Omdia research finds AI models collapsing traditional cloud architecture layers, integrating multiple services into single API calls.

Omdia research reveals that AI models are fundamentally reshaping cloud architectures by consolidating multiple functional layers into single API calls and transforming billing models from "pay-per-layer" to "pay-per-token."

According to Omdia's *Global AI Cloud Stack 2026: Rethinking Cloud in the Agentic AI Era*, traditional single, vertically layered cloud architectures are breaking down. While cloud-native layers have long been converging in specific use cases, the rise of large language models since 2023 has accelerated this collapse across three distinct dimensions.

**Architectural Integration.** A single model API call now consolidates functionality spanning four to five traditional cloud-native layers, absorbing previously independent components into the model itself. This vertical integration fundamentally restructures how cloud services are decomposed and delivered.

**Billing Model Shift.** Enterprise procurement is migrating from per-layer pricing to per-token billing. As the token becomes the primary commercial unit, the original layered structure loses independent business value and increasingly functions as a technical component supporting AI Cloud infrastructure rather than a standalone product offering.

**Human-Machine Relations.** Models and agents are assuming labor-like roles. Salesforce's Agentforce, Microsoft's Copilot expansion, and Anthropic's dual-track Claude strategy (Cowork and Managed Agents) exemplify machines taking on autonomous work as humans relinquish operational control.

These forces eliminate the independent technical and commercial significance of layering and challenge the foundational logic of human control upon which cloud-native architectures were built.

Omdia identifies two distinct approaches to AI model deployment. The "API-first" camp prioritizes speed and cost-efficiency, using off-the-shelf model capabilities for general-purpose tasks like chatbots and content generation, expecting model upgrades to address future needs. The "Agent-centric" group tackles complex, domain-specific problems through orchestration—for example, a supply-chain agent using model APIs to analyze trends while autonomously triggering workflows to adjust inventory, coordinate with suppliers, and generate reports.

Against this backdrop, Omdia has redefined the 2026 AI Cloud technical architecture into three interdependent markets: **AI Cloud Infra** (multi-tenant IaaS and dedicated AI BMaaS); **Model-as-a-Service** (model production and consumption services); and **Agent-as-a-Service** (agent platforms and labor services).

In 2025, the global AI Cloud Infra market reached **$65.74 billion**. The AI IaaS segment totaled **$24.79 billion** (North America 45.7%, China 22.3%, Europe 15.4%), while AI BMaaS reached **$40.95 billion** (North America 54.5%, China 13.7%, Europe 9.2%). The Model Production Service market totaled **$18.54 billion**, with model development and training comprising 63.5% of the segment. Regional leaders were North America (43%), Europe (16.8%), and China (14.8%).

Raymond Zhan, Senior Principal Analyst for Cloud & AI at Omdia, observes that the global AI Cloud market simultaneously supports industry-wide AI applications and drives data-center demand. "From a technological perspective, 'model-as-a-resource' is becoming reality, and as the resource layer rises, traditional infrastructure vendors gain reach into application-layer market dynamics. However, validating application-layer ROI will require proven success in industry-specific AI deployment." Zhan also emphasizes that China's market operates under distinct architectural, billing, and supply-chain logic, requiring independent analytical frameworks.

The emerging AI Cloud technology stack should center on "model-as-a-resource," with market momentum gradually shifting from infrastructure to applications. Omdia intends to publish annual tracking reports documenting this migration path.