AI is forcing enterprises to rethink their cloud strategy

AI demand is pushing enterprises toward hybrid cloud

AI compute is changing cloud strategy. Enterprises need more GPU capacity to train and run AI models, but adding that capacity can increase costs and deepen dependence on a single provider. ISG identifies preparation for broader AI use as a central reason enterprises are changing their cloud strategies.

The response is a shift toward hybrid cloud. Companies are investing in GPU-enabled architectures, distributed data platforms, and hybrid AI operating models. A hybrid approach can combine public cloud services with private infrastructure and other environments. This gives enterprises more options for deciding where AI workloads and data should run.

The main constraint is no longer access to cloud services in general. It is access to the right computing capacity, at an acceptable cost, with enough control over data and infrastructure. AI workloads can require expensive GPU resources and move large amounts of data. These characteristics make workload placement and infrastructure economics more important.

For executives, hybrid cloud should therefore be treated as an operating decision rather than simply an infrastructure choice. The goal is not to spread workloads across more providers for its own sake. It is to place each workload where performance, cost, governance, and business requirements are best met while retaining the ability to change that placement as AI demand evolves.

This also reduces some exposure to vendor lock-in. But hybrid cloud does not remove that risk automatically. AI platforms can create dependencies at the model, data, application, and management layers even when infrastructure is distributed. Enterprises need architectures and commercial agreements that preserve practical options to move workloads.

Hybrid cloud needs one operating model for performance, cost, and resilience

Adding cloud environments creates a management problem. Enterprises may gain more infrastructure choices, but those choices become difficult to control when monitoring, automation, financial management, security recovery, and operations remain separate. ISG finds that organizations increasingly want these functions brought into a unified platform.

Cost and performance are especially important. AI workloads can consume expensive GPU capacity, making weak utilization or poor workload placement costly. Executives need visibility into what infrastructure is being used, which workloads are driving spending, and whether that spending produces the required performance. Financial management must therefore operate alongside technical monitoring rather than after infrastructure decisions have already been made.

Resilience is part of the same problem. An organization cannot manage business continuity effectively without understanding dependencies across private and public clouds, applications, data, and infrastructure. Cyber recovery also needs to be incorporated into this operating model so critical services and data can be restored after an attack or major failure.

The executive priority should be operational consistency. A hybrid architecture that requires independent processes and tools for every environment can add complexity faster than it adds flexibility. Unified observability and automation help teams identify problems across environments, while integrated financial controls make the resulting infrastructure costs visible to business and technology leaders.

This changes how enterprises should assess cloud providers and management platforms. Raw compute capacity is only one requirement. Providers must also demonstrate that they can support consistent operations, cost controls, recovery, and performance management across the environments an enterprise intends to use.

AI demand is creating a larger market for specialized cloud providers

AI workloads need large amounts of GPU capacity. That requirement is creating room for providers built specifically around AI infrastructure. Gartner expects neocloud providers, which specialize in AI-optimized computing, to capture 20% of the $267 billion AI cloud market by 2030. That would represent about $53.4 billion in annual market value if Gartner’s market estimate and share projection are realized.

This growth matters because enterprises now have more options than traditional hyperscale public clouds. Specialized providers can offer GPU-focused capacity and infrastructure designed around AI workloads. Established technology companies and investors are also expanding their offerings. Apple opened access to private cloud compute in June, while Blackstone and Google launched a compute-as-a-service offering in May.

For executives, more supplier choice creates both leverage and complexity. A specialized provider may offer attractive AI compute capacity or economics, but infrastructure selection should account for more than GPU availability. Data movement, integration with existing systems, security controls, geographic coverage, service reliability, and contractual terms can materially affect the total cost and operational value of a deployment.

Vendor concentration also deserves attention. Moving an AI workload to a new provider does not automatically eliminate lock-in. Dependence can shift to proprietary AI services, management tools, data formats, or specialized infrastructure. Enterprises should know which components can move between providers and which would require significant engineering work to replace.

The broader conclusion is that AI infrastructure sourcing is becoming more diverse. Enterprises can use that competition to match computing resources more closely to workload needs. The strongest strategy will preserve choice without creating unnecessary operational fragmentation.

