Pick your cloud provider from the software you already run, not from a benchmark chart. If your team lives in Microsoft 365 and Windows Server, Azure will cost you the least effort. If you are starting fresh and want the widest service catalogue and the deepest pool of engineers who have seen your problem before, AWS is still the default. If your work is data and machine learning, Google Cloud earns a look. Oracle, Alibaba and Huawei Cloud are specialists: brilliant in a narrow set of situations, awkward outside them.
Where the market actually sits in 2026
Synergy Research Group reported that enterprise spending on cloud infrastructure services reached $143 billion in the second quarter of 2026, a jump of more than $43 billion on the same quarter a year earlier. That is 43% growth, which Synergy called the highest rate in eight years. The market has roughly doubled in size over eleven quarters, and AI training and inference are doing most of the pushing.
Share of that spend matters to you for one practical reason: it decides how many people know the platform, how many third-party tools integrate with it, and how quickly a problem you hit at 2am has already been answered in public. CRN’s reporting of the same Synergy data for Q2 2026 puts AWS at 28%, Microsoft at 20% and Google Cloud at 15%, with the three together taking 67% of global spend.
The tail is not static. Synergy’s Q2 2026 note lists CoreWeave, OpenAI, Oracle, Crusoe, Nebius, Anthropic and Nscale among the fastest-growing tier-two providers, and says nine so-called neocloud companies now sit inside the top 40 cloud providers by infrastructure revenue. If you need dense GPU capacity rather than a general-purpose platform, those names are worth a quote. For ordinary business workloads, they are not a substitute.
What each provider is genuinely good at
Here is the short version of a long argument. Every one of these providers can run a web application, a database and a backup target. The differences that survive contact with a real project are the identity story, the data services, the regional footprint and the licence maths.
| Provider | Strongest fit | Published footprint | Watch out for |
|---|---|---|---|
| AWS | Broadest service catalogue, biggest hiring pool, most third-party integrations | 124 Availability Zones across 39 regions | Service sprawl: it is easy to end up with eleven ways to do one thing |
| Microsoft Azure | Organisations already on Microsoft 365, Entra ID and Windows Server | Microsoft documents the widest region count of any provider | Licence benefits are the sell; model them before you commit |
| Google Cloud | Data analytics, BigQuery, Kubernetes-first teams | 43 regions and 130 zones | Smaller partner ecosystem outside data and AI work |
| Oracle Cloud (OCI) | Oracle Database estates and licence-bound workloads | Expanding, with multicloud links into Azure and Google | Thin outside the Oracle stack; check service parity per region |
| Alibaba Cloud | Mainland China presence and APAC commerce workloads | Strong China and South-East Asia coverage | Separate accounts, separate compliance regime, separate tooling |
| Huawei Cloud | Markets where Huawei already holds the network contract | Concentrated in China, Asia, Africa and the Middle East | Export-control exposure and a smaller international partner network |
AWS publishes 124 Availability Zones inside 39 geographic regions, with more announced. Google Cloud publishes 43 regions and 130 zones. Both numbers are useful for one question only: is there a region close enough to your users, and does it have at least three zones so you can survive a single data centre failing? If the answer is yes for two providers, footprint stops being the tiebreaker.
Read the SLA, not the slogan
Uptime marketing and uptime contracts are different documents. Amazon’s Compute Service Level Agreement commits to a 99.99% Monthly Uptime Percentage at region level only when your running instances are deployed across two or more Availability Zones. A single EC2 instance carries a 99.5% instance-level commitment. Google’s Compute Engine SLA is shaped the same way: instances in multiple zones get 99.99%, a single instance of most families gets 99.9%, and memory-optimised single instances get 99.95%.
| Commitment | AWS EC2 | Google Compute Engine |
|---|---|---|
| Spread across zones | 99.99% monthly, region level | 99.99% monthly, multiple zones |
| Single instance | 99.5% monthly | 99.9% (most families), 99.95% memory-optimised |
| First credit tier | 10% credit below 99.99% | Credits per the service-specific SLA |
| Largest credit | 100% credit below 95.0% | Credits per the service-specific SLA |
| What you must do | Deploy concurrently in 2+ AZs | Deploy in 2+ zones |
- A service credit pays back a slice of your bill, not your lost revenue.
- Single-instance commitments are the ones most small deployments actually get.
- Architecture, not provider choice, is what moves you from 99.5% to 99.99%.
