Enterprise leaders examining constraints on their AI investments’ performance could be looking for the wrong culprits. Executives might point the finger of blame at LLM models, GPUs, and compute capacity, without identifying quieter, more fundamental bottlenecks to AI success, such as legacy data infrastructures and tactical approaches to their upkeep.
Companies can certainly store enough information for yesterday’s data access needs but cannot reliably deliver the real-time context and governance that AI applications require.
This performance gap has evolved because enterprises have treated file storage as a passive layer that is refreshed in spending cycles and patched up on an ad hoc basis. Until recently, data requests were driven by humans, which older data infrastructures could cope with. In contrast, today’s AI applications demand continuous, system-level access to complete and consistent data sources. Many organisations assume their tech stack meets AI’s demands with data storage infrastructures that, in reality, are outdated in terms of strategic investment, access and governance of data.
Outdated thinking
Our research of 1000 organisations in US, UK, France, and DACH shows how many C-suites have yet to identify and respond to these emerging obstacles to AI success.
For example, many enterprises haven’t yet committed to investing strategically in data layers that underpin higher AI performance. In our survey, almost half (46%) of organisations say their #1 storage infrastructure challenge is the need to boost spending because of data growth (46%). Slightly fewer – 43% – say their top challenge is balancing their budget between AI and storage infrastructure, while the same number (43%) say their biggest worry is the strain that regular storage hardware refreshes or expansion cycles put on their daily operations.
The urgent need for boards to prioritise infrastructure spending is accompanied by other silent obstacles – the unprecedented pressures AI technologies are putting on enterprises’ information access capabilities.
Obstacles to AI success
AI tools create a raft of pressures – surging data volumes, increased access frequency, and higher data governance demands. These factors can be managed in isolation, but together, they expose the limits of organisations’ fragmented data infrastructure.
Overconfidence in managing data volumes
Despite the AI economy’s surging data levels, senior executives’ overconfidence in their organisation’s ability to deal with service disruption suggests they have yet to fully understand these infrastructure-level risks to their AI ambitions.
While two-thirds (66%) of organisations believe they can recover their unstructured data from service disruption or downtime, only four in ten (38%) have the systems in place to minimise downtime and achieve a rapid recovery. This contradiction suggests many companies haven’t fully grasped the complexities of using traditional backup amid the AI economy’s increased data volumes.
Unstructured data unleashes AI potential
Industry leaders have long forecast that enterprises which consolidate their unstructured data – the images, video, and R&D data that make up 80% of organisations’ information – will enable their AI models to harness this undiscovered intellectual property for game-changing competitive edge. Here again, survey responses revealed another silent obstacle to improved AI performance.
Our study found that an overwhelming 94% of companies struggle to manage their unstructured data, but only 16% rank it as a top investment priority. This paradox highlights continued under-investment in the data foundations that could enable AI tools’ success since AI requires data access that preserves full context across many different systems and locations. But as AI usage expands, slow retrieval of data sets, which used to be viewed as a low-level inefficiency, will become an organisation-wide barrier to growth.
Centralisation for better governance
With its demand for consistent data access and security guardrails across organisations’ data sets and storage infrastructures, AI has pushed data governance and security to the top of the enterprise IT agenda.
Organisations with centralised data storage environments deliver more consistent performance across multiple locations and recover faster from disruption. In our study, only 21% of businesses have a centrally managed environment delivering consistent performance everywhere, with the vast majority (79%) leaving their teams with patchy access and fragmented platforms for sharing and collaboration.
A lack of centralisation also heightens security risks. Our survey found that, on average, organisations took four weeks to recover from a cyberattack. But almost two-thirds (62%) of firms that take less than one week to recover from their last attack use some form of centralised file environment, showing how such approaches can enhance cyber resilience.
While the vast majority (94%) of enterprises worry about risks from file sharing, fewer perceive the security issues surrounding AI. With around 78% of employees now estimated to be using personal AI tools for work, only 30% of our study participants say they’re worried about such risks from shadow AI (non-authorised AI use at work).
From capacity thinking to data value thinking
The organisations whose AI projects deliver transformative innovation and workflow efficiencies will be the ones that design strategic data layers that can address and accommodate previously silent obstacles – the surging data volumes, increased data access frequency, and higher data governance demands of the AI era.
Forward-looking enterprises that can make this shift from capacity thinking to data value thinking will ensure critical data remains usable, regardless of where it is created or accessed. But the laggards will continue to invest in sophisticated AI products without rethinking their data storage infrastructures as strategic data layers will continue to find their AI tools’ performance shackled by the same underlying limitations of legacy infrastructures.
