A 2025 report from the MIT NANDA initiative reveals that 95% of enterprise generative AI pilots have failed to produce a measurable impact on profit-and-loss statements. This lack of return is often attributed to data gravity, a phenomenon where the cost and complexity of moving large datasets hinder the scalability of AI programs.
Many organizations encounter hidden costs 12 to 24 months after initial AI deployment. These expenses arise from the need to maintain data pipelines, ensure synchronization between source systems and copies, and manage governance controls. Such operational burdens frequently exceed initial budget projections for computing capacity.
The author suggests that companies should shift from a centralization-first strategy to an architecture that provides context where data already resides. This approach avoids the need to move restricted or regulated datasets, allowing AI systems to access information while respecting existing sovereignty and governance rules.
Before finalizing AI infrastructure investments, organizations are advised to evaluate four key factors: regulatory restrictions on data movement, existing governance ownership, synchronization requirements for data copies, and the feasibility of deploying intelligence directly to the data location.
