Microsoft Azure

What Microsoft Azure actually is

Microsoft Azure is the enterprise cloud: hundreds of services, but at the core, virtual machines in Microsoft's datacenters, tied into the same identity, networking, and management stack that runs Windows Server, SQL Server, and Microsoft 365. The platform went commercially live on February 1, 2010 as Windows Azure; virtual machines and the wider IaaS story followed over the next few years, and today compute spans general-purpose, memory-optimized, storage-optimized, and GPU N-series families (NVIDIA-powered, GA since late 2016) plus Arm-based SKUs from the Cobalt and Ampere lines.

The honest headline from Microsoft's own marketing is that "more than 95% of Fortune 500 companies" use Azure. Treat that as a vendor claim — it counts any Azure touch inside a giant organization, and third-party breakdowns put real infrastructure commitment considerably lower — but the direction is right: if you work in a large company, Azure is probably already in the building. That entrenchment is the product's core advantage. Entra ID (formerly Azure AD) integration, hybrid connectivity to on-premise Windows estates, Group Policy-style governance, Azure Arc for managing on-prem and other-cloud machines from one plane, and procurement paperwork that enterprise legal teams already know.

The VM catalogue itself is conventional: burstable B-series (including the cheap Bps v2 and Bpls v2 Arm SKUs), D-series general purpose, E-series memory-heavy, L-series storage, F/H high-performance compute, and N-series GPUs, on Linux or Windows Server. Availability sets spread instances across fault and update domains within a datacenter; availability zones spread them across physically separate datacenters in a region. We track Azure under VPS providers, GPU cloud, and Kubernetes managed — AKS is a major draw on its own.

Pricing: what you really pay

Azure publishes full per-SKU rates, and the current cheapest Linux VM on the official pricing page is instructive: the B2pts v2 — 2 vCPU, 1 GiB RAM, Arm-based, no local temp disk — is $6.132/month pay-as-you-go. The next step up, B2pls v2 with 2 vCPU and 4 GiB, is $24.528/month. One- and three-year savings plans cut the B2pts v2 to $4.1099/month (~32%) and $2.7594/month (~54%), and spot capacity prices it at $5.5188/month (~9% off that SKU — spot discounts vary wildly by SKU and region). Billing is metered by the second.

Put those numbers next to the value VPS market — our Hetzner CX22 benchmark machine is 2 vCPU / 4 GB for €4.35/month — and the price-performance gap is blunt: the B2pls v2 costs roughly five times as much for comparable RAM. You are not buying Azure compute; you are buying Azure.

The free entry is decent: new accounts get a 30-day free trial, and the free tier includes 750 hours per month for 12 months of B1s, B2pts v2 (Arm), and B2ats v2 (AMD) burstable VMs, Linux or Windows. That is enough to run one small server continuously and learn the platform.

Then come the line items that decide the real bill. Managed disks are billed separately from the VM (the Bps v2 SKUs ship with 0 GiB temporary storage by design). Public IP addresses are metered. Egress to the internet is billed per GB on standard rates. Backup, monitoring, load balancing tiers, and Defender plans all meter separately. Reserved instances and Azure Hybrid Benefit (using existing Windows Server / SQL licenses) change the math substantially for enterprises; for a solo developer paying list price, the meter adds up faster than on any value provider.

Performance & infrastructure

The footprint is the largest claim in the industry: Microsoft's global-infrastructure pages say 70+ regions and over 400 datacenters (marketing pages now say 80+ regions and 500+ datacenters), spanning every continent, with EU, US, and APAC data residency and sovereign clouds for government and China workloads. Each region is a set of datacenters; availability zones are physically separate facilities with independent power and cooling inside a region.

Performance depends entirely on SKU and disk choice. The burstable B-series throttles to a baseline CPU credit balance — fine for dev boxes and low-traffic services, wrong for sustained compute. The Dv5/Fsv2 families are the normal production choices; the N-series covers GPU workloads from inference to training. Pairing VMs with Premium SSD managed disks is both the performance and the availability decision: a single VM with Premium SSD or Ultra Disk carries a 99.9% connectivity SLA, a single VM on Standard HDD drops to 95%, two or more VMs in an availability set earn 99.95%, and two or more instances across two or more availability zones earn 99.99%. Service credits step 10% / 25% / 100% below 99.99% / 99% / 95% respectively.

We have not benchmarked Azure VMs in our own runs — our measured numbers at /benchmarks cover Hetzner, DigitalOcean, UpCloud, Vultr, and Linode — so any Azure performance figure here would be guesswork. What we can say is that the networking layer (private VNets, global load balancing, 370,000+ miles of fiber per Microsoft) is serious infrastructure, and that per-instance specs are honestly listed in the pricing tables.

