AWS vs Azure vs Google Cloud: Which to Open First in 2026

AWS vs Azure vs Google Cloud is the wrong question for a first cloud account, and the three vendors know it: every comparison page ranks features that a beginner won't touch for a year, while the things that actually decide your first six months (how the free credit works, what closes your account, which console makes sense at 11 p.m., and whose bill surprises you first) are buried in the fine print. This guide answers the question people actually type: which one should I open an account with, this month, if I want to learn cloud, run a small project, or get hired. Every price, credit and time limit below is current as of September 2026, and the honest answer changes depending on which of those three goals is yours.

The three tabs Jake had open, and the one that closed his account

Jake runs a phone-repair shop and wanted a cheap server for a booking page and a place to learn "cloud", because a customer's IT son had told him it was the thing to know. He opened three tabs: AWS, Azure, Google Cloud. All three said free. All three wanted a credit card.

He picked the one with the biggest number on the page ($300) and forgot about it for four months. When he came back, the project was gone: the trial had ended, he had never upgraded, and after the grace period the resources were deleted. Nothing was charged. Nothing was there, either.

"I thought free meant it just sits there," he told Ethan.

"Free means three different things on those three pages," Ethan said. "One of them means free for 90 days and then we delete it. One means free for 30 days, then some things stay free for a year. One means free for 6 months, then the account closes itself. None of them is bad. They're just three different clocks, and the page never puts the clocks side by side."

That is what this post does. Not "which cloud is best" (there isn't one), but which cloud is right for the specific first thing you're trying to do, with the clocks written down.

The three free tiers, with the expiry rules that matter more than the credit

Read this table twice. The first column is the number the marketing page shows. The last two are the numbers that decide whether your project still exists in the spring.

Three things the table makes obvious that the vendor pages don't:

- AWS's clock is the longest but it has a hard stop. Six months is a semester. But the Free Plan account closes; it doesn't quietly become a paid account with a bill. That is a safety feature for beginners and a surprise for people who assumed it would keep running. - Azure's headline is the smallest and shortest, but the 12-month tail is the most generous for compute. Three different burstable VM families at 750 hours a month each is more free server time than either rival gives after the trial. - Google's headline is the biggest, and its Always Free VM is the only one that never expires. But the 90-day trial deletes things when it ends, and the grace period is the only net.

"So the $300 one is the one that deleted my server," Jake said.

"The $300 one is the one you didn't upgrade in 90 days," Ethan said. "It told you. It's just that you read the number, not the clock."

What each free tier actually gives you, service by service

The credits pay for anything. The always-free allowances are the part that matters after the credits are gone, because that's what a hobby project runs on for years. Here is what each vendor gives you, restricted to the services a first-year user touches.

AWS: the Free Plan and the 30+ always-free services

The 2025 change replaced the old "12 months of t2.micro" model with credits. On sign-up you get $100 of credit; completing a handful of guided activities (launching an instance, setting a budget and so on) can add up to $100 more, for a ceiling of $200. Credits last six months. The Free Plan is "limited to select services," so some services (the expensive ones) simply aren't available until you upgrade.

Independently of the plan, AWS keeps a list of services that are always free within monthly limits. The ones that matter early: Lambda's monthly request allowance, DynamoDB's free storage and throughput, a bundle of CloudWatch metrics and alarms, a modest SNS and SQS allowance, and the 100 GB a month of data transfer out to the internet that applies across all AWS services. Our free tier types post lists all three kinds (always-free, trial, credit) on one page, and the credits model explainer walks through what happens at month six.

What isn't free and catches beginners: a NAT gateway (hourly plus per-GB), an unattached Elastic IP, an idle load balancer, and RDS left running. None of those is exotic; all of them appear in our list of the ten most common surprise charges.