GPU-heavy AI workloads make cost visibility a core infrastructure requirement

AI changes the economics of cloud computing. GPU resources can be costly, and enterprises are deploying them across increasingly varied environments. According to ISG, this growth is raising demand for cost transparency and optimization across private cloud, Kubernetes, edge, and sovereign cloud infrastructure.

The main constraint is utilization. Buying or reserving expensive compute does not create value if GPUs sit idle, workloads use more capacity than necessary, or teams cannot identify where spending originates. As AI deployments grow, executives need a clear view of resource consumption, workload performance, and costs across infrastructure types.

Traditional cloud cost controls may not be sufficient. A hybrid AI environment can spread spending across cloud services, owned infrastructure, container platforms such as Kubernetes, and specialized providers. These environments use different pricing and accounting models. Enterprises therefore need financial management that can translate infrastructure consumption into comparable business and workload costs.

Cost optimization must also be linked to performance. The cheapest infrastructure is not necessarily the correct choice if it slows a critical AI service or cannot meet availability requirements. Conversely, premium GPU capacity is difficult to justify for workloads that can meet their performance targets on less costly hardware. Companies need enough technical and financial visibility to make those trade-offs deliberately.

For C-suite leaders, this makes AI infrastructure economics a continuous management issue rather than a one-time procurement decision. Teams should be able to identify who is consuming GPU resources, what business workload those resources support, and whether capacity can be resized or moved without harming performance.

Data sovereignty and AI governance are driving cloud diversification

AI is making cloud location and control strategic issues. ISG identifies sovereignty as a major reason enterprises are diversifying their cloud environments. Companies want more control over infrastructure, data residency, and AI governance as they build long-term AI plans.

Data residency is a practical constraint. Enterprises operating across countries may need to control where sensitive or regulated data is stored and processed. AI adds complexity because applications can move data through model training, inference, retrieval, logging, and monitoring systems. Leaders therefore need visibility into both where data resides and how AI services use it.

Sovereign cloud environments can address some of these requirements by keeping data, infrastructure, or operational control within defined jurisdictions. Private and regional cloud options can provide additional control. But sovereignty is not achieved simply by selecting a particular deployment model. Enterprises still need to examine provider contracts, infrastructure dependencies, administrative access, encryption controls, data movement, and applicable legal requirements.

AI governance adds another requirement. Companies need policies that define which data models may access, who can deploy or modify AI systems, how outputs are monitored, and how sensitive information is protected. These controls become harder to enforce when workloads span multiple environments. A diversified cloud strategy therefore needs a consistent governance framework rather than separate policies for each provider.

For C-suite leaders, the objective should be selective control. Highly regulated or strategically sensitive workloads may justify sovereign or private infrastructure, while other applications can remain on public cloud platforms. The right architecture depends on the requirements of the workload, data, jurisdiction, and business.

This approach can also preserve long-term flexibility. AI technology, regulation, and provider offerings will continue to change. Enterprises that understand where their data and workloads operate, while maintaining consistent governance across those environments, will be better positioned to adapt without redesigning their entire AI infrastructure strategy.

Main highlights

• Rethink cloud strategy around AI: Rising GPU demand, costs, and vendor dependence are strengthening the case for hybrid cloud. Leaders should place AI workloads based on performance, cost, governance, and portability requirements.
• Unify hybrid cloud management: More infrastructure choices can create operational complexity. Prioritize platforms that integrate observability, automation, financial management, and cyber recovery across environments.
• Use growing AI cloud competition strategically: Gartner expects neocloud providers to capture 20% of the $267 billion AI cloud market by 2030. Evaluate specialized providers for GPU capacity and economics while protecting against new forms of vendor lock-in.
• Make GPU economics visible: AI infrastructure costs require continuous oversight across private cloud, Kubernetes, edge, and sovereign environments. Track GPU utilization and workload-level costs alongside performance to avoid paying for unused or unnecessary capacity.
• Build sovereignty into AI architecture: Data residency, infrastructure control, and AI governance are influencing where enterprises run workloads. Match deployment models to regulatory and business requirements while applying consistent governance across providers.