DataCentersX > Workloads > Cloud and Enterprise

DC Cloud & Enterprise Workloads

Cloud and Enterprise is the workload cluster that accounts for the largest share of global datacenter compute capacity by volume. It covers the general-purpose business and consumer applications that run the economy: SaaS platforms, cloud-native services, traditional enterprise applications such as ERP and CRM, and the data warehouses and analytics platforms that sit downstream of them. The three children below group together because they share the defining infrastructure profile of enterprise-grade compute: CPU-dominated workloads with selective accelerator use, elastic microservices architecture, steady interactive load with diurnal and seasonal peaks, and deployment primarily on hyperscale cloud and colocation infrastructure.

What distinguishes this cluster from AI Training is the absence of accelerator-density demand: a SaaS platform does not need 120 kW racks. What distinguishes it from HPC is the absence of tightly-coupled scientific compute patterns: business applications rarely run MPI jobs across thousands of nodes. What distinguishes it from Regulated Industries is that regulation does not drive facility design, even though many individual cloud and enterprise workloads carry regulatory obligations at the application layer. What distinguishes it from Communications and Content is that the deployment model centralizes rather than geographically distributes: cloud and enterprise workloads consolidate in a smaller number of large facilities, not in hundreds of edge points of presence.

The three children

Child Workload	Characteristic Profile	Dominant Deployment
Cloud and SaaS	Elastic multi-tenant microservices; API-driven; continuous deployment	Hyperscalers (AWS, Azure, GCP), cloud-native SaaS platforms
Enterprise Apps	Transactional business systems; integration-heavy; long lifecycle	Enterprise on-premise, colocation, hybrid cloud, cloud migration in progress
Big Data and Analytics	Large-scale data ingest and query; batch and streaming; OLAP-optimized	Hyperscale cloud data platforms, enterprise lakes and warehouses

Shared infrastructure demands

The three children share a set of infrastructure requirements that distinguish them from other workload clusters.

CPU-dominated compute. Cloud and enterprise workloads run primarily on general-purpose server CPUs with high core counts, large memory footprints, and optional accelerator attachment for ML-adjacent workloads. The accelerator share of the cluster is growing but remains a minority of deployed capacity. Server SKUs favor memory-optimized and balanced configurations rather than the GPU-dense reference designs characteristic of AI factories.

High-fan-out Ethernet networking. Unlike AI training and HPC which demand low-latency lossless fabrics such as InfiniBand and RoCE, cloud and enterprise workloads run on standard Ethernet with leaf-spine or Clos topologies. East-west traffic within a cluster matters, but the microsecond-scale latency requirements of training fabrics do not apply. Network design optimizes for aggregate capacity and fault tolerance rather than minimum latency.

Mixed storage tiers. The storage profile spans object stores for unstructured data and backups, block storage for virtual machines and databases, and increasingly NVMe for transaction-intensive workloads. Parallel file systems appropriate for HPC are rare; the dominant storage layer is hyperscale object storage (S3, Azure Blob, GCS) for analytics and cloud-native applications, with enterprise storage arrays (Pure, NetApp, Dell) persisting in on-premise deployments.

Elastic capacity with diurnal peaks. Cloud workloads scale up and down continuously in response to demand; enterprise workloads have more predictable daily cycles tied to business hours and batch processing windows; analytics workloads have spiky peaks around reporting cycles. Facility capacity planning assumes averaged utilization well below peak, with autoscaling and scheduling policies smoothing aggregate load.

Air cooling remains sufficient. Rack densities for cloud and enterprise workloads typically sit in the 8 to 20 kW per rack range, well below the air cooling economic ceiling. Liquid cooling is uncommon in this cluster, unlike in AI training or HPC where it is effectively required. Facilities hosting cloud and enterprise workloads can be optimized for power and space efficiency rather than for thermal extremes.

Where cloud and enterprise workloads run

The three children deploy across several facility types, with the mix driven by organizational preference, regulatory posture, and the specific sub-workload.

Deployment Context	Typical Workloads	Rationale
Hyperscaler DCs	Cloud and SaaS, cloud-native applications, cloud data platforms	Scale economics, continuous deployment infrastructure, managed services
Colocation DCs	Enterprise apps, hybrid cloud extensions, data warehouse tiers	Carrier density, cross-connect to cloud providers, controlled environment without owning real estate
Enterprise DCs	Legacy ERP and CRM, on-premise applications, data residency-constrained workloads	Full operational control, data sovereignty, depreciation of existing assets
Hybrid cloud (cloud + colocation + enterprise)	Enterprise apps with cloud bursting, analytics pipelines spanning on-premise and cloud data	Workload placement optimized per application; migration in progress

The cloud migration arc

Cloud and Enterprise is the workload cluster most affected by the ongoing decade-long migration from enterprise on-premise infrastructure to hyperscale cloud. SaaS reached cloud-native maturity first, with platforms born in the cloud never having run on-premise. Enterprise applications have migrated in stages, with new deployments typically cloud-first and legacy systems modernized or replaced over time. Big Data and Analytics followed a similar arc, with cloud data warehouses and lakes displacing on-premise Hadoop and Teradata installations at many organizations.

The migration is not complete. Regulated industries retain significant on-premise footprint for data residency and control reasons; some enterprises have executed partial repatriation of workloads back from cloud to colocation or on-premise for cost reasons; and the AI-driven demand for GPU compute has produced a new generation of enterprise and sovereign private cloud deployments. The cluster's deployment profile continues to evolve, with the aggregate trend still pointing toward cloud consolidation but with meaningful exceptions at both the regulated and cost-optimized ends.

Where this cluster sits in the workload taxonomy

Cloud and Enterprise is the default workload cluster: the general-purpose compute that runs the business and consumer software layer of the economy. AI Training and HPC and Simulation are specialized clusters defined by accelerator and tight-coupling demands. Communications and Content is a distributed cluster defined by geographic proximity requirements. Regulated Industries is the cluster where regulation shapes facility design. Cloud and Enterprise is what remains when those specializations are factored out, and it accounts for the majority of deployed datacenter capacity when measured by server count or aggregate power draw.

Related coverage