In the Age of AI, “Deliberate Choice” Is Becoming More Important Than “Distribution” in Multi-Cloud Environments
With the proliferation of AI, corporate multi-cloud strategies are also evolving.
In the past, multi-cloud was often used as a defensive strategy to reduce dependence on a specific cloud provider or to ensure system redundancy.
However, since AI workloads require vastly different infrastructure depending on the model, GPU, data location, and regulatory requirements, it is now becoming increasingly important to move beyond simply using multiple clouds and adopt a strategy of intentionally selecting the most suitable environment based on business and technical requirements.
Gartner Senior Analyst Dennis Smith described this as “Intentional Multicloud.”
Why It Is Difficult for a Single Cloud to Handle All AI Workloads
There are three main reasons for the expansion of multicloud in the AI era.
First is the variation in AI capabilities among cloud providers.
As the strengths of the large language models (LLMs), AI services, GPUs, and training infrastructure provided by each cloud service provider (CSP) vary, situations are becoming more common where key data resides in one cloud while the necessary models or training resources are located in another environment.
Consequently, rather than moving the entire dataset, it is becoming increasingly important to connect and leverage the specialized capabilities of different clouds.
Second is data sovereignty and data residency.
As regulations regarding the location of data storage and processing tighten by country and industry, it is becoming increasingly difficult for global companies to consolidate all their data into a single cloud environment.
Third is the supply constraint on GPUs and specialized hardware.
As demand for AI grows rapidly, it has become difficult to guarantee that the necessary GPU resources will always be available on a specific cloud, making it increasingly important to have a structure that allows resources from other clouds or data centers to be utilized as needed.
Workload Placement Is More Important Than “Multicloud”
The core of a strategic multi-cloud approach lies not in simply using multiple clouds, but in having clear criteria for determining which environment to deploy each AI workload in.
For example,
Sensitive data processing
→ An environment that meets data sovereignty and security requirements
Large-scale model training
→ A cloud with sufficient GPU supply or a GPU cluster
General service operations
→ An environment with high cost-effectiveness and operational efficiency
Utilizing
specific AI models → A CSP with strengths in that particular model or AI service
As shown above, deployment strategies can vary depending on the nature of the work.
In this structure, the ability to coordinate distributed infrastructure and AI resources within a single operational framework becomes more important than simply purchasing cloud services.
The core of AI multicloud is “Day 2 operations”
Complexity in a multi-cloud environment increases significantly during the operations phase rather than during deployment.
This is because when AI workloads are distributed across multiple environments, you must manage the cost policies, data transfer costs, security policies, and the status of models and GPU resources for each cloud separately.
The article particularly emphasizes the importance of “Day 2 Governance”—that is, the ongoing management that begins once the actual service is up and running.
The following items were identified as key operational challenges:
Cost Management
: Holistically track usage and data egress costs across clouds and optimize them based on FinOps
Security and Access Management
: Applying consistent identity and access policies and data protection frameworks across multiple clouds
AI Asset Management
: Lifecycle management of models, GPU resources, AI agents, and other assets left unused after experimentation
Integrated Observability:
Consolidating cloud-specific telemetry, performance metrics, distributed tracing, and alerts into a single system
In particular, “Shadow AI”—where individual business units directly procure the AI resources they need, resulting in operations outside the IT organization’s scope of management—was also identified as a new management issue.
Integrated Technologies Enabling Multi-Cloud
In the past, operational technologies were not sufficiently mature relative to the concept of multi-cloud, making integration costs and complexity major challenges.
Recently, however, advancements in related technologies—such as cross-cloud networking, integrated control planes, and resource pooling across multiple data centers—have expanded the foundation for managing disparate infrastructures within a single operational framework.
Consequently, the core of multicloud has shifted from simple infrastructure distribution to
toward integrating workload placement → resource management → policy enforcement → cost optimization → observability
.
QUANTUM View
A key point to note in this article is that multi-cloud in the AI era is shifting from a “strategy of using multiple clouds” to a “strategy of deploying workloads to the most suitable environment and managing them in an integrated manner.”
Unlike general-purpose applications, AI infrastructure involves numerous factors to consider, such as GPU availability, model characteristics, data location, and security policies.
Therefore, rather than locking all resources into a single environment, future AI infrastructure will
On-Premise / Private Cloud / Public Cloud / GPU Cluster
as needed and managing them under a single operational framework.
In particular, what is crucial in this process—more so than multicloud itself—is multi-cluster management, resource scheduling, unified policies, cost and performance visibility, and observability.
As enterprise AI expands beyond proof-of-concept (PoC) to the actual service operation phase, platform capabilities that enable the consistent operation and control of distributed AI workloads and resources—in addition to technologies that connect diverse infrastructures—are expected to become increasingly important.
View Original Article
IT DAILY
“[Op-Ed] ‘Intentional Multi-Cloud’ in the AI Era”
Dennis Smith, Gartner Senior Analyst · Yang Seung-gap, Reporter
August 31, 2026.