Omdia: The Agentic Era Collapses Cloud Layers

retained text

Omdia: The Agentic Era Collapses Cloud Layers
Omdia research shows AI models are collapsing traditional cloud architectures by integrating multiple layers into single API calls and shifting billing from “pay-per-layer” to “pay-per-token.”
According to Omdia’s report, Global AI Cloud Stack 2026: Rethinking Cloud in the Agentic AI Era, the different layers of cloud-native architectures have long been moving toward joint scheduling in specific scenarios. At the same time, signs are emerging that traditional single, vertical layered architectures are beginning to break down.
Critical Impacts to the Traditional Cloud Stack in AI Native Era
Since 2023, the rise of large language models (LLMs) has further disrupted the cloud-native computing architecture paradigm in three aspects:
First, model technology has significantly accelerated the breakdown of vertical architecture. A single model API call can now cover functions across four to five layers of the cloud-native architecture, gradually integrating and restructuring these previously independent layers into functional components within the model itself.
Second, enterprise procurement logic is shifting from “pay-per-layer” to “pay-per-token.” With the rise of the token economy, tokens have become the core billing unit for enterprise procurement. Consequently, the original layered structure has lost a significant portion of its independent commercial value, increasingly becoming critical technical components supporting AI Cloud rather than independently deliverable product components.
Third, the production relationship between humans and machines has fundamentally changed. Models, agents, and AI as a whole are increasingly taking on the form of “labor.” Salesforce has launched Agentforce covering various job types; Microsoft’s Copilot has expanded across multiple office scenarios; and Anthropic has adopted a dual-track strategy, releasing both a digital coworker version (Claude Cowork) and a developer version (Claude Managed Agents). As humans gradually cede control to machines, the cybernetic logic on which cloud-native architectures rely is beginning to loosen.
For these reasons, cross-layer integration of functions removes the original layering of its independent technical significance; token-based billing reduces the independent commercial meaning of “layering”; and the emergence of machines as labor challenges the layering logic based on “the depth of human control over machines.”
Apart from the changes on the architecture of Cloud stack, Omdia also captured that the demands of using AI models can be divided into 2 different groups:
The “API-first” camp tends to prioritize speed-to-market and cost-efficiency, leveraging off-the-shelf model capabilities without scenario-grade customization. They often operate in scenarios where general intelligence suffices, such as customer service chatbots or basic content generation, betting that model upgrades will automatically address emerging needs.
In contrast, the "Agent-centric” group focuses on solving complex, domain-specific problems that require orchestration. For example, a supply chain Agent might use model APIs to analyze market trends, then trigger workflows to adjust inventory levels, coordinate with suppliers via email, and generate reports for stakeholders. Here, the Agent acts as a decision-maker and autonomous executor with model intelligence serving as just one part of its cognitive architecture.
As layer collapse and the theory of user control plane loosening continue to reshape the market, and the co-existence of 2 beliefs regarding the use of AI model, Omdia has redefined the technical architecture of AI Cloud in 2026 into three layers: AI Cloud Infra / MaaS / AaaS, corresponding to three distinct markets:
AI Cloud Infra includes multi-tenant AI IaaS and physically dedicated resource AI BMaaS.
Model-as-a-Service (MaaS) includes Model Production Service (MPS) and Model Consumption Service (MCS).
Agent-as-a-Service (AaaS) includes Agent Platform Service and Agent Labor Service.
The market spaces defined by Omdia for each track are mutually independent. The statistical scope for these tracks is based on recognized revenue, without specific cost breakdowns.
Global AI Cloud Infra Market
In 2025, the global AI Cloud Infra market size was valued at $65.74 billion. Within this, the global AI IaaS market was valued at $24.79 billion. Regionally, the top three markets were North America (45.7%), China (22.3%), and Europe (15.4%).
The global AI BMaaS market was valued at $40.95 billion. Regionally, the top three markets were North America (54.5%), China (13.7%), and Europe (9.2%).
Global Model Production Service (MaaS-MPS) Market
The Model Production Service market comprises four sub-markets: data and knowledge engineering, model development and training, model evaluation and alignment, and model governance and management.
The overall market size reached $18.54 billion, with the model development and training segment accounting for 63.5% of the market, making it the largest segment.
Regionally, the top three markets were North America (43%), Europe (16.8%), and China (14.8%).
Raymond Zhan, Senior Principal Analyst, Cloud & AI at Omdia, states that the global AI Cloud market serves both as a supporting market for industry-wide AI applications and as a demand driver for data centers. From the perspective of technological architecture evolution, ‘model-as-a-resource‘ is gradually becoming a reality. Meanwhile, as the resource layer shifts upward, more traditional infrastructure vendors gain the ability to reach the market logic of the application layer. However, validating the market logic of the application layer will still require successful ROI verification in industry-specific AI.
Raymond also notes that the Chinese market is also influenced by heterogeneous underlying architectural logic, different billing structures, and relative independence in the supply chain market. Therefore, it requires a relatively independent analytical framework for market tracking and insights.
The future AI Cloud technology stack should be built around the concept of “model-as-a-resource,” and market momentum will gradually shift from the resource layer to the application layer. Omdia aims to provide the world with quantitative data support that clearly demonstrates this market migration path, publishing relevant data tracking reports annually at the end of the year.
Omdia’s Global AI Cloud Stack 2026 report analyzes global AI cloud technology architecture trends, outlines corresponding market classification rules, defines mapping rules for major vendor products, and provides quantitative trend forecasts for key markets.

document fetched 2026-07-09 09:31 · canonical https://datacenterknowledge.com/cloud/omdia-the-agentic-era-collapses-cloud-layers