Azure: $200 for 30 days, then 12 months of the good stuff

Azure's free account is the odd one out because the 12-month allowances are, for compute, the richest of the three. In the first 12 months you get 750 hours a month each of B1s (Intel), B2pts v2 (Arm) and B2ats v2 (AMD) burstable VMs, 750 hours of a Flexible Server B1MS instance for PostgreSQL and for MySQL with 32 GB storage and 32 GB backup each, a Standard-tier container registry with 100 GB, and a list of storage and networking allowances. Beyond the year, Microsoft counts "65+" always-free services, including Azure Functions (a million requests a month), a small Azure SQL Database, Azure DevOps and a set of AI services with monthly free calls.

What catches beginners on Azure: the resource-group model. A VM is not one thing; it is a VM, a disk, a network interface, a public IP and a network security group, each a separate resource. Delete the VM and leave the disk and the IP behind, and you keep paying for both. Azure's fix is to delete the resource group, which takes everything with it. Learn that on day one.

Google Cloud: $300 for 90 days, and the VM that never expires

Google's trial is the largest headline credit and the shortest clock. Its Always Free tier is where it wins for a certain kind of user: one non-preemptible e2-micro VM per month in three US regions (Oregon, Iowa, South Carolina) with 30 GB of standard disk; 5 GB of regional Cloud Storage; 1 TiB of BigQuery queries and 10 GiB of BigQuery storage a month; 2 million Cloud Functions invocations and 2 million Cloud Run requests a month; and 100 GB of outbound data transfer from North America. The e2-micro is small (shared-core, 1 GB memory) but it runs a tiny site, a bot or a learning box forever, for nothing.

What catches beginners on Google Cloud: the region rule on the free VM (only those three US regions), and the fact that the trial's end deletes resources rather than pausing them. Upgrade to a billing account before day 90 if you want anything to survive, and set a budget alert the day you do.

The job-market answer, honestly

If the goal is get hired, the comparison stops being about features. Three things are true in 2026 and have been for years:

1. AWS has the most job postings that name a specific cloud, in the US, the UK, India and Australia alike. Azure is second and closest in enterprise and government roles, especially where Microsoft 365 is already the office suite. Google Cloud is third overall and strongest in data engineering, analytics and machine-learning roles.
2. Skills transfer. A virtual machine, a bucket, an identity policy, a load balancer, a managed database: all three sell the same nouns under different names. Learn one properly and the second takes weeks, not months. The equivalence table below is the map.
3. Certifications track the same order. AWS Cloud Practitioner and Solutions Architect Associate are the most-requested entry certifications, with Azure Fundamentals (AZ-900) and Azure Administrator close behind in Microsoft-heavy markets. Google's Associate Cloud Engineer is respected but requested less often outside data roles. Our AWS certification changes post covers what to book and when.

Jake's customer's IT son was right, in other words, but for a narrower reason than he gave: not "AWS is best," but "AWS is where the most doors are, and a key that opens AWS doors is easy to re-cut for the other two."

The one-hour console test (do this before you decide)

Every comparison ignores the thing that decides whether you'll actually keep learning: whether the console makes sense to you. Here's a test that takes about twenty minutes per vendor and costs nothing. Open an account (or use the free trial), and do exactly these four things:

1. Launch the smallest Linux VM the free tier allows, connect to it from the browser, and run uptime.
2. Create a storage bucket, upload a photo, and make that one file public.
3. Find the page that shows you the bill so far and the page that lets you set a budget alert.
4. Delete everything, and then confirm, from the billing page, that nothing is still running.

Time each step. Note where you got lost. The vendor where step 4 was easiest is the vendor least likely to surprise you with a bill; the vendor where step 1 was easiest is the one you'll practice on. In our experience the answers are roughly:

The last row is the one to remember: Azure and Google let you delete a container of everything (a resource group, a project). AWS has nothing quite like that at the beginner level, which is why "I stopped everything and I'm still being billed" is an AWS search far more than an Azure or Google one. It's also why we wrote a three-layer billing alert setup; the alerts are the substitute for a delete-everything button.

