The AI confidence trap: why enterprises are still missing the mark

Although enterprise AI adoption continues to gather momentum, questions remain about the ability of those efforts to achieve success. According to a survey HPE released in 2024, enterprise overconfidence in their AI capabilities created significant blind spots that threatened deployment benefits.

So, what’s changed over the past year? To see if and how, enterprises have evolved their AI adoption approaches since the previous survey, HPE fielded a new study, ONE YEAR ON: Architecting an AI advantage. It surveyed 1,775 IT leaders across nine global markets to offer an updated perspective on progress and stumbling blocks.

Download ONE YEAR ON: Architecting an AI Advantage Report

The survey found that enterprise AI initiatives have both progressed and regressed. More organizations have operationalized AI (22% currently vs 15% last year). Yet fewer than half of enterprises term their overall deployment efforts a success, with over a third (35-40%) of deployment use cases only garnering limited success.

To understand the reasons enterprises still struggle with achieving success, the study took a closer look at the numbers and came up with several key insights.

Fragmented approach continues to impede benefits

Despite a clear majority of enterprises (72%) recognizing that a holistic AI approach —meaning a left-to-right view for strategy, resource considerations, and infrastructure investments— is vital to netting intended benefits, many are pursuing the opposite. In fact, a third (34%) of respondents say their businesses still have multiple, separate AI strategies.

Further, goalsetting is equally fractured. Less than half (42%) of enterprises have collaborated on the best practice of creating a single set of AI goals. More concerning, over a third (35%) are actually actively setting (or planning to set) separate goals across their lines of business.

“Enterprises that approach AI with a holistic strategy that aligns goals, data architecture, and governance are best positioned to build agile, scalable platforms capable of handling massive data volumes, ensuring sovereignty, and unlocking AI’s full potential,” said Brian Gruttadauria, CTO for Hybrid Cloud at HPE. “In contrast, those taking a fragmented or siloed path risk accumulating technical debt that could take years and significant investment to overcome.”

Data maturity still lags

Like last year, enterprises still recognize robust data management capabilities as one of the most critical elements for AI success.

However, data maturity—while improved—remains concerningly low. For example, fewer than half (45%) of organizations can run real-time data pushes/pulls to enable innovation and external data monetization.

It’s a similar story for establishing data governance models that can run advanced analytics, where only 43% report competence. Much the same findings are revealed in shared data models and centralized business intelligence, where just 41% report competence.

Infrastructure barriers persist

The introduction of generative AI and, more recently, agentic AI touched off a flurry of innovation by compute and networking providers. New solutions exist that are aimed at rapidly moving beyond pilots to building and scaling AI in production and do so by addressing capacity, performance, governance, security and simplicity.

However, enterprise fluency has failed to keep up. For instance, fewer than half of survey respondents are confident in their ability to understand their actual compute or networking requirements for most phases of the AI lifecycle.

Further, a significant percentage (42%) still operate under the notion that they can rely on a mix of components assembled internally, from existing infrastructure, despite most organizations lacking the in-house expertise required for designing, developing, and deploying AI-appropriate, fit-for-purpose compute and networking infrastructure.

Unsurprisingly, this is reflected in the survey’s findings around the ability to launch and evolve AI initiatives in production. Nearly half (45%) of enterprises say they’re already struggling to manage their AI projects pipeline, with a similar number (47%) questioning their organization’s ability to scale AI projects effectively.

Compliance, ethics, and security take a step backwards

A holistic AI approach also addresses ethics, compliance, and security, including data sovereignty related to running workloads in public vs private clouds. Yet, when it comes to developing their AI strategies, a concerning percentage of enterprises are collaborating with their legal and HR stakeholders less often, instead of more.

• This year, nearly a third (30%) say their legal teams are uninvolved in setting AI strategies, as opposed to only 21% last year.
• From an ethics perspective, almost 4 in 10 enterprises (39%) now say their HR teams are uninvolved, whereas less than a third (32%) reported omitting HR in 2024.
• Equally concerning is the drop in CISO-level involvement in AI strategy development. Last year almost half (46%) of enterprises included their CISO, or equivalent leadership, in AI decision-making. Now it’s just 36%. This finding was surprising as, intuitively, CISO involvement would be expected to rise given today’s complex threat environment.

“No matter how successful an AI prototype is at passing tests in isolation on synthetic data, it can all be undermined if the developers fail to pressure test their models against the real-world security, regulatory, and IT conditions of a particular enterprise,” said Kirk Bresniker, HPE Fellow and HPE Labs Chief Architect, “Have you engineered robustness? Prepared for day 1 and day 3650? Run the supply chain due diligence on the model that you'd run on minerals? With risks of damaged brand reputation, security breaches, legal battles and fines, it’s imperative for enterprises to push back on their developers and vendors by simply insisting on modern compliance, ethics, and data security best practices."

Put proven AI expertise to work

Regardless of where enterprises are on their journey, most enterprises will either increase their AI budget over the next 12 months or even double it—according to the survey. Although this momentum is positive, properly directing investment is critical for turning tactical deployments into strategic successes that enable fully realize AI rewards.

Years of AI-centric expertise at HPE has demonstrated that a holistic approach is key. Further, a holistic approach is enabled by validated, pre-integrated, modular AI infrastructure solutions, along with strong consultancy services that are already proven in the real-world.

To find out more about how HPE can help drive your company’s AI ambitions, see www.HPE.com/AI/Insights. To gain a fuller perspective of the survey and its results, go to ONE YEAR ON: Architecting an AI advantage.