Global cloud infrastructure spending hit $102.6 billion in Q3 2025, up 25% year over year, as enterprise AI use moved from pilots to real production deployments worldwide.
Cloud spending jumps in Q3
Omdia said total spending on cloud infrastructure services reached $102.6 billion in Q3 2025, extending a run of more than 20% annual growth for five consecutive quarters. The research links the acceleration to a clear shift in enterprise demand—away from small experiments and toward scaled, production-grade AI systems that need steady capacity, security controls, and reliability. In Omdia’s definition, “cloud infrastructure services” includes hosted services such as IaaS and PaaS, plus related cloud-native services delivered over the internet.
Top providers are still pulling away. Omdia said AWS, Microsoft Azure, and Google Cloud together accounted for 66% of global cloud infrastructure spending in Q3 2025, and their combined revenues rose 29% year over year.
| Provider (Q3 2025) | Market share | YoY growth rate (Q3 2025) | Key point highlighted by Omdia |
| AWS | 32% | 20% | Omdia said growth improved as compute supply constraints eased and demand rose, with enterprise AI partnerships supporting momentum. |
| Microsoft Azure | 22% | 40% | Omdia pointed to expanding model access and enterprise agent tooling as adoption moved into production environments. |
| Google Cloud | 11% | 36% | Omdia said growth was driven largely by enterprise AI offerings, and it reported a Q3 backlog figure of $157.7 billion as of Sept. 30. |
| Top 3 combined | 66% | 29% | Omdia described demand as resilient, supported by rising backlogs at leading providers. |
Why AI production is driving spend
Omdia said the market’s center of gravity is moving from model “benchmarks” to platform capabilities that help companies run AI safely and reliably in day-to-day operations. That matters because production AI typically requires more than raw computing power—it needs governance, monitoring, cost controls, and the ability to switch between models depending on price, performance, and risk. Omdia framed “multi-model” support as a production requirement because enterprises want resilience and flexibility across generative AI workloads.
Separate survey evidence suggests many firms are still early in the journey, even as overall usage becomes mainstream. McKinsey’s 2025 global survey said 88% of respondents report regular use of AI in at least one business function, but nearly two-thirds say their organizations have not started scaling AI across the enterprise. The same survey said interest in AI agents is high—62% report at least experimenting with agents—yet only 23% report scaling an agentic AI system in at least one function.
| Enterprise AI adoption signal (McKinsey, 2025) | What it suggests for cloud demand |
| 88% report regular AI use in at least one business function. | Broader AI usage increases baseline cloud consumption for data, apps, and model access. |
| Nearly two-thirds say they have not begun scaling AI enterprise-wide. | A large “migration wave” is still ahead as pilots convert into production systems. |
| 62% are at least experimenting with AI agents; 23% are scaling agents in at least one function. | Agents can increase compute and monitoring needs because they run multi-step workflows continuously, not just one-off prompts. |
How hyperscalers are responding
Omdia said hyperscalers are competing by expanding “build-and-run” platforms that make production AI easier, including tools that support multiple model families (proprietary, third-party, and open-weight models). It pointed to managed AI platforms and model catalogs—Amazon Bedrock, Azure AI Foundry, and Vertex AI’s Model Garden—as examples of how cloud providers are packaging model choice with governance and operational tooling. Omdia also said providers are investing in standardized building blocks for “AI agents,” meaning software that can plan and execute tasks across multiple steps, because real-world deployment is proving harder than early experimentation suggested.
In its Q3 snapshot, Omdia described provider-specific moves aimed at making AI workloads easier to run at scale. For AWS, it highlighted updates to Bedrock’s model choice and “guardrails” features, along with new model and agent announcements showcased at re:Invent 2025 and an expansion of AWS regional capacity with a New Zealand region launched in September. For Microsoft, Omdia emphasized expanded model access in Azure AI Foundry, the launch of a Microsoft agent framework for multi-agent orchestration, and new regional infrastructure plans including expansion in Malaysia and a new India datacenter region planned for 2026.
For Google Cloud, Omdia said enterprise AI offerings were a key growth driver, and it highlighted continued expansion of Vertex AI’s model lineup plus the launch of “Gemini Enterprise,” positioned as an enterprise platform that combines the Gemini model family with agents, no-code tools, and governance features. Omdia also reported Google Cloud backlog growth from $108.2 billion in Q2 to $157.7 billion as of Sept. 30, framing it as stronger demand visibility.
What comes next
One near-term implication is that cloud growth is increasingly tied to whether enterprises can convert AI pilots into stable, governed production services that deliver measurable value. McKinsey’s survey found that only 39% of respondents attribute any level of enterprise-wide EBIT impact to AI so far, even though many report cost or revenue benefits at specific use-case levels. That gap reinforces why cloud providers are pushing beyond “model access” into operational platforms—monitoring, security, compliance support, and agent tooling—because these are often the barriers that slow production rollout.
Longer term, broader public-cloud budgets are still expected to rise sharply, which sets the backdrop for sustained infrastructure buildout. IDC forecasts worldwide spending on public cloud services will reach $805 billion in 2024 and continue expanding through 2028, with multiple regions projected to post five-year growth rates above 20%.
