Data strategy: the deciding factor in AI leadership.

As AI expands from pilots to production, companies are discovering that outdated data architectures have become a serious obstacle to scaling AI. New global research from Cloudera finds that enterprise leaders are fundamentally reassessing their data foundations to support AI growth. In fact, Cloudera’s just-released survey of 1,500 IT architects reports that while 77% of organizations are actively using AI, nearly all (95%) have had to delay or cancel AI projects in the past year due to data governance, compliance, or regulatory challenges. To address these pain points, 72% say their current data architecture needs a “significant overhaul” to meet future AI requirements. This mass revamp of data infrastructure – what one industry report calls the “Great AI Re-Architecture” – highlights a broader truth: the ability to manage and deliver quality data is now a make-or-break factor in AI success.

A key driver of this architectural soul-searching is not raw computing power, but the demands of security, compliance and scale. In many firms, sensitive data and skyrocketing AI workloads have exposed the limits of one-size-fits-all cloud strategies. Executives rank security, data governance, and regulatory compliance as the top reasons for rethinking data architecture (42%), even above performance (35%) or scaling AI across the business (33%). To avoid new risks to their most sensitive data, organizations are embracing more flexible designs. This often means adopting hybrid cloud and multi-cloud environments that let them place AI workloads where it makes most sense for cost, speed, and compliance. Notably, two-thirds of companies surveyed have already moved some AI workloads from public cloud back to on-premises or private clouds within the last year. Hybrid architectures have in fact become the most common choice for hosting AI models, slightly edging out pure public cloud usage.

In practice, AI leaders are investing in “AI-ready” data platforms that break down silos and unify data across the enterprise. Approaches like the data lakehouse – which combines the scalability of data lakes with the governance and performance of data warehouses – are gaining traction as a way to ensure all types of data (structured and unstructured) are accessible to machine learning. Vendors are also evolving their tools to support these needs. Just this week, IBM announced a new capability to integrate vector similarity search (used for finding relevant information via AI embeddings) directly into its Netezza data warehouse platform. By incorporating this kind of semantic search function right into the database, enterprises can support AI applications like generative AI chatbots or recommendation systems without spinning up separate, specialized “vector” databases. Other major players are similarly doubling down on open data formats (for example, Apache Iceberg) and unified data catalogs that let different analytics and AI tools access one shared source of truth. The end goal is a data architecture that can flexibly deliver the right data to the right AI systems in real time, while maintaining consistent governance and security controls across environments.

The payoff for this re-engineering is not just technical efficiency, but strategic agility. A modern, well-structured data architecture allows organizations to rapidly experiment with new AI capabilities and scale successful pilots into enterprise-wide applications. As one report put it, the gap between AI leaders and everyone else will increasingly be shaped by architecture. In other words, companies that build data architectures capable of bringing AI to data “wherever it resides” will be far better positioned to achieve AI-driven growth, while those clinging to legacy data stack decisions risk falling behind.

While cutting-edge models grab headlines, many AI projects are quietly running aground on more mundane issues of data quality, governance, and ownership. Multiple studies in the past year show that as many as 70% to 85% of corporate AI initiatives fail to meet their objectives or ROI targets, and data problems are a primary culprit. Gartner recently reported that by the end of last year over half of all generative AI projects were abandoned at the proof-of-concept stage – in large part because of poor data quality, missing risk controls, or ballooning costs. And according to one multi-industry survey, 42% of U.S. enterprises have already scrapped at least one AI project specifically due to data issues. These sobering numbers reinforce what many CIOs have long suspected: if the data feeding an AI isn’t clean, well-governed, and accessible, the project is likely doomed before it ever gets off the ground.

The urgency of data governance and quality as AI enablers was a recurring theme in this week’s industry updates. Nearly all of the enterprises in Cloudera’s new survey cited earlier also pointed to data governance complexity as a growing barrier – 95% said they had delayed or canceled at least one AI initiative in the past year because of governance, compliance or regulatory issues. Moreover, 73% of those surveyed noted that integrating AI has made data governance even more complex across increasingly distributed data environments. Simply put, as AI use widens in an organization, it shines a harsh spotlight on inconsistent data definitions, quality errors, and unclear data ownership. Projects that looked promising in the lab often stall when they hit the reality of messy, siloed data and strict compliance requirements in production.

To turn AI aspirations into real value, leading companies are treating data governance and quality as strategic business imperatives rather than IT afterthoughts. In a recent report on AI readiness, 40.9% of global IT leaders said improving data governance is their top data priority for 2026 – even above specific AI investments. These organizations understand that better data practices translate directly into business outcomes like faster time-to-market and reduced risk. They are establishing unified governance frameworks that cover both data and AI systems in tandem, clarifying data ownership and accountability across the enterprise. As Info-Tech Research Group analyst Pooja Khandelwal put it, "Feed AI and automation noise, and they’ll scale confusion; feed them clarity, and they’ll scale intelligence". In practice, that means investing in data quality improvement, standardizing data definitions, and ensuring privacy and security controls are baked into every AI project from day one. With trustworthy, well-prepared data, AI initiatives have a far greater chance to move from experimentation to impactful deployment.

Another central insight for executives this week is that data itself has become both the ultimate competitive asset and a source of new obligations. As AI models rapidly commoditize, unique data is emerging as the key differentiator between companies that lead in AI and those that lag. Tech magnate Larry Ellison captured this sentiment in a now-viral statement, arguing that since most advanced AI models are trained on the same public internet information, "the true competitive advantage is no longer the model itself, but the ability to acquire exclusive proprietary datasets, which may be the only remaining moat". In other words, the long-term winners in the AI race will be those with valuable, hard-to-replicate data reserves. This perspective is already reshaping strategy: organizations are increasingly focused on expanding their data assets and preventing others from accessing them. Industry observers note a surge in data partnerships and acquisitions as AI companies rush to shore up their data moats. In fields from finance to healthcare, businesses that have spent years curating rich stores of domain-specific data – or that control streams of real-time customer and sensor data – are leveraging that trove to train superior AI models and services that competitors struggle to match.

But as companies double down on proprietary data as intellectual property, they face a parallel rise in responsibilities and regulations around that data. New laws are coming into effect that demand greater oversight of how data is used in AI. In the European Union, the first requirements of the sweeping AI Act took effect in early August, mandating heightened transparency for AI systems and granting regulators stronger authority to enforce data governance and compliance standards. On the U.S. side, California’s new AI law (SB 942), also effective this month, now requires companies to embed provenance data in AI-generated images, audio, and video content and provide disclosure mechanisms. These regulatory “guardrails” mean that organizations must treat data not only as a competitive advantage but as a regulated asset. In practice, that entails investing in robust data lineage tracking, consent management, and compliance auditing for AI initiatives. It also reinforces the importance of data localization and sovereignty considerations: with 84% of organizations prioritizing control over where their data resides and who can access it, many firms are weighing hybrid and local data architectures to meet regional data residency laws and protect privacy.

Ultimately, the twin trends of proprietary data moats and stricter data regulations are two sides of the same coin. For senior technology and data leaders, the takeaway is clear: success with AI depends on getting your data house in order. That means building unique, high-quality datasets that drive competitive advantage, while simultaneously ensuring those data assets are governed and utilized in a way that meets evolving compliance standards and ethical expectations. In an era where AI can amplify both the wisdom and the risks hidden in your data, the companies that thrive will be those that treat data strategy as central to business strategy.