人工智能数据互联互通状况报告：2026年展望

Over 200 data and AI leaders say data infrastructure is the biggest The Top Line The August 2025 MIT report, The GenAI Divide: State of AI in Business 2025,1made waves amongbusiness leaders and AI product owners largely due to its headline statistic: 95% of generative AI pilotsat companies are failing. With the unprecedented scale of investment and the high expectations for While the accuracy of that specific statistic continues to be debated, the core issue it surfaces isnot: a large share of companies are failing to realize meaningful ROI from their AI efforts. The more We surveyed 200+ data and AI leaders, both from enterprises with internal AI adoption initiatives aswell as software companies embedding AI copilots and agents into their products. And here’s whatwe learned: enterprise AI is no longer limited by models. It’s constrained by data infrastructure and The strongest predictor of AI success in 2026 is the maturity of the underlying data In fact, 60% of companies at the highest level of AI maturity also have the most mature datainfrastructure. And the inverse is also true: 53% of companies with immature AI have immature In this report, AI maturity refers to the extent to which an organization has operationalized AI, movingbeyond experimentation to measurable business impact. Our framework considers dimensionssuch as model deployment, data integration maturity, governance, and ROI tracking. We categorize “ The paradox of AI readiness is that our data infrastructurebecomes more powerful not through endless adaptability,but through intentional semantic boundaries that give LLMsthe predictable contracts they need to orchestrate complex — Carlisia Campos, AI Software Engineer, Grokking Tech As anyone using enterprise AI tools like ChatGPT, LangChain, or Agentforce can attest, it’s no surprisethat context plays a defining role in AI maturity. Large language models depend heavily on it foraccurate, reliable, useful outputs. Whatissurprising is how few organizations are actually set up to Other findings from the research highlight the specific challenges standing between intention and The survey results point to a sobering truth: generative and agentic AI aren’t bottlenecked by thecapabilities of foundational AI models, but by access to connected, contextualized, controlled data.And the AI landscape is rife with data integration issues, from fragmented systems to a lack of That’s the bad news. The good news? There are enterprises and software providers that are gettingit right, and the survey surfaced the key initiatives, priorities, and investments behind their success.If you’re an enterprise looking to self-assess your AI maturity or the current state of out-of-the- The report is made up of two major parts: 1.Enterprise AI adoption and data challenges:A deep dive into how enterprise organizations are 2.Product AI strategy among software providers:An exploration of how product leaders areembedding AI into their platforms and why data integration remains a critical dependency. Together, these sections form a comprehensive picture of how data connectivity, infrastructurematurity, and integration strategy dictate AI success in both enterprise and product contexts. Table of Contents Survey Methodology and Respondent Demographics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6Part I: Enterprise AI Adoption and the Data Infrastructure Gap. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9Enterprise AI isn’t on the horizon: it’s in production. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9Stuck in the middle: most enterprises are implementing and scaling AI, but very few are leading. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10Knowledge assistance and customer service automation are the most prevalent applications of enterprise AI. . . . . . . . . . . . . . . . . . . .11 Survey Methodology and The insights in this report draw from two complementary surveys conducted in 2025; one capturingthe perspective of enterprise AI implementation leaders, and the other from product leadersat software providers. Together, they offer a dual view of how organizations are adopting andoperationalizing AI: from enterprises embedding AI into their operations, to software providers building Part I methodology:We used an independent research firm to blind survey 100 enterprise data and AIleaders, across industries and sizes ranging from startup to over $10B in annual recurring revenue. Nearly half (49%) of the respondents were C-level executives responsible for technology, IT, data,and AI functions. Including the 22% VPs and directors who responded to the survey, the data set is Seventy-four percent