Cloud and infrastructure software

Why the concentration exists

Cloud infrastructure software transforms physical servers into metered and schedulable compute resources through virtualization and orchestration layers. The billing model shifted from capital expenditure to pay-as-you-go, which means poor capacity planning can lead to 30-50% wasted spend on idle resources. Because utilization and cost are now directly linked to software decisions, procurement lead times and workload placement have become strategic concerns rather than routine IT tasks. Capacity planning has consequently re-emerged as a rising operational concern for data center operators.[1][3]

The software stack that manages this resource abstraction concentrates value in orchestration interfaces and platform margins rather than raw hardware rental. Hyperscalers require the same cloud infrastructure to operate their own businesses, which reinforces their integrated software and hardware position. This self-reinforcing dynamic means the platforms that control the orchestration layer also dictate the terms of access to underlying compute. As a result, the software that schedules and monitors workloads becomes a more durable competitive advantage than the physical facilities themselves.[2]

Hybrid cloud adoption further entrenches the role of infrastructure software because organizations must orchestrate workloads across both on-premises and public environments. Over 90% of organizations will adopt hybrid cloud through 2027, according to Gartner. Managing this complexity requires sophisticated platform software that can move workloads and maintain policies across disparate environments. The software layer therefore becomes the essential integration point that binds physical infrastructure into a coherent operational system.[9]

What the evidence shows

Who supplies it

In specific national markets, government certification requirements further narrow the field of qualified suppliers. In Japan, only Amazon Web Services, Google Cloud, Microsoft Azure, Oracle, and Sakura Internet are certified as providers to the Japanese government under the Government Cloud program as of June 2025. This regulatory gatekeeping limits which infrastructure software platforms can serve public-sector workloads. It also creates a protected tier of approved suppliers that domestic agencies must choose from.[4]

What it depends on, and what depends on it

The software layer takes raw GPU and CPU capacity and parcels it into virtual machines and container environments tailored for AI workloads. Google Cloud offers fractional G4 VMs using NVIDIA vGPU technology with slice sizes of one-half, one-quarter, and one-eighth of a GPU. Azure HDv2 virtual machines feature nearly 500 physical AMD EPYC CPU cores and 4 terabytes of RAM for high-density compute. These virtualized offerings demonstrate how infrastructure software converts fixed hardware configurations into flexible, right-sized units of consumption.[11][12]

Downstream, the orchestration platforms feed AI model training and inference pipelines that depend on their scheduling and monitoring capabilities. AWS provides over 100 foundation models through Amazon Bedrock for software companies embedding AI into SaaS products. Microsoft and OpenAI published a paper on power stabilization for AI training datacenters that can reduce power overshoot by 40%, showing how infrastructure software extends into physical power management. The infrastructure software thus mediates every interaction between AI developers and the physical compute they require.[5][10]

Where it sits in the stack

What would break it

Software-level substitution also threatens the hyperscaler position by enabling workloads to move across environments more easily. VMware Cloud Foundation 9.1, announced on May 5, 2026, enables secure infrastructure for production AI workloads with up to 40% reduction in server costs through intelligent memory tiering. IBM is making management interfaces across its cloud, storage, power, and Z platforms MCP-compatible, turning infrastructure into building blocks that agents can reason about and orchestrate. These abstraction layers reduce the friction of moving workloads away from a single hyperscaler stack.[8][13]

Policy interventions represent another potential shift in the supply picture by mandating capacity or restricting provider choice. The European Commission proposed the Cloud and AI Development Act on July 2, 2026, with the aim of tripling EU data center capacity within five to seven years to better meet the needs of EU businesses and public administrations by 2035. The public consultation period on the Action Plan runs from July 2, 2026, to August 27, 2026. If enacted, this legislation could create new European infrastructure capacity that weakens the grip of non-EU hyperscalers on regional workloads.[7]

What to watch

Several major capacity expansions are due to come online in the near term, which will alter the physical supply available to infrastructure software platforms. Microsoft will open its East US 3 datacenter region in the Greater Atlanta Metro area in early 2027 and will add Availability Zones to the North Central US region by the end of 2026. Google Cloud plans to be among the first cloud providers to offer NVIDIA Vera Rubin NVL72 rack-scale systems in the second half of 2026. These additions will increase the addressable GPU and CPU capacity that orchestration software can schedule and meter.[6][11]

New hardware-software integration milestones are also approaching that will reshape how infrastructure software partitions and manages compute. Azure HXv2 virtual machines featuring 176 AMD 6th Gen EPYC CPU cores with clock frequencies exceeding 5 GHz are entering deployment. Microsoft announced the expansion of Azure AI and HPC infrastructure with AMD on July 20, 2026, bringing AMD's Helios AI platform and next-generation EPYC datacenter processors to Azure. These platforms will require updated orchestration software to exploit their specific performance characteristics.[12]

Related nodes

No other tracked node in the atlas currently shares this parent.

Sources

1. oreilly.com · 2026-03-03T00:00:00
2. oecd.org
3. cloudtoggle.com · 2026-04-30T08:51:54
4. trade.gov · 2025-11-18
5. aws.amazon.com
6. azure.microsoft.com · December 9, 2025
7. hunton.com · 2026-07-02
8. cio.com · late 2024
9. atos.net
10. azure.microsoft.com · October 13, 2025
11. cloud.google.com
12. blogs.microsoft.com · 2026-07-20
13. news.broadcom.com · May 5, 2026