2026年人工智能基础设施状况报告

What600 LeadersReveal AboutBuilding the Foundation for AI Success Executive Summary AI has entered its production era— but theinfrastructure powering it is holding many back. New research from Vanson Bourne, commissioned by DDN in partnership with Google Cloud and Cognizant, surveyed600 IT and business leaders across industries in the US to understand what’s helping and what’s hindering AI to scale.The findings are clear: AI success is no longer determined by algorithms or budgets — it’s determined by infrastructure. 97% 93% 98% 99% of surveyed IT andbusiness leaders reportinefficiencies in their AIworkloads. agree that cloud platformswill play a critical role inscaling their AI initiativesover the next year. are actively looking toreduce the energy impactof their AI workloads. acknowledge AI skillsgaps within their teams. As AI demand grows exponentially, data complexity,energy costs, and skill shortages are exposing the cracks intraditional infrastructure. Those who treat infrastructure asa strategic asset are able to see a quicker ROI, than thoseacting reactively. Those who don’t think strategically becometrapped in a cycle of inefficiency and underperformance —spending more time fixing than innovating. All of this leads to: 54% 65% having experienceddelayed or cancelledAI projects within thelast two years. claiming their AIenvironments are toocomplex to manage. AI success isn’t about model size or budget.It’s about theoptimized infrastructure beneath it. Table of Contents Executive Summary2 Infrastructure Challenges & Insights Section One:AI Infrastructure Has Become the Make-or-Break Factor4Section Two:Cloud Infrastructure Is Critical to AI Strategy10Section Three:AI Is Exposing Power, Cooling, and Efficiency Gaps14Section Four:Partnership Is the Accelerator of AI Success17 Practical Guidance & Implementation Section Five:Your AI Stack Starts Here - 6 Steps to Build Faster and Scale Smarter22 Appendix25 A. How DDN Can Help: Infrastructure That Works as Hard as Your AIB. About DDN and the Research PartnersC. Key Statistics from the 2026 State of AI Infrastructure ReportD. MethodologyE. About Vanson Bourne AI InfrastructureHas Becomethe Make-or-Break Factor of respondents report inefficiencies in their AI workloads, and more than half (54%)have delayed or cancelled AI initiatives in the last 24 months as a result.99% AIis growing faster than the systems designed to support it, andcomplexity slows down access to ROI. In fact, those who agree thattheir AI environments are too complex for their teams to manage canexpect to wait an extra three months to see ROI on AI investments. across multiple locations or environments remains a critical challenge— even as they seek to take advantage of distributed resources andaccelerators. With this mix of tools and access points, it’s no surprisethat pipelines, storage, and compute often sit in silos, rarely workingtogether. It slows innovation before it even starts. Most IT and business leaders manage their infrastructure like apatchwork of different tools rather than a single, unified ecosystem.Only 38% of respondents say they access their data through aunified data platform, which can be largely beneficial for efficencyand visibility. But even with these advantages, multi-region trainingand inference can still add additional levels of complexity that mustbe managed. For the rest without a unified platform, managing data The result of this complexity? A widening divide between those stillexperimenting with AI, and those scaling rapidly withAI factories(software-defined environments designed to deliver intelligence atindustrial scale). This shift signals that AI isn’t just a workload, it’sbecoming a full-stack operational model that depends on repeatability,observability, and efficiency from end to end. Jensen Huang,NVIDIACEO The Data AI Demands Is OutpacingInfrastructure Across Every Industry of leaders admit their AI environmentsarealready too complex for their teams to manage.65% But complexity isn’t being felt equally, and in someindustries, it’s becoming an existential threat to progress.For instance, in automotive and manufacturing,massivestreams of sensor and simulation dataare overwhelminglegacy systems.Public agenciesface governance andinteroperability challenges that slow every project.Financial firms are straining to meet real-time datademands for complianceand risk management. Evendata-driven industries likelife sciences reportthat thespeed and scale of AI research are surpassing whattraditional infrastructure can handle. This pressure is only growing. Over the next 12 months, AI workloads are set to more than double,growing by 110%across environments, with the sharpest increases expected in hybrid (+162%) or edge deployments (+227%)(see figure 1). For already stretched teams, that’s not just growth, it’s acceleration without control. We face challenges every day and must learn to deal with them. Integration, scalabilit