The gap between AI adoption and AI readiness
Cloudera's newly released survey of 1,500 enterprise leaders, published August 11 under the banner "The Great AI Re-Architecture," puts a hard number on a problem most CIOs already feel in their bones: 95% of enterprises have delayed or canceled AI initiatives because of governance, compliance, or regulatory challenges. That statistic lands alongside a second one that makes it more striking, not less: 77% of the same organizations report actively using AI already. The problem is not adoption. It is that adoption has outrun the infrastructure and governance capacity needed to sustain it.
Sergio Gago, Cloudera's chief technology officer, framed the finding directly: "This current era of AI is forcing organizations to rethink the foundations of their technology infrastructure. Many enterprises are discovering that the architectures built for traditional analytics weren't designed for the scale, governance, and flexibility AI demands today." That is a pointed statement from a data infrastructure vendor, but the survey's own numbers back it up. Analytics-era architecture was built for query speed and reporting accuracy. AI-era architecture has to additionally support governed data movement, model traceability, and compliance controls that most legacy stacks never had to provide.
Governance, not model quality, is the actual bottleneck
The most useful number in the survey for anyone setting AI budget priorities is this one: 55% of enterprises delayed or canceled more than six AI projects in the past 12 months specifically due to governance issues. Model accuracy was not the cause. Compute cost, a real problem in its own right, was not the cause either. Governance was. That means the primary failure mode for enterprise AI projects right now sits squarely in organizational and architectural readiness, the ability to deploy models under acceptable compliance and control conditions, well ahead of any shortfall in the models themselves.
This finding should reorder how technology leaders sequence their AI investment. A team that spends its budget evaluating and licensing the best available models while treating data governance as a parallel, lower-priority workstream is optimizing the part of the problem that survey data flags as a secondary constraint. 73% of respondents report that AI has made data governance meaningfully more complex, and 97% now move data between environments at least monthly. Governance tooling and data lineage capability have moved out of the back office. They now sit on the critical path to shipping AI at all, ahead of model selection, prompt design, or any of the work that typically absorbs an AI team's early attention.
The retreat from public-cloud-only architecture
One of the more counterintuitive findings is that 66% of enterprises have shifted AI workloads from public cloud back to private cloud or on-premises environments. For the past decade, the default enterprise architecture conversation assumed a one-way migration toward public cloud, with on-premises infrastructure treated as legacy debt to be retired rather than a deliberate architectural choice. The Cloudera data suggests AI workloads are breaking that assumption, with cost, data sovereignty, and control concerns pulling some workloads back on-premises even as overall cloud spending continues to rise elsewhere in the business.
This is not a rejection of cloud computing. It is a maturing recognition that different AI workloads have different governance and latency requirements, and that a single-environment strategy cannot serve all of them well. 84% of respondents report that AI workloads have increased their infrastructure costs, and 25% say they will prioritize hybrid-first architecture over the next two years. Read together, these numbers describe an enterprise landscape that is actively rearchitecting itself around AI's actual requirements, rather than retrofitting AI onto whatever infrastructure happened to be in place already.
What Liberty Mutual's approach signals about the fix
The survey's write-up points to Liberty Mutual as an example of enterprises adapting rather than starting over: the insurer has built model-agnostic layers on top of its existing mainframe systems, rather than attempting a wholesale replacement of core infrastructure before deploying AI. That approach lets a company preserve decades of institutional investment in core systems while still gaining the flexibility to route AI workloads to whichever environment and model best fits the task, public cloud, private cloud, or on-premises, and it does so without waiting years for a core systems replacement project to clear the runway first.
That model-agnostic, environment-agnostic pattern is a more realistic fix for most large enterprises than a multi-year infrastructure replacement project. Ripping out legacy core systems to build AI-native infrastructure from scratch is rarely fundable or fast enough to matter competitively. Building an abstraction layer that lets AI workloads move freely across whatever infrastructure already exists, while still meeting governance requirements, is the more pragmatic version of the same goal, and it is the pattern the survey suggests is gaining ground.
The decision this data forces onto the next budget cycle
For CIOs building next year's technology budget, this survey is an argument for funding data architecture and governance capability ahead of, or at minimum alongside, additional AI model licensing and experimentation spend. The organizations Cloudera surveyed did not fail to adopt AI. They adopted it, ran into governance and architecture limits, and then had to cancel or delay projects after the fact, which is a more expensive and more disruptive sequence than addressing those limits up front.
The practical takeaway is straightforward even if the execution is not: audit whether your data architecture can support governed, traceable AI deployment before committing to the next wave of AI projects, and treat a hybrid, model-agnostic infrastructure layer as a prerequisite rather than a nice-to-have. The enterprises in this survey that are furthest ahead are not the ones with the most advanced models. They are the ones that fixed their foundation first.
