The AI dependency trap: When the infrastructure powering innovation becomes a business risk

By Michael Sell, SVP & Global Head, GARP

As artificial intelligence moves into enterprise operations, the supporting infrastructure and dependencies are becoming an important business risk factor. Concentration across cloud providers, foundation models, data centres and semiconductor supply chains can transmit disruption beyond technology teams, potentially affecting business continuity and operational resilience.

As adoption accelerates, and deepens, these dependencies become harder to identify and isolate. An organisation may know its AI vendors but not the cloud, computing and data services supporting them. A June 2026 global IBM study of 1,000 senior executives found that 91% of respondents did not fully understand the full scope of their organisations’ reliance on AI vendors, models and infrastructure.

The concentration risk beneath AI adoption
Organisations may deploy several AI applications and models, yet rely on a concentrated group of cloud providers, data centre operators and infrastructure partners. As a result, different services may share underlying technology platforms or computing capacity. Diversifying AI vendors does not necessarily diversify infrastructure exposure. An AI model disruption could interrupt customer-facing operations, while failures in data pipelines could compromise reporting and decision-making.

External developments can compound these exposures. Geoeconomic tensions, trade restrictions and supply-chain pressures can influence access to strategic technologies. For enterprises, these dynamics make concentration and substitutability important considerations in operational and supply-chain risk assessments.

Rethinking vendor risk and flexibility
Risk teams must assess which critical operations rely on each provider, the business impact of interruption and whether an alternative could be activated within an acceptable timeframe.

Alternative vendors do not necessarily guarantee an optimal technology transition. Differences in data formats, model behaviour, integrations, security approvals and proprietary tools can make migration difficult. An alternative technology solution works only if its independence, portability and performance have been tested.

Maintaining multiple providers can add costs, integration challenges and governance complexity. Enterprises should therefore weigh the cost of redundancy against the potential impact of disruption, prioritising safeguards for the most critical workloads. Contractual protections, data portability and exit arrangements matter, but a well-documented strategy is not the same as stress tested recovery capability.

Extending resilience beyond traditional recovery Vendor flexibility is only one part of preparedness. Traditional business continuity and risk mitigation plans may not fully capture the interconnected dependencies of AI-enabled operations, where models, data pipelines, cloud environments and third-party services may not recover simultaneously. Organisations must therefore question the continuity of critical business services when key technology components remain unavailable, including through alternative processing or controlled human intervention.

The assessment should extend to fourth-party relationships: including key infrastructure and service partners direct vendors rely on. These upstream dependencies can be hard to identify and assess. Scenario testing should examine cloud outages, restricted computing capacity and loss of a critical AI service. It should also establish who can suspend an AI-enabled process, what fallback remains workable, and how quickly a business service can recover.

Embedding AI infrastructure risk into enterprise governance
Managing these exposures requires shared accountability across technology, security, procurement, business and risk functions. CIOs, and technology teams understand architecture and substitutability; business owners understand the impact of interrupted services; CROs and CISOs can challenge concentration assumptions and connect them to risk appetite. Senior leaders need clear ownership and agreed escalation thresholds.

Material AI infrastructure dependencies should inform enterprise risk assessments and investment decisions alongside performance, scalability and cost. For boards, the test is whether an organisation can identify its critical dependencies and demonstrate how essential services would continue if one failed. An alternative matters only if it can be activated within a timeframe the business can tolerate. The goal is not to eliminate dependency, but to understand where it matters and to build resilience to withstand periods of stress.

As AI becomes more deeply embedded in enterprise operations, infrastructure resilience becomes a critical business decision rather than a purely technical safeguard. The measure of preparedness is not simply whether an AI application performs when everything works, but whether the organisation can withstand the vulnerability or failure of a key dependency.

AIAI DependencyBusiness Risk
Comments (0)
Add Comment