A new report by data centre solutions provider Flexential reveals that 95% of IT leaders believe their organisations’ growing investments in artificial intelligence (AI) have heightened their vulnerability to cyber threats.

The 2024 State of AI Infrastructure Report is based on a survey of 350 IT leaders from companies with annual revenues exceeding $100m. The survey includes 100 respondents from firms with revenues over $2bn.

“As AI becomes more embedded into business processes, it’s clear that enterprises need help safeguarding their organisation’s critical data as well as educating and up-skilling their entire staff,” said Flexential’s chief innovation officer, Jason Carolan. “Our report findings reinforce the importance of resilient and secure IT infrastructure solutions that can reduce vulnerability while supporting the high-performance demands of AI workloads.”

CIOs and other senior leaders increasingly nervous about securing AI

While AI is rapidly becoming central to business processes, the report highlights significant gaps in organisations’ ability to manage these deployments securely.

The findings show that 40% of respondents acknowledged their cybersecurity teams lack the necessary understanding to protect AI applications and workloads effectively. This represents a major concern as organisations expand their AI capabilities, facing an increased attack surface due to the inherent complexity of AI applications, as noted by 54% of IT leaders surveyed.

Moreover, 39% of respondents reported that AI investments have resulted in the storage and processing of more sensitive data than before, which has further escalated the risks.

The report identifies a marked shift in strategy regarding where AI workloads are deployed. In response to growing privacy and security concerns, 42% of organisations have moved their AI applications and workloads from public cloud environments to colocation data centres. This approach allows companies to exert tighter control over sensitive data by leveraging on-premise or third-party data centres and private clouds, thereby mitigating potential risks associated with public cloud infrastructures.

The strategic shift towards colocation is part of a broader trend highlighted in the report, where 59% of respondents emphasised increasing IT infrastructure investments as a crucial part of their AI roadmaps.

Flexential’s report also emphasises the challenges associated with AI infrastructure performance. Over the past 12 months, 82% of surveyed organisations reported encountering performance issues with their AI workloads. Among these, 43% cited bandwidth shortages, and 41% reported unreliable connections as significant barriers to efficient AI operations. These performance challenges underscore the need for robust infrastructure capable of handling high-density computing requirements and low-latency AI applications, which most existing data centres are not originally designed to support.

Furthermore, the report sheds light on the growing pressure on IT leaders to align AI infrastructure with evolving business and regulatory requirements.

Emergent skills gap

A significant 51% of respondents indicated that to optimise access and performance, they are moving sensitive data storage closer to the network’s edge, despite the increased cybersecurity vulnerabilities this can create. This strategy aligns with the demand for more agile and secure data handling, reflecting the complex balancing act between performance, cost efficiency, and security.

The survey results also pointed to significant skills gaps that could impede AI deployment success. 53% of respondents cited staffing shortages related to the management of specialised computing infrastructure, while 47% pointed to deficiencies in managing advanced networking technologies such as SDN and NFV. These gaps suggest a growing reliance on third-party colocation data centres to process data closer to the edge of the network, a strategy chosen by 51% of organisations to reduce latency and enhance performance.

Read more: Just how high can Nvidia fly?