Price anchors you can verify yourself
Prices move, so quote them with a date and a link. Everything below was read from the vendors’ own pricing material in October 2026, in their cheapest common US regions, and will drift.
| What you are buying | Published price | Where it comes from |
|---|---|---|
| General-purpose VM, 4 vCPU (Azure D4s v5, Linux, East US) | $0.192 per hour | Azure retail price list |
| The same VM on Azure Spot capacity | $0.040531 per hour | Azure retail price list |
| Object storage, hot tier (Amazon S3 Standard, first 50 TB, N. Virginia) | $0.023 per GB-month | AWS price list |
| Object storage, hot tier (Azure Blob Hot LRS, East US) | $0.0208 per GB-month | Azure retail price list |
| Internet egress after the free allowance (Azure, from North America or Europe) | $0.087 per GB | Azure bandwidth pricing |
| Internet egress (AWS worked example, Ireland to internet) | $0.09 per GB | AWS S3 pricing page |
| Three-year commitment discount (AWS Savings Plans and Reserved Instances) | Up to 72% off on-demand | AWS documentation |
Two things fall out of that table. First, spot or pre-emptible capacity is the single biggest lever in cloud pricing: the same Azure D4s v5 drops from $0.192 to about four cents an hour if your workload can tolerate eviction, which is a reduction of roughly 79%. Second, egress is the line nobody models. Moving 5 TB out to the internet in a month at Azure’s $0.087 per GB rate is around $440 before you have served a single extra customer. AWS now gives 100 GB of internet egress free each month, and Azure’s published table makes the first 100 GB free from North America and Europe, which covers a brochure site and nothing more.
Nobody has ever been fired for choosing the second-cheapest provider. Plenty of teams have been embarrassed by an egress bill they never modelled.
When Oracle, Alibaba or Huawei is the right answer
Oracle Cloud Infrastructure makes sense when Oracle Database licensing already dominates your costs, or when you want Oracle’s database services sitting a few milliseconds from workloads in another provider through its multicloud interconnects. Oracle also appears in Synergy’s Q2 2026 list of the fastest-growing tier-two providers, driven largely by AI capacity deals rather than general business migration.
Alibaba Cloud is the pragmatic answer for mainland China. If you sell into China, you need in-country hosting, an ICP filing and a provider who operates there daily. Huawei Cloud wins where it already holds the network or carrier relationship, which in practice means parts of Asia, Africa and the Middle East. CRN’s Q1 2026 breakdown of Synergy’s data lists the top five global providers as AWS, Alibaba, Google Cloud, Microsoft and Oracle, so these are real businesses, not curiosities.
The honest caution is operational rather than technical. A second provider means a second billing relationship, a second identity model, a second set of audit evidence and a second on-call runbook. Unless there is a reason you can say out loud in one sentence, do not add one. Our view on that trade-off is set out in single cloud versus multi-cloud.
How to run the decision in a week
You do not need a three-month evaluation. You need five decisions written down, and the discipline to stop once they agree.
- List your constraints first. Data residency, a regulator, an existing licence agreement, or a customer contract that names a region. Constraints usually eliminate half the field before any technical work starts.
- Map your top five workloads to managed services. Not to VMs. If four of the five land on services you would actually use, that provider is a fit.
- Model one month of real traffic, including egress. Use the published rates above and your own analytics numbers, not a vendor’s sample workload.
- Check the SLA shape against your architecture. If you will run one instance, you are buying a 99.5% or 99.9% commitment regardless of the logo.
- Write the exit paragraph. How would you leave in 90 days, what would it cost, and what is proprietary? If you cannot answer, you are not choosing, you are committing.
That is the process we run with clients remotely, and it usually lands on one primary provider plus a cheap second home for backups. If you want help running it on your estate, our cloud solutions work covers the assessment, the migration and the month-two bill review. Before you compare vendors, it also helps to be clear on what you are actually buying: IaaS, PaaS and SaaS are priced and supported very differently, and if the cloud idea itself is new to your team, start with our introduction to cloud technology.
Frequently asked questions
Is AWS still the safe default in 2026?
For general business workloads, yes, mostly for boring reasons. It holds 28% of global cloud infrastructure spend according to Synergy’s Q2 2026 data, which means the widest pool of engineers, documentation and third-party tools. Azure overtakes it the moment your estate is already Microsoft-centred, because the identity and licensing alignment saves more than any hourly rate difference.
How much cheaper is one provider than another, really?
On like-for-like general-purpose compute the hyperscalers sit within a few percent of each other. The differences that matter are commitment discounts of up to 72% on AWS Savings Plans, spot capacity at roughly a fifth of on-demand, and egress rates of around $0.087 to $0.09 per GB. Architecture and purchasing discipline beat vendor choice.
Should we use two providers for resilience?
Only if you can state the failure you are protecting against and you are willing to pay for two sets of skills. A second provider as a backup destination is cheap insurance. A live second production estate doubles your operational surface and usually lowers reliability in the first year.
Do we need a provider with a region in our country?
You need one close enough for latency and lawful enough for your data. Those are different tests. Check whether a regulator or a customer contract names a jurisdiction first, then check round-trip latency from your actual users. Many businesses are better served by a region one country away with three zones than a single-zone local option.
Can Eudora manage this if we have no in-house cloud engineer?
Yes. We work remotely with clients worldwide, so the assessment, build and ongoing management happen over calls, shared documents and your own consoles. If you need hands-on-site work in Sri Lanka, that is handled through eudora.lk instead.
Want a short, honest provider recommendation for your workloads rather than a sales deck? Tell us what you run and we will map it. Get in touch with Eudora Technology to talk about your project.
_-_detail_view.jpg){kind=link}
_-_overview.jpg){kind=link}
.jpg){kind=link}
