IBM Cloud

Enterprise cloud with mainframe integration

3.2(0 reviews)freemium

free tierhourly billinggpu availableeu datacenterus datacenterapac datacenterapi access

Integrates with

IBM Cloud offers Virtual Server Instances, Kubernetes, bare metal, and strong mainframe (Z-series) integration. Used by financial services needing IBM Z connectivity. 60+ global datacenters.

What IBM Cloud Virtual Servers actually is

IBM Cloud is the enterprise outlier in our /category/vps-providers comparison. Where Hetzner, Vultr, and DigitalOcean sell one clean compute platform, IBM runs two: Classic Infrastructure (the SoftLayer estate IBM bought in 2013, still widely used for lift-and-shift) and the newer Virtual Private Cloud generation with its own instance profiles, block storage, and load balancers. The product we are reviewing here — Virtual Server Instances — exists on both, and that split is the single most important thing to understand before you sign up.

On the VPC side you get ten instance profile families: Balanced, Compute, Memory, Very High Memory, Ultra High Memory, GPU, Storage Optimized, Confidential Computing, Flex, and High Frequency. On the classic side you get public multi-tenant, dedicated single-tenant, transient (spot) servers billed at a much lower rate, reserved instances on one- or three-year terms, and dedicated hosts with up to 20 Gbps network throughput. Both run Intel Xeon hardware; the VPC generation also uses AMD EPYC Milan chips in its AMD profile lines (IBM notes compute capabilities are limited to EPYC Rome levels on those).

What genuinely separates IBM from the rest of this category is what sits alongside the VMs. Bare metal is a first-class product, not an afterthought. Power Virtual Server gives you IBM AIX and IBM i workloads in the cloud. And Hyper Protect Virtual Servers run on IBM LinuxONE — the IBM Z mainframe line — which is the only mainstream way to run a VM inside a FIPS 140-2 Level 4 hardware boundary. If you are a bank that runs core banking on a Z mainframe and you want burst capacity that speaks to it natively, this is realistically the only shortlist entry that fits. Everybody else in our comparison sells commodity x86 and stops there.

Pricing: what you really pay

IBM's pricing is legible once you are inside the provisioning calculator, and opaque everywhere else. In the VPC provisioning UI we were quoted $0.112/hr for a Flex bxf-2x8 profile (2 vCPU, 8 GiB RAM, 3 Gbps network), which works out to roughly $82 per month for the instance alone. Boot volumes are billed separately: $1.02/month for 10 GB, $5.09/month for 50 GB, and $10.18/month for 100 GB general-purpose volumes (max 3,000 IOPS at that tier). GPU profiles are priced where you would expect enterprise GPU rental to land: a 24-vCPU/120 GiB profile with one NVIDIA A100 at $4.633/hr, and the same shape with an L40S at $3.232/hr.

Put that next to our /benchmarks reference numbers and the gap is stark: Hetzner's CX22 (2 vCPU/4 GB/NVMe) is €4.35 per month, and even DigitalOcean's Premium AMD 4 GB is €28. A comparable 2-vCPU IBM instance is around $82 per month before storage. We are not saying IBM is overpriced for what it is — we are saying you should know that most of that premium buys you compliance certifications, contracting flexibility, mainframe adjacency, and a support organization that will sign enterprise paperwork. If none of those are on your list, the premium is dead money.

The entry point is friendlier than the list prices suggest. A Lite account is free with no credit card and no expiration, covering more than 40 services in limited tiers. Pay-as-you-go accounts have carried a USD 200 credit for new customers, and IBM rotates heavy promotional discounts — at the time of writing, 60% off VPC virtual server profiles with code BUYVPC, a USD 1,000 credit for Bare Metal Servers for VPC and surrounding VPC services, and GPU profiles on L40S, A100 PCIe, and Intel Gaudi 3 at 50% off. Object Storage runs a "One Rate" plan bundling storage, retrieval, API calls, and egress into one flat rate advertised from USD 10/TB per month. Classic virtual servers bill hourly or monthly and can be suspended; transient classic servers are deprovisioned first-on, first-off at steep discounts and are honest about being unsuitable for anything that needs constant uptime.

Our practical advice: treat the promo pricing as a reason to try, not a reason to commit. Renewal economics on enterprise cloud rarely match the first-year discount, and reserved classic capacity on one- to three-year terms is where the real per-unit savings live — with the corresponding lock-in.

Performance & infrastructure

The infrastructure story is stronger than the VM story. IBM's multizone regions are built from three or more physically separate data centers with independent power, cooling, and network, and IBM states sub-2-millisecond latency and more than 1,000 Gbps of bandwidth within a zone. Regional services spread across those zones generally deliver 99.99% availability, and the SLA for Virtual Server for VPC is published at 99.99% with tiered service credits, a 60-day claim window, and credits applied to your next invoice. Block Storage for VPC carries a 99.999% figure. That is a credit-backed SLA rather than marketing copy, which we count as one of IBM's real strengths.

