Companies are investing heavily in AI, but only 6 percent of data and AI leaders believe that their infrastructure can actually support AI projects. This is according to a survey of more than 200 IT leaders conducted by CData.
The gap between AI ambitions and implementation appears to be greater than expected. Although 78 percent of organizations have now passed the pilot phase, only 17 percent have achieved a mature AI implementation in which ROI is measurable. CData’s study, “The State of AI Data Connectivity: 2026 Outlook,” shows that the biggest barrier is not AI models or vendor solutions, but the underlying data infrastructure.
However, it is not all doom and gloom. Of the companies that have achieved the highest level of AI maturity, 60 percent also have the most mature data infrastructure. The reverse is also true: 53 percent of organizations with immature AI implementations have weak data systems.
Data integration eats up time
A striking finding from the report is how much time AI teams spend on ‘data plumbing’. No less than 71 percent of teams spend more than a quarter of their implementation time on data integration. This involves tasks such as data modeling, setting up ETL pipelines, and configuring connectors.
This means that significant resources are tied up in infrastructure work rather than innovation. The impact of this is tangible: 71 percent of the organizations surveyed report higher costs and resource pressure due to integration challenges. In addition, 45 percent say that AI features are delayed by data connectivity issues.
“The era in which AI was limited by models is over. Today, AI is limited by data,” says Amit Sharma, CEO and co-founder of CData. “The organizations that win with AI are not those with the best algorithms, but those with connected, contextual, and semantically consistent data infrastructure.”
Real-time data as a hard requirement
All respondents agree: access to real-time data is essential for AI agents. Yet 20 percent do not yet have the capability to achieve this. The survey shows that organizations with real-time integrations are significantly more advanced in AI maturity. All companies in the highest AI maturity class (“leading”) support real-time integration, while this is the case for only 40 percent of the lowest category.
In addition, “AI-native” software vendors require three times as many external data integrations as traditional vendors (46 percent versus 15 percent have more than 26 integrations). This suggests that AI features are inherently difficult to integrate.
Semantic intelligence as a solution
Data access is therefore a requirement, but building a layer for this is apparently not a major stumbling block. No less than 83 percent of organizations have built a centralized layer for data access or are about to do so. So there is hope that organizations will eventually be AI-ready en masse.
In addition, 60 percent of organizations are investing in data governance, quality, and lineage as a top priority for AI readiness. This is more than double the number of those who want to build their own models and have this as a priority (9 percent). Real-time connectivity is in the top five investment priorities for 42 percent.
AI drives organizations that can connect, manage, and understand their data at scale. It is not the best AI models that make the difference, but the best “plumbing” around them.
Read also: The rise (and fall?) of shadow AI