The Castle Builders of AI

The Independent was not involved in the creation of this sponsored content.

Data is hailed as the new moat of business: a defensive trench that doubles as the reservoir powering agentic and generative AI. The metaphor is appealing, and 95% of executives say they intend to build such fortifications by turning themselves into AI and data platforms within three years—seven in 10 want to do so in less than one year. Yet only 13% of these enterprise leaders have figured out how to keep their moats full. These firms reap five times the return on investment from generative and agentic AI, and mainstream twice as many applications as their peers. The rest, for all their ambition, are digging ditches that drain as quickly as they are dug.

So what separates the leaders from the laggards? These three findings, from EDB’s “Sovereignty Matters” study of 2,050 executives across 13 economies representing ~$48T in GDP, show the difference.

Sovereignty first—or not at all

The first mistake of the 87% is failing to treat sovereignty over data as a foundational need. A leaking moat is a liability, not protection. In modern terms, enterprises that do not centralize, secure, and make governed access to their data possible across silos simply lack the material to defend themselves or fuel AI.

Among the winning 13%, nine out of 10 made sovereignty a mission-critical priority. Operationally, having sovereignty means that the enterprise decides where data and AI run, who can use them, and under what rules—consistently across clouds, data centers, and the edge. Data is available in real-time, integrated across functions, and protected from misuse.

The laggards, by contrast, did not view sovereignty with the same weight of explicit criticality. They were still experimenting with piecemeal solutions and had less than a one-in-10 chance of success in moving their AI projects beyond experimentation or initial pilots. The lesson is stark: “Before companies can dream of AI-driven moats, they must decide that owning, securing, and governing their data is the ground on which everything else is built,” says Kevin Dallas, CEO of EDB.

Governance as glue

The second stumbling block is governance—or, rather, the absence of it. Too many firms treat governance as a compliance afterthought, when in fact it is the very layer that allows experiments to scale into production.

In the laboratory, experiments often rely on synthetic or isolated data sets. But in the real world, data is messy, regulated, and dynamic. Regulatory and compliance obligations turn governance from a nice-to-have into an existential necessity. Yet, insights from EDB’s “Sovereignty Matters” research reveal that many enterprises aiming to evolve into data platforms often overlook critical elements like security and governance in their sovereign architectures.

This explains why many experiments never leave the lab. Without governance frameworks for security, external interactions, data quality, and hybrid operations, even the most innovative agentic AI prototypes collapse when tested against real-world conditions. By contrast, the 13% who succeed are far more likely to run hybrid architectures—40% or more of their workloads span multiple environments—and nearly one in five are managing critical workloads in open source databases such as Postgres^®.

As Dallas puts it, “The governance layer is what enables agentic and GenAI to move from incubation to mainstream production. It is what bonds all the pieces together as a working moat for your business.”

Thinking ecosystems, not pieces

The final reason most firms fail is cultural rather than technical. They think in pieces instead of platforms. Many companies design AI projects as discrete initiatives, hoping that success in one area can later be replicated elsewhere. The result resembles a collection of bottle rockets—entertaining bursts, but no lasting propulsion.

The winners, by contrast, design for scale from the outset. Their agentic and generative AI projects are built to operate across functions, with the architecture to handle wider adoption. This allows them to double the range of mainstream applications compared with their peers, and to deploy them at higher density—66% of a given AI initiative in mainstream use, versus half that in less successful firms.

This ecosystem mindset echoes lessons from history. Medieval castles evolved from crude walled camps into intricate urban centers. Their moats were not barriers alone but channels of trade, lifelines for the surrounding community. The best castles were built on elevated ground, offering visibility and control. In the same way, companies that create AI and data moats as ecosystems—secure, integrated, and scalable—gain the high ground. They see both opportunity and liability more clearly and can defend and exploit them more effectively.

“The best castle builders of AI run on unified platforms where data and AI work as one—often in hybrid architectures, which enable seamless movement of workloads across clouds, on-premise, and the edge to optimize for compliance, efficiency, performance, and resilience," explains Dallas.

The high ground ahead

The global data on these differences is instructive. Success rates are not evenly distributed. Only 10% of United Kingdom firms with more than 500 employees had managed to make data into a moat, compared with 17% in the United Arab Emirates and Saudi Arabia. Yet the patterns of success are universal. Whether in Europe, Asia, or the Middle East, sovereignty, governance, and ecosystem thinking separate the 13% from the rest.

The urgency is only growing. With most enterprises determined to become AI and data platforms within three years, and seven in 10 demanding results within one, the moat-building race is underway. But unless firms treat sovereignty as the foundation, governance as the glue, and ecosystem design as the blueprint, most will find themselves digging ditches that never hold water.

Sovereignty over data is not about isolation; it is about control. Like the bridges across a medieval moat, data sovereignty allows enterprises to open themselves selectively to the outside world, enabling both trade and defense. In the age of generative and agentic AI, those bridges are vital.

The conclusion is as unromantic as it is urgent: If data is to be the moat, build it properly—or risk being outpaced by those who do.