As AI infrastructure becomes critical to more businesses, providers need to assess the wider consequences of failure
The expansion of AI is changing what is at stake when computing infrastructure fails and businesses providing that computing power need to rethink risk management around the customers that depend on them, warns Simon Gilbert, CEO of Elmore Insurance Brokers.
Greater computing density and heat loads are increasing reliance on power, cooling and connectivity. Yet assessing these systems individually can leave businesses with an incomplete picture of how a failure could spread, interrupt customers’ operations and expose leadership decisions to scrutiny.
This applies particularly to specialist AI cloud businesses known as NeoClouds, GPU-as-a-service providers renting out computing power, GPU cluster operators, AI infrastructure firms and cloud integrators. These businesses may not own the data centre building, but can still have substantial exposure through their equipment, customer contracts and financing arrangements.
Simon Gilbert comments:
“AI is turning data centres into highly concentrated critical infrastructure. For the businesses using these facilities, the risk profile can no longer be understood solely through the value of the buildings and equipment they contain. It also depends on the economic activity they support and the consequences when that infrastructure becomes unavailable.
“Providers need to ask how far the impact of a failure could extend, which customers depend on the same systems and whether their insurance reflects those connections. Treating property, cyber and business interruption as separate exposures can obscure the bigger picture.”
Insurance reviews should distinguish between the facility operator and the compute business using it. Large numbers of valuable graphics processing units, or GPUs, can be concentrated in one location, making specialist hardware cover important where appropriate. Businesses also need to consider promises to keep services available, potential liability to customers and whether their own cover responds when a third-party facility fails.
This also raises a governance question as AI infrastructure becomes more complex and more important to customers. A major breach or operational failure could bring the adequacy of procedures and protections under scrutiny, potentially triggering claims against directors and senior management.
The drive for more sustainable infrastructure adds another dimension. Renewable energy, battery storage and newer cooling technologies need to be assessed as part of the wider operating system, with attention to how they interact and what happens when one component fails.
Simon Gilbert continues:
“The challenge is to understand how resilience and sustainability work together as AI infrastructure evolves. Insurers need to keep pace with the technology being deployed, while providers need to demonstrate how the systems supporting their growth are managed and protected.
“New technologies may have limited claims history, but that does not make the consequences of failure any less significant. Underwriting needs to reflect how a compute business actually operates, including its reliance on third-party power, storage, cooling and connectivity.”
This thinking should extend beyond preventing an incident to maintaining insurability after one. A significant claim changes how insurers view a business, making a clear account of the failure, remedial action and strengthened protections essential to securing future cover.
Financing adds another dimension as newer technology can affect the future value of GPU equipment. At Elmore, we are exploring the potential role of residual value insurance in GPU financing, an emerging area where any cover would depend on the equipment, financing arrangements and available insurance capacity.
“As AI computing services become more critical to their customers, risk management needs to keep pace with providers’ commercial ambitions. Understanding the wider consequences of disruption is fundamental to building infrastructure that businesses can depend on.”