Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Your data foundation matters more than your model

The AI infrastructure gap.

[Photo: Getty Images]

Profile photo of Haroldo Jacobovicz

Founder · Arlequim Technologies S/A

It seems like a new version of an AI model arrives every few weeks, claiming to beat the last one and any of the competitors. And every few weeks, teams ask the same question: Should we switch?

For most, it’s the wrong question. The fiercest, most visible race in artificial intelligence is over the best model, the newest capability, the highest benchmark. Underneath that race is a question that matters far more: Can the data and infrastructure a company connects to a given model turn it into a useful result? A firm can license the most advanced model on the market and still get mediocre output if its data is fragmented, stale, or locked in systems that don’t talk to each other.

MIT’s NANDA initiative reported in 2025 that 95% of enterprise generative AI pilots delivered no measurable impact on profit or loss. The researchers were direct about the cause: The models were rarely the weak link. What failed was the fit between those tools and the way organizations work.

A FORMULA 1 CAR ON A DIRT ROAD

It’s like putting a Formula 1 car on a poorly maintained dirt road. The engineering under the hood is extraordinary, but the road ultimately decides how fast you go.

The road, in this case, is the data architecture, cloud infrastructure, compute, system integration, and the governance that controls what information a model can reach. When those are weak, a better engine produces the same outcome: a powerful machine crawling over bad ground.

In Gartner’s survey of 1,203 data leaders, 63% of organizations either lack the right data management practices for AI or are unsure whether they have them, and Gartner projects that 60% of AI projects will be abandoned through 2026 without what it calls AI-ready data. At this point, projects rarely fail because the model wasn’t smart enough, but they fail nonetheless.

AI-READY ISN’T THE SAME AS A LOT OF DATA

Having a lot of data is not the same as being ready for AI.

AI-ready data needs to reflect the actual use case, including the errors, outliers, and edge cases a model needs to learn from. Volume doesn’t deliver that.

Many companies sit on enormous stores of information that are duplicated across departments, inconsistent between systems, badly labeled, or simply wrong. Feeding large amounts of that data into a generative AI assistant doesn’t improve its answers. The old rule holds with new force: garbage in, garbage out.

Being AI-ready means close to the opposite of a data dump. Data is organized, current, reachable by the systems that need it, and governed so people know what it contains and who may use it. That’s meticulous, unglamorous work. It’s also the work that decides whether the model on top produces anything of value.

THE ADVANTAGE OF BORING INVESTMENTS

There is an uncomfortable truth here for leaders who feel behind on AI: The companies best positioned for AI today are often the ones that made data infrastructure investments years ago.

A business that modernized its cloud, cleaned up its data, and integrated its systems five years ago did not do it for AI. But that organizational work has become one of the most valuable things it owns. A competitor that skipped those steps and rushed to buy access to the newest model is now finding that the model has nothing solid to stand on.

In four decades in technology, I have watched this cycle repeat. Each wave of innovation pulls attention to the most visible layer, and the companies that capture the most value are usually the ones that invested earlier in the layers underneath, where nobody was looking.

This doesn’t mean the model doesn’t matter. Of course it does. Beyond a certain level of capability, though, switching to a marginally better model produces smaller gains than fixing the foundation it runs on, and at an unnecessary cost. A company can choose a model in an afternoon. The foundation the model runs on takes time to build and must be lived with for years.

DATA AND DURABILITY

The market rewards what it can see, and right now it’s watching the models. The more durable advantage is forming out of view, in the data and infrastructure that decide how well any model performs.

Over the coming years, the companies that win with AI will be the ones that did the quiet infrastructure work first. They cleaned up their data and integrated their systems while everyone else compared model benchmarks.

Haroldo Jacobovicz is the founder of Arlequim Technologies.