40% of Enterprises Now Say Infrastructure Is What Holds AI Back, Up From 9% in 2024

Two years ago, almost nobody blamed the data center. When Digital Realty's 2026 Global Data Insights Survey asked 2,131 IT decision-makers across 19 countries what was holding their AI initiatives back, 40% named a lack of specialized infrastructure as the primary constraint. In 2024 the same answer came from 9%. That shift happened in the same two years that most organizations spent treating AI as a software and talent problem.

The Bottleneck Moved From the Model to the Floor

The shift makes sense once you look at what respondents are now trying to do. Seventy-nine percent expect to deploy AI initiatives in 2026, and 98% expect to be running real-time AI applications within 12 months. Real-time inference does not tolerate data sitting in the wrong place, which is why 92% of respondents now tie data location decisions directly to their AI plans, up from 73% two years earlier. Eighty-eight percent have adopted a distributed data strategy, 86% are pursuing sovereign AI initiatives, and more than half host AI workloads in private cloud environments.

That is a very different estate from the one most AI pilots were built on. A proof of concept can run on a single cloud region with a copied dataset. A production system serving real-time decisions across several regions needs the data, the compute, and the network path between them to sit where regulation, latency, and cost all allow. Digital Realty's Chief Revenue Officer, Colin McLean, summed it up: "Infrastructure beneath AI is now the gatekeeper between promise and payoff." Respondents are not pulling back either. They expect AI spending to rise 32% over the next year, and only 3% report no measurable return from AI so far. The money is there. The floor it has to run on is not.

Most Teams Think They Are Ready. Fewer Than Four in Ten Are.

If infrastructure is the problem, why did it take until 2026 to show up in the surveys? Seagate's Data Infrastructure Readiness Report 2026, fielded by Recon Analytics with 2,712 enterprise technology decision-makers across seven countries in May and June, offers an answer. Eighty-three percent of respondents say they are fully or mostly prepared for AI's demands. Only 38% say they are fully prepared for its long-term data demands. The gap between "mostly" and "fully" is where the infrastructure surprise has been hiding.

The demand side of that equation is not in doubt. Ninety-nine percent expect AI to increase their storage requirements over the next three years, 70% expect those requirements to grow by at least 26%, and 32% expect growth of more than 50%. Forty-three percent already identify storage infrastructure as a key constraint, and 53% cite data quality and readiness as their leading deployment challenge. Sustainability is complicating the build-out as well: 77% have delayed or restructured an expansion because of sustainability concerns, and only 39% describe their storage operations as very sustainable today. Organizations that were "mostly" ready for a pilot are discovering what "mostly" costs at production scale.

The Data That AI Needs Is the Data Nobody Manages

The hardest part of that data is the part that never lived in a database. A Nasuni survey of 1,000 senior IT and procurement decision-makers found 94% struggle to manage unstructured data, the documents, design files, and media that carry most of an enterprise's institutional knowledge. Only 16% treat managing it as a strategic IT priority. Seventy-nine percent report inconsistent file access across locations, 90% cite data security, integration, and trust as barriers to scaling AI, and only 43% of AI initiatives meet their original objectives. Nearly every firm is piloting AI agents, but just 18% have deployed them at scale.

Cloudera's Data Readiness Index, a survey of 1,270 IT leaders at companies with more than 1,000 employees, shows the same confidence pattern Seagate found. Eighty-five percent say they have a clear data strategy and 84% feel confident in their data's accuracy and completeness, yet only 18% say their data is fully governed and nearly four in five admit their AI initiatives are constrained by limited data access across environments. "AI is only as effective as the data that fuels it," said Sergio Gago, Cloudera's Chief Technology Officer. A distributed data strategy that 88% of enterprises have adopted is only an advantage if the data in every one of those locations can actually be reached, governed, and secured.

• Separate "mostly ready" from "fully ready" in your own planning. Seagate's 45-point gap between the two answers is where storage, power, and network shortfalls surface at production scale.
• Model storage growth before approving the next AI program. With 70% of Seagate's respondents expecting at least 26% more storage within three years, capacity planning belongs in the business case, not the post-mortem.
• Treat data location as an architecture decision, not a procurement one. Ninety-two percent now tie location to AI plans; decide where data must live for latency and sovereignty before choosing where compute runs.
• Put unstructured data on the strategic agenda. Nasuni found 94% struggle with it and only 16% prioritize it, which is the single widest gap in any of these surveys.
• Make every new data location a security boundary from day one. A distributed, sovereign, partly private-cloud estate multiplies the places policy has to hold, and 90% already cite security and trust as barriers to scale.
• Measure governance, not confidence. Cloudera found 84% confident in their data and only 18% fully governed; report the second number to the board, not the first.

Nine percent to 40% in two years is not a trend line that flattens on its own. The organizations that looked ready for AI in 2024 were mostly being asked to run pilots, and pilots hide infrastructure problems well. Production does not. The enterprises that close this gap in the next twelve months will be the ones that stop asking whether their models are good enough and start asking whether the floor underneath them can carry the weight.