Cloud compute platforms

Microsoft Azure locks enterprise GPU workloads to its infrastructure, preventing qualified second sources from achieving economic viability.

Services that expose GPU and accelerator hardware through APIs and web consoles, eliminating direct infrastructure management. Enables AI developers to rent rather than own compute. Hyperscalers capture enterprise lock-in; neoclouds compete on speed and price.

Businesses that abstract physical GPU/accelerator infrastructure into on-demand, API-accessible compute services.

What the evidence shows

In Q1 2026, AWS held 28% market share, Microsoft Azure 21%, Google Cloud 14%.

Amazon, Microsoft, Google combined held 63% of cloud spending in Q3, rising from 62%.

srgresearch.com

New cloud suppliers must match the investment levels of largest suppliers to compete.

RESCORED JUL 2026oligopolyscaling24 companies

Who controls it

Microsoft Azure+23 more tracked

No independently verified market-size figure is published for this node yet.

Where it sits in the stack

depends onAI accelerators and silicon Facility ownership and operation Servers, systems, accelerator boards, and racks Data centre physical Hyperscale self-build owners/operators Neocloud facility owners/operators Edge and micro data centre operators Compute hardware and components

feeds intoCluster orchestration and scheduling Container orchestration (Kubernetes and AI extensions)HPC batch schedulers (Slurm and successors)AI-native cluster and fabric managers Virtualisation and multi-tenancy MLOps and training/serving toolchain Experiment tracking, pipeline orchestration and model registry Training infrastructure platforms Inference serving, engines and runtimes Inference serving engines and runtimes Data infrastructure for AI Data lakes, warehouses and lakehouses Vector databases and embedding stores Feature stores and data pipelines +9 more in the atlas

8 upstream · 23 downstream

Related nodes

Cluster orchestration and scheduling Virtualisation and multi-tenancy MLOps and training/serving toolchain Inference serving, engines and runtimes Model optimisation, compilers and kernels Data infrastructure for AI

Full scorecard, owner shares, supply edges and the full tracked roster are in the desk letter.

GET THE BRIEFING