The global cloud infrastructure services market has crossed a symbolic threshold: $102.6 billion in spending in the third quarter of 2025, with a year-over-year growth of 25%. The data, compiled in a report by Omdia published in London on December 22, points to a pattern that no longer seems circumstantial: the sector has now experienced five consecutive quarters of growth above 20%. More importantly, it is growing for a very specific reason. Artificial Intelligence is transitioning from a testing laboratory into a steady, ongoing production workload within companies of all sizes.
This phase shift—from “pilot projects” to operational deployments— not only increases compute and storage consumption but also changes the nature of competition: the battle is no longer just about who has the best model, but about who offers the most robust platform to run multiple AI models and agents seamlessly, with operational continuity and cost control.
Three giants, two-thirds of the market
In the quarter, AWS, Microsoft Azure, and Google Cloud maintained their top positions and collectively accounted for 66% of global spending, according to Omdia. This concentrated leadership continues to grow: together, these three saw an increase of 29% year-over-year. This indicates that the market isn’t being “divided” at the expense of the leader but is instead expanding strongly among the leading providers.
AWS retained its first position with a 32% market share and a 20% growth rate— the best pace for the company since 2022, according to the analyst. Omdia attributes this acceleration partly to better availability of compute capacity and partly to the surge in demand related to AI initiatives. At the end of the quarter, AWS reported a backlog of $200 billion, a metric often seen as an early indicator of future consumption.
Azure, meanwhile, maintained second place with a 22% market share and a 40% YoY growth. Omdia relates this momentum to the push of enterprise AI and the strengthening of its ecosystem of tools for building, deploying, and managing AI solutions. The renewal of Microsoft’s agreement with OpenAI, mentioned in the report, is part of this competitive landscape: cloud is becoming the place where AI is “industrialized,” not just where it is trained.
Google Cloud closed the trio with an 11% market share and a 36% growth rate. Its momentum is tied to the adoption of AI services by large enterprises and increasing visibility of future revenue: its reported order backlog reached $157.7 billion, a significant jump from the previous quarter.
From model race to platform race
A key nuance in 2025 is crucial for understanding the market’s current state: many organizations no longer want to rely solely on a single model. A “multi-model” strategy is emerging, where commercial, open, and specialized task models coexist. This approach requires platforms capable of managing dependencies, permissions, costs, monitoring, and lifecycle management with a discipline akin to that of a mission-critical system rather than a trial experiment.
In this realm, hyperscalers are positioning themselves as the “operational layer” for AI: managed environments that integrate proprietary and third-party models, deploy agents, and maintain their operation in real-world scenarios. Omdia describes this shift as a transition from incremental model performance improvements to platform capabilities focused on reliability and continuous operation.
The practical consequence is that cloud is no longer only sold as elasticity; it is being marketed as production infrastructure for AI, with guarantees of availability, data governance tools, and a catalog of services ready to connect to business needs.
The secondary effect: rising appeal of private cloud
However, the growth in public cloud spending has a less visible reflection: as AI moves into production, more companies are reevaluating where to run their most sensitive workloads. This isn’t a wholesale retreat but rather a hybrid reordering: some workloads remain in public cloud for speed and managed services; others shift to private cloud for control, predictability, and— in some cases— compliance or data residency requirements.
In this context, VMware remains a key player in many organizations’ private cloud strategies… but it is also one of the catalysts for ongoing debate. Changes in licensing and the transition to per-core subscription models, along with controversies over minimum contract requirements, have rekindled interest in alternatives where total cost of ownership and provider dependency matter as much as performance.
Proxmox gains ground as an alternative and accelerates the conversation
Among the most recurring options on the radar is Proxmox VE, an open-source virtualization platform that in 2025 has shifted from being “the bold choice” to an option frequently compared directly with traditional stacks. This shift is no coincidence: as AI spending rises, so does the pressure to optimize every layer of infrastructure.
In Spain, this movement is influencing the provider market: some companies are strengthening private cloud offerings that provide continuity for current VMware users and structured migration paths for those looking to switch. Stackscale, which operates private cloud infrastructure in Spain, is positioning exactly this dual approach: support for VMware-based projects alongside specific services to migrate from VMware to Proxmox with evaluation, pilot implementation, and controlled deployment— plus options for centralized storage and high-availability architectures.
The core takeaway is clear: public cloud is growing fueled by AI demand, but private cloud is also benefiting from the same trend, through a different route. In a world where AI already operates in production, the conversation is shifting to platforms, costs, governance, and operational continuity. Both hyperscalers and private infrastructure ecosystems are adjusting their strategies to win the next phase of the market.
Frequently Asked Questions
Why is infrastructure cloud spending growing so much in 2025?
Because many companies have moved from testing AI to deploying it in production, which increases sustained consumption of compute, storage, and managed services.
What does it mean that competition is shifting “from the model to the platform”?
It means that offering a good model is not enough: companies now prioritize platforms capable of reliably running multiple AI models and agents, with governance, monitoring, and cost control.
Why are some companies considering migrating from VMware to Proxmox?
Due to a combination of licensing and cost changes, the drive to reduce dependency on a single vendor, and the need to maintain virtualization efficiency as workloads—including AI—grow.
When does it make sense to opt for private cloud for AI projects?
When cost predictability, compliance or data residency requirements, latency, and operational control are especially important, or when integrating AI with internal systems without exposing sensitive data outside the organization.