"So AWS is harder to clean up," Jake said.

"AWS gives you more pieces," Ethan said. "More pieces, more to forget. The other two put the pieces in a box you can throw away. That's a real difference, and it matters more for a beginner than any feature list."

The equivalence table: the same 20 things under three names

Most "AWS vs Azure vs GCP" confusion is vocabulary. The services below are the same idea; learn the idea once and the names are a lookup.

Two honest footnotes. Google Cloud's functions product has been folded into Cloud Run, so older tutorials that say "Cloud Functions" are describing what the console now calls Cloud Run functions. And Azure's identity system has been renamed twice (Active Directory → Entra ID), so the name you see in a 2022 course is not the name in the portal.

The prices that actually matter in year one

Feature-by-feature pricing comparisons are a full-time job and go stale in a month. For a first account, four prices decide the bill, and here they are for the cheapest useful options, in the vendors' primary US regions, as of September 2026. Treat them as the shape of the market, and check the vendor's calculator before you commit.

What the table says, read plainly:

- The cheapest VM is within a cent an hour on all three. Nobody chooses a cloud on VM price any more. What differs is what's free: Azure's 750 hours a month of three VM families for a year, and Google's one e2-micro forever. - Data out is where all three make their money. The 100 GB a month free is the same on all three; the per-GB after it is within a few cents. Our data transfer pricing map explains why "free in, expensive out" is the rule everywhere. - The small-print prices are where accounts get hurt. NAT gateways, static IPs, load balancers, snapshots, logs. Every cloud has these; AWS's versions are the most searched, which is why our AWS costs series exists. The equivalent Azure surprise is the orphaned disk; the Google one is the forgotten Cloud SQL instance.

Where each one is genuinely better (not marketing, the reasons people switch)

Every vendor page claims leadership in everything. Here's what practitioners actually cite when they pick one over another:

AWS: breadth, maturity and the job market

- The most services and the most regions, which means the odd thing you need (a managed queue, a specific database engine, a GPU type) is more likely to exist and to have been there for years. - The most third-party tooling and the most answers online. Every error message has been hit by a thousand people before you. That is worth real money when you are stuck at midnight. - The largest share of job ads, in every English-speaking market. - Weaknesses people cite: the console is the busiest of the three; the pricing has the most line items; permissions (IAM) are powerful and unforgiving, which is why "AccessDenied" is the most searched AWS error there is.

Azure: the enterprise and Microsoft integration

- If your company runs Microsoft 365, Windows Server and Active Directory, Azure's identity (Entra ID) is already your login, and hybrid tools (Azure Arc, Entra Connect) are built for that exact shape. - Licensing perks. Azure Hybrid Benefit lets existing Windows Server and SQL Server licenses be reused on Azure VMs, which can make a Windows workload materially cheaper there than anywhere else. - The richest 12-month free tier for compute and databases, as the table above shows. - Weaknesses people cite: the portal's depth (resource groups, subscriptions, management groups, tenants); naming that changes; and a history of service outages that make the news, which is partly a function of how much of the world's office software sits on it.

Google Cloud: data, networking, Kubernetes and the cleanest console

- BigQuery is the reason a lot of teams have any Google Cloud account at all: a warehouse where you pay per query, with 1 TiB a month free. The Always Free tier reflects that: it's built for someone who wants to analyze things. - Kubernetes was born at Google, and GKE is the managed version most people find easiest. - Networking is Google's own backbone, and the project-based console is the simplest of the three for a beginner to reason about (everything lives in a project; delete the project and it's gone). - Weaknesses people cite: fewer regions and services than AWS; a smaller job market; and a track record of retiring products, which makes some teams nervous about building on the newest thing.

Jake's booking page, costed on all three clouds for its first month

Tables of unit prices are abstract. Here is one real, small workload priced three ways: a booking page for a repair shop. One small Linux VM running a web app, a tiny managed PostgreSQL database, 10 GB of object storage for photos, a domain with DNS, and about 20 GB of traffic out a month. Prices are on-demand rates for the primary US regions in September 2026, rounded, and the "free-tier month" column assumes a brand-new account.

