Cloudera survey: 89% of enterprises redesigned AI use cases over costs

Cloudera has published its State of Enterprise AI 2026 report, drawing on a survey of 1,500 global IT decision-makers commissioned from Wakefield Research in September 2026. The headline finding is stark: 89% of respondents have already limited or fundamentally redesigned AI use cases because of rising token consumption costs, a figure that suggests the economics of generative AI in production are proving harder to manage than many organisations anticipated when they began deployments.

The report paints a picture of enterprises caught in what Cloudera frames as a "scale trap": early pilots succeed, workloads are promoted to production, and then token costs, governance overhead and data-sovereignty obligations compound in ways that were not modelled at inception.

The cost and governance squeeze

Token economics sit at the centre of the report's findings. Forty-five per cent of respondents said they had already moved AI workloads in response to higher or more volatile costs, a significant operational burden given that 58% also reported that relocating workloads requires moderate or major redesign. Governance pressure is equally acute: 75% cited compliance and governance as a constraint on scaling AI, while 55% said governance costs had grown faster than expected.

The most striking single statistic in the survey may be that 98% of respondents had pulled back at least one AI deployment due to failure, hallucinations, or unexpected model behaviour. That near-universal figure implies that the industry is still working through a maturation phase in which production readiness remains genuinely difficult to achieve and maintain.

Data sovereignty is reshaping workload architecture as well. Eighty-four per cent of respondents said residency, sovereignty, or jurisdiction-specific requirements influence where AI workloads run, a finding consistent with accelerating regulatory activity in the EU, UK and a number of Asia-Pacific markets.

Brian Martin, CTO of Applied AI at Cloudera, said: "Costs change, models evolve, regulations shift, and the right place to run a workload changes with them. Leaders need an AI strategy built for that reality, one that gives them the freedom to adapt without giving up control of their data, governance, or spend."

Market context and competitive landscape

Cloudera is not a neutral party here: the report is a marketing asset designed to support its hybrid and multi-cloud data platform positioning. The vendor's core argument, that enterprises should run compute close to governed data rather than move data to AI, is a direct competitive response to hyperscaler pull-in strategies from AWS, Microsoft Azure and Google Cloud, each of which benefits when customers consolidate AI workloads on a single public cloud.

The findings nevertheless align with a broader pattern visible across the enterprise software market. Vendors including Databricks, Snowflake and IBM have each published research or made product announcements in 2026 addressing governance friction and cost predictability for AI workloads. The convergence of messaging suggests a genuine market pain point, even if individual survey methodologies favour the commissioning vendor's narrative.

On the regulatory side, the EU AI Act's obligations for general-purpose AI models are phasing in through 2026 and 2027, and the UK's proposed AI liability framework is under consultation. Both regimes raise the compliance stakes for enterprises running customer-facing or high-risk AI applications, lending credibility to the 75% governance-constraint figure.

If given the chance to restart their AI journey, 49% of respondents said they would embed governance earlier, 46% would prioritise data quality from the outset, and 45% would modernise infrastructure sooner. These retrospective preferences point to an emerging consensus that the foundational work, data pipelines, access controls, cost attribution, and model-monitoring tooling, was systematically under-resourced in the first wave of enterprise AI adoption. The next phase of enterprise AI investment is likely to flow disproportionately toward those layers.