The profile system is granular — Storage Optimized profiles attach temporary SSD instance storage at roughly 1 vCPU to 300 GB, Confidential Computing profiles use processor enclaves to keep data encrypted in use, and High Frequency profiles target latency-sensitive work. That granularity is genuinely useful for databases and regulated workloads. It also means your cost model has many more dials than it would at a developer-friendly host, and every dial is metered: boot volume, data volumes, egress, load balancers, and floating IPs all bill separately.

We have not run our own sysbench, fio, or iperf3 passes against IBM Cloud instances, so we will not quote performance numbers for them; our /benchmarks page covers the providers we have actually tested. What we can say editorially is that IBM's VPC generation is a competent modern platform — per-second metering, VPC networking, snapshots, image capture, autoscaling — that has closed most of the gap with AWS and Azure on basic virtualization, while the classic generation increasingly looks its age.

Where it falls short

The dual-platform reality is the biggest operational cost. Classic and VPC do not share APIs, images, or networking primitives; a migration between them is a project, not a checkbox. Documentation is extensive but sprawled across both worlds, and finding the current answer often means reading three pages to work out which platform they refer to.

DDoS protection is not bundled with your VM. IBM sells it as Cloud Internet Services, powered by Cloudflare's network of 165-plus points of presence, priced as its own product with a free trial. Security groups and ACLs cover you at the VPC level, but volumetric protection at the edge is an add-on — a different posture from hosts in our /category/firewall-ddos comparison that include baseline scrubbing in every plan.

Reliability has a footnote. IBM's SLA is credit-backed and honest on paper, but the platform has had high-profile global outages — the June 2020 incident took the console, DNS, and Kubernetes control planes down worldwide — and a 10% maximum credit does not reimburse the business you lost that day. Support is tiered: the base level is not what enterprise buyers assume, and the responsive support IBM is famous for is a paid tier or a contract line item.

Finally, the developer experience lags the competition. Provisioning flows assume more infrastructure knowledge than a DigitalOcean droplet or a Hetzner Cloud server, the catalog is enormous and unevenly maintained, and account-level cost management is a discipline of its own. Teams without a dedicated cloud engineer should budget time for this.

Who should pick this

We pick IBM Cloud when the workload is anchored to IBM's other platforms: core banking near IBM Z, AIX and IBM i estates on Power Virtual Server, SAP and Oracle workloads under managed services, or regulated financial services workloads that need the certification set and contracting vehicle only an enterprise vendor carries. Government buyers get FedRAMP and FISMA-oriented regions built to agency specs. Hybrid estates using Direct Link to stitch on-premises data centers to cloud regions are exactly the use case this platform was built around.

We would not pick it for a side project, a startup MVP, or a cost-sensitive SaaS backend. For those, the value leaders in our /category/vps-providers comparison deliver equivalent virtualization at a fraction of the monthly cost, and you can rent mainframe adjacency later if you ever actually need it.

Key takeaways

Two platforms in one account (SoftLayer-era Classic and the newer VPC) — confirm which you are buying before comparing prices.
Real list prices: 2 vCPU/8 GiB Flex at $0.112/hr (~$82/month) plus boot volume from $1.02/month; A100 GPU profiles at $4.633/hr — roughly 10x the value leaders in our benchmark set.
Credit-backed 99.99% SLA on Virtual Server for VPC, with multizone regions of three or more physically separate data centers and 99.99% regional availability targets.
Unmatched ecosystem reach: bare metal, Power Virtual Server, and Hyper Protect on IBM Z — the only mainstream mainframe-integrated cloud VMs.
Lite accounts are free with no credit card, and new-account credits plus rotating promo codes make trial cheap; long-term economics are enterprise-priced.
DDoS scrubbing is a separate Cloudflare-powered product (Cloud Internet Services), so budget for it if edge protection matters.

Sources

Pros

IBM Z (mainframe) and Power Virtual Server integration that no commodity host offers
Credit-backed 99.99% SLA on Virtual Server for VPC with multizone regions of three or more separate data centers
Ten instance profile families including Confidential Computing and storage-optimized shapes
Bare metal, spot-style transient servers, reserved capacity, and hourly billing in one contract surface
Free Lite account plus generous rotating new-account credits for trials

Cons

Two overlapping platforms (Classic and VPC) with different APIs, images, and pricing
Roughly 10x the monthly cost of value leaders for a comparable 2-vCPU VM
DDoS scrubbing is a separate paid product (Cloud Internet Services)
Metered extras (boot volume, egress, load balancers) make cost prediction work
Base support tier is not the enterprise support IBM's reputation implies

Best For

Enterprises anchored to IBM Z, Power, SAP, or Oracle estates, and regulated financial services teams needing certification coverage and mainframe adjacency