Three lessons hide in that table:

- Month one is free everywhere, give or take a DNS zone. The free tiers really do cover a small app. The differences start in month two. - Azure's year-one is the cheapest way to run a small VM plus a managed database, because both the B1s VM and the Flexible Server B1MS are in the 12-month free list. After month 12 Azure looks like AWS. - Google's Always Free VM makes the cheapest long-run bill for a tiny site, but the moment you want a managed database you're paying, and its smallest Cloud SQL costs about what AWS's does.

For Jake's specific shop, the honest recommendation was Azure for year one, then a fixed-price bundle (an AWS Lightsail $5-$12 plan, or Azure's equivalent) once the free hours ended, because a fixed monthly number is worth more to a shop owner than the last dollar of savings.

"So I'd have saved money by not picking the $300 one," Jake said.

"You'd have saved the project," Ethan said. "The money on all three is close to nothing for what you're running. The clock was the whole story."

Regions for UK, Australian and Indian readers: what changes

Everything above uses US prices because that's what the vendors publish first. If you're reading from the UK, Australia or India, three things change, in this order of importance.

Prices are higher outside the US, by a predictable margin. The same small VM costs roughly 10-20% more in London, Sydney or Mumbai on all three clouds, and data out is a few cents a gigabyte more. The shape of the comparison (which one's free tier is best, which one wins for Windows) doesn't change. The free-tier VMs mostly do: Google's Always Free e2-micro is US regions only; AWS credits and Azure's 12-month free hours work in your local region.

Data residency is a real requirement, not a nice-to-have. UK GDPR, the Australian Privacy Act and India's DPDP Act all push customer data toward local storage. All three clouds have regions in London, Sydney (and Melbourne), and Mumbai (AWS also has Hyderabad; Azure has Pune and Chennai; Google has Delhi). Pick the local region for anything with customer records in it, on day one, because moving a database between regions later is a migration, not a setting.

Latency to your users decides the region more than price does. A booking page in Sydney served from Oregon adds 150-plus milliseconds to every request. That's the difference between "snappy" and "a bit slow". Serve from the region your customers are in, and use a CDN (CloudFront, Front Door, Cloud CDN) for the static parts if they're spread out.

One more regional note for the job-market question: the AWS-first advice holds in the UK and Australia (with Azure a strong second in government, banking and anywhere Microsoft 365 rules, which is a lot of the UK public sector), and it holds even more firmly in India, where AWS and Azure dominate the postings and Google Cloud is mostly data roles.

Pick by goal: the decision table

"That table has AWS on top for the job, Google for the free server, and Azure for the shop," Jake said. "So the answer is really 'it depends'."

"The answer is 'it depends on one thing you already know about yourself'," Ethan said. "Which is different from 'it depends' as an excuse. You know whether you want a job, a project, or a cheaper version of what you already run. Pick the row."

The seven first-account mistakes, on all three clouds

These cost money or projects on every vendor. Do the fixes on day one.

1. Using the root or owner account for daily work. Create a second user with admin rights and MFA; lock the root away. Our root vs IAM user guide, and the same logic holds for the Azure subscription owner and the Google project owner.
2. No MFA on the account that has a credit card. Three minutes (AWS steps); the single most effective thing you can do against a stolen password.
3. No budget alert. Every cloud has budgets; none is on by default. Set one for a number that would make you act.
4. Leaving a "stopped" VM's disk and IP alive. Stopped is not deleted, on any of the three. Delete the resource group (Azure), the project (Google), or the volumes and Elastic IPs (AWS).
5. Picking the wrong region. Prices, free-tier eligibility (Google's free VM is US-only) and latency all depend on it. Our region and availability zone map is written for AWS, and the idea is identical elsewhere.
6. Committing keys to a public repository. Bots scan GitHub in minutes. If you ever do, the first 30 minutes after a leaked key are the same on every cloud: revoke, rotate, check the bill in every region.
7. Assuming the free tier is a plan. It's a runway. Know the clock (6 months, 30 days plus a year, 90 days) and set a calendar reminder two weeks before it ends.

