Executives have poured money into artificial intelligence projects for years. Returns remain elusive. A new study from Confluent lays bare the issue. Most organizations do not face an investment shortfall. They confront a data shortfall.
Three in four IT leaders, or 72%, report that poor real-time data infrastructure stops them from scaling AI effectively. The findings come from TechRadar. Real-time data processing tops the list of barriers at 72%. Data lineage uncertainty follows at 66%. Fragmented data ownership sits at 65%. Only 32% of organizations have agentic AI systems in production. The rest endure delays and disappointing returns on investment.
Shaun Clowes serves as chief product officer at Confluent. “Models need to be connected to the systems, events and signals that reflect what is happening across the business,” he said. Current setups were never built for continuous intelligence. That explains why companies of every size and sector hit the same walls.
But the data problem runs deeper than pipelines and ownership. Legacy systems cannot feed AI the fresh, accurate, contextual information it demands. And infrastructure demands now compound the trouble. Power. Cooling. Connectivity. These constraints have moved from background noise to front-page obstacles.
Deloitte’s 2025 AI Infrastructure Survey captured the shift. Power and grid capacity stand as very or extremely challenging for 72% of data center and power company executives. AI data centers in the US could need 123 gigawatts by 2035. That marks a thirtyfold jump from 4 gigawatts in 2024. A single five-acre facility with GPUs might draw 50 megawatts. Traditional centers needed just 5. Deloitte detailed the gaps.
Supply chain disruptions worry 65% of respondents. Security concerns affect 64%. Skilled labor shortages hit data center operators hardest at 63%. Seven-year waits for grid connections have become common. Meanwhile data centers can finish construction in one to two years. The mismatch creates bottlenecks that no amount of capital seems able to fix quickly.
CoreSite’s analysis of 2025 drove the point home. That year AI stopped being a headline story about new models. It became an infrastructure problem. Hyperscalers raced ahead. Enterprises scrambled to answer basic questions about power, cooling, connectivity and the movement of petabytes of data. Juan Font, president and CEO of CoreSite, captured the mood. “I see urgent questions emerging about the future of AI infrastructure – ones that need actionable, concrete answers, and quickly.” CoreSite outlined the stakes.
Colocation facilities have emerged as one practical response. They offer dense power and advanced cooling without forcing enterprises to rebuild their own data rooms. High-speed links to multiple cloud providers and AI services come built in. Flexibility follows. Organizations can shift workloads or swap models without tearing out hardware. The approach turns infrastructure from a fixed cost into a variable capability.
Environmental costs add another layer of pressure. Researchers at Cornell University modeled the trajectory. By 2030, unchecked AI computing growth could release 24 to 44 million metric tons of carbon dioxide each year. That equals the emissions of 5 to 10 million additional cars on American roads. Water consumption could reach 731 to 1,125 million cubic meters annually. The volume matches household use for 6 to 10 million people. Fengqi You led the study. “Artificial intelligence is changing every sector of society, but its rapid growth comes with a real footprint in energy, water and carbon,” he said. “This is the build-out moment. The AI infrastructure choices we make this decade will decide whether AI accelerates climate progress or becomes a new environmental burden.” The work appeared in Cornell Chronicle.
Smart siting, grid decarbonization and operational efficiency could cut carbon emissions by roughly 73% and water use by 86% compared with worst-case paths. Yet those choices require coordination among industry, utilities and regulators that has so far proven elusive.
Data quality issues have worsened in parallel. BARC research showed the share of organizations citing data quality as a top AI challenge more than doubled from 19% in 2024 to 44% in 2025. Gartner had already warned that organizations would abandon 60% of AI projects through 2026 if those projects lacked AI-ready data. Production rates hover around 48% for AI initiatives overall. Chief data officers report that 65% cannot meet their goals this year. Almost all, 98%, have suffered major data-quality incidents.
The pattern repeats across reports. Flexential’s surveys highlight skills gaps around high-density infrastructure. Networking bottlenecks and scaling difficulties delay time to revenue. CIO Magazine noted that AI’s appetite for data has broken traditional storage systems. Companies now race to rebuild architectures for speed and reliability before budgets collapse. McKinsey projects AI will drive nearly 70% of data center capacity growth through 2030. The World Economic Forum sees the global data center market more than doubling to $584 billion by 2032.
Yet spending alone has not translated into value. MIT’s Project NANDA found 95% of organizations report zero measurable return from generative AI investments. Grant Thornton points to weak prioritization, governance and operational integration as root causes. The technology works. The surrounding systems do not.
So what separates the few organizations seeing progress? They treat data foundations as seriously as model training. Eighty percent of respondents in the Confluent study now prioritize enterprise data for AI initiatives. Data streaming platforms earn endorsement from 88% of IT leaders. That outranks even AI and machine learning tools themselves at 82%. The message is clear. Better data infrastructure unlocks everything else.
Power remains the ultimate constraint for many. Goldman Sachs analysis estimates global data center power demand will surge 160% by 2030. Marginal supply increases cannot keep pace. Data centers could account for 40% of US power demand growth in 2026. The gap between need and buildable capacity widens fast. AEI warned that the race to construct new facilities has accelerated electricity requirements to alarming levels.
Water scarcity presents a parallel risk. A single large AI data center can consume millions of gallons daily for cooling. Projections for 2028 reach 32 billion gallons annually across US AI-related facilities. That volume could support indoor water use for 360,000 households. Communities in growth corridors already feel the strain. Residential electricity prices have risen faster than average in eight of nine major data center markets.
These pressures have pushed some leaders to rethink location strategy. Midwest states with abundant wind power and water resources score well on combined carbon and water metrics. New York benefits from a low-carbon grid but must adopt water-efficient cooling methods. Choices made now will shape outcomes for the rest of the decade.
Enterprises that once viewed AI through the lens of experimentation have shifted to production expectations. The infrastructure to support those expectations lags. Complexity has grown unmanageable for many. DDN’s 2026 State of AI Infrastructure Report found 99% of organizations report inefficiencies in AI workloads. Sixty-five percent say their environments have become too complex to manage. More than half have delayed or canceled projects because of infrastructure limitations. Energy consumption, data movement and GPU underutilization now outweigh model capability or budget as constraints.
The picture that emerges is sobering. Billions flow into GPUs, cloud credits and consulting engagements. Many projects stall at proof of concept. Others reach production only to deliver marginal gains. The common thread is rarely the algorithm. It is the data that feeds it and the physical plant that runs it.
Fixing the problem demands more than incremental upgrades. Organizations must modernize data architectures for real-time flows, clear lineage and unified ownership. They must secure access to dense, flexible infrastructure capable of handling high-density AI workloads without prohibitive delays. And they must weigh environmental impacts in every siting and design decision.
Those who act on these fronts stand to pull ahead. The rest risk watching their substantial AI investments generate more heat than value. The data crisis is here. Infrastructure limitations have made it impossible to ignore.
WebProNews is an iEntry Publication