmart's approach of giving business units model access while centralizing data
- 2:04contracts is directionally correct. But even they've acknowledged, and this came out in their
- 2:10Q120-26 earnings commentary, that the metadata governance piece lagged the deployment pace.
- 2:17They're now backfilling that. It's a common pattern. Deployment outruns governance,
- 2:22and the CDO spends 18 months cleaning up what could have been designed correctly upfront.
- 2:27If a CDO is sitting in a board meeting next week trying to make the case for owning this
- 2:31infrastructure mandate, what's the sharpest argument they can make?
- 2:35Ask the board one question. Which of our competitors would you be comfortable letting
- 2:40write the rules for how we use data? Because if you don't build this internally,
- 2:44you're effectively outsourcing that governance to whoever sold you the platform.
- 2:49Databricks, Snowflake, Microsoft. They all have opinions baked into their architectures.
- 2:55The CDO's job is to make sure those opinions serve the company's strategy,
- 2:59not the vendor's roadmap. Final question. One thing a CDO should do differently
- 3:04starting Monday morning. Audit every active AI initiative and ask,
- 3:09what data contract is this running on and who owns it? If the answer is unclear,
- 3:14shut it down or pause it. Expensive experiments become infrastructure only when the foundation
- 3:20is explicit. Start there. To go further, Leaders Insights offers strategic training programs,
- 3:26diagnostic tools, and an executive self-assessment designed for CMOs, CDOs, CFOs, and leaders
- 3:34aiming for board level performance. Explore everything at MBA-training.com.
- 3:39Thank you for listening to the Leaders Insights podcast. See you tomorrow for another episode.