Where it falls short

Price-performance is the standing problem for non-enterprise buyers. List-price VMs cost multiples of value providers, and the discipline required to avoid waste — reservations, spot, Hybrid Benefit, right-sizing, auto-shutdown — is itself work. Cloud cost management is a category because of bills like these.

Complexity is the second. The portal is dense, and a "simple VM" pulls in a virtual network, network security group, public IP, NIC, disk, and resource group, each with its own settings and costs. Azure has improved the defaults, but newcomers still spend their first weeks learning vocabulary rather than shipping. Our standing advice: budget onboarding time, not just money.

The SLA is honest but conditional in ways that catch people out. A single VM with Standard HDD has a 95% SLA — over a day of allowed downtime a month — and the 99.99% number only exists if you architect across zones yourself. Support is where Azure diverges from value providers most sharply: the plan you pay for determines whether you talk to an engineer in an hour or open a forum thread. Basic support is included; production response is an extra monthly commitment.

Finally, the developer experience for small teams is weaker than AWS's or DigitalOcean's. Everything works, nothing is lightweight, and the happy path for a two-person startup is rarely Azure.

One underrated point in Azure's favor: the governance story. Azure Policy, role-based access control, resource locks, and cost budgets can be applied across an entire estate, and for organizations that answer to auditors, those controls are worth more than the per-hour rate. A cheap VM with no policy engine is expensive the first time a compliance review comes around.

Who should pick this

We pick Azure VMs when the organization runs Microsoft: Windows Server and SQL Server estates, Active Directory/Entra ID in the middle of everything, hybrid connections to on-premise hardware, government or compliance regimes that want the certification catalogue and sovereign regions, and enterprises where the negotiated agreement already discounts list prices. Teams with .NET workloads, Visual Studio subscriptions (which include monthly Azure credits), or an existing Microsoft Enterprise Agreement should run the numbers before looking anywhere else — with Hybrid Benefit and reservations, Azure can be competitive in ways list prices never show.

GPU buyers with enterprise requirements (N-series, compliance, private networking) and Kubernetes teams standardizing on AKS also belong here.

We would not pick it for a personal server, a small SaaS's first VMs, or any workload where raw compute per euro is the objective. For those, a value provider measured in our benchmarks wins on every axis that matters, and you can still put Azure in front of it later if the company grows into one.

Key takeaways

• Enterprise cloud (commercially live since February 2010) with VMs on Linux or Windows across burstable, general-purpose, GPU, and Arm families.
• Cheapest current Linux VM is the Arm-based B2pts v2 at $6.132/month (2 vCPU, 1 GiB); comparable RAM on the B2pls v2 runs $24.528/month — several times value-provider pricing.
• Free tier: 750 hours/month for 12 months of B1s, B2pts v2, and B2ats v2 VMs; new accounts also get a 30-day trial.
• SLA ladder is conditional: 99.9% single VM on Premium SSD/Ultra, 99.95% in an availability set, 99.99% across two availability zones, 95% single VM on Standard HDD.
• Dual-stack IPv6 is supported (public and private IPv6 on VMs and scale sets); platform-level DDoS infrastructure protection is free, enhanced DDoS tiers are paid.
• Buy it for Microsoft integration, hybrid identity, and compliance — not for raw price-performance.

Sources

• https://azure.microsoft.com/en-gb/pricing/details/virtual-machines/azure-linux
• https://azure.microsoft.com/en-us/pricing/details/virtual-machines/azure-linux-previous
• https://azure.microsoft.com/en-us/products/functions
• https://azure.microsoft.com/en-us/products/ddos-protection
• https://azure.microsoft.com/en-us/explore/global-infrastructure
• https://learn.microsoft.com/en-us/azure/reliability/regions-overview
• https://learn.microsoft.com/en-us/answers/questions/1114758/sla-for-virtual-machines
• https://learn.microsoft.com/en-us/answers/questions/2287902/i-have-taken-a-azure-free-subscription-and-unable
• https://learn.microsoft.com/en-us/azure/virtual-network/ip-services/create-vm-dual-stack-ipv6-portal
• https://azure.microsoft.com/en-us/blog/choose-azure-like-openai-did-to-power-your-compute-intensive-workloads-and-add-intelligent-interactions-to-all-your-apps/
• https://docs.microsoft.com/en-us/azure/virtual-machines/availability-set-overview
• https://en.wikipedia.org/wiki/Microsoft_Azure
• https://en.wikipedia.org/wiki/Microsoft
• https://getpanto.ai/blog/microsoft-azure-statistics