"Should I learn all three?" and the multi-cloud question

For a beginner: no, not at once. Learn one to the point where you can build and tear down a small three-tier app (VM or container, database, storage, DNS) without a tutorial. That takes two to four months of evenings. Then the second cloud takes weeks, because it's the same nouns.

For a company: "multi-cloud" is usually a fact rather than a plan. Teams end up on two clouds because of an acquisition, a data product that lives on BigQuery, or a Microsoft licensing deal. Deliberately splitting one workload across clouds for resilience is rare, expensive (data out charges apply between clouds too), and mostly done by companies with dedicated platform teams. If you are reading a beginner's comparison, you are not that company yet, and that's fine.

Ethan's rule for Jake: "Be fluent in one, conversational in a second, and able to read the menu in the third."

Switching later: how hard is it really?

Easier than the vendors imply for the basics, harder for the clever bits. A VM image, a Postgres database and a bucket of files move between clouds in an afternoon. Code that leans on a vendor-specific service (Lambda triggers wired to DynamoDB streams, Azure Functions bound to Cosmos DB, BigQuery scheduled queries) takes redesign, not copying. That is why the migration-strategy literature talks about rehost (move it as-is) versus refactor (rebuild it for the new platform); our on-premises to AWS migration guide walks through the same seven strategies, and they apply just as well cloud-to-cloud.

The practical advice for a first project: keep the app boring. A Linux VM or a container, Postgres, object storage, a plain DNS record. Boring apps move. Clever apps marry.

AWS vs Azure vs Google Cloud: the questions people search

Which is best for beginners: AWS, Azure or Google Cloud?

Google Cloud has the cleanest console and the only free VM that never expires; AWS has the most tutorials, jobs and answers online; Azure is easiest if you already use Microsoft 365. For a beginner with a career goal, AWS; for a beginner with a small project, Google Cloud; for a Microsoft-shop admin, Azure.

Which cloud has the best free tier in 2026?

It depends on what "best" means. Biggest credit: Google ($300 for 90 days). Longest safe clock: AWS (6 months, $100-$200, account closes rather than bills). Most free compute after the trial: Azure (750 hours a month each of three VM families for 12 months). Only VM that never expires: Google's e2-micro.

Does the AWS free tier still give 12 months of a free t2.micro?

Not for accounts opened since the 2025 change. New accounts get a Free Plan with $100 credit on sign-up (up to $200 with activities) for six months, plus the always-free services. The old 12-month model applies only to accounts created before the change.

What happens when the Google Cloud free trial ends?

After 90 days or when the $300 is used, the trial ends. If you have not upgraded to a paid billing account, your account is suspended and resources are deleted, with a 30-day recovery window. Upgrade before day 90 if you want anything to survive; the Always Free allowances continue on a paid account.

What happens after the Azure $200 credit?

After 30 days or once the $200 is used, you must upgrade to pay-as-you-go to keep using paid services. There is no automatic charge. Once upgraded, the 12-month free amounts continue until the account's first anniversary, and the always-free services continue indefinitely.

Is AWS cheaper than Azure or Google Cloud?

For the basic pieces (small VMs, storage, data out), the three are within cents of each other and the answer changes with each price cut. Real cost differences come from licensing (Azure Hybrid Benefit for Windows/SQL Server), free tiers, committed-use discounts, and how much data you send out. Compare your specific workload in each vendor's calculator rather than trusting a headline.

Which cloud is best for getting a job?

AWS has the largest share of cloud job postings in the US, UK, India and Australia. Azure is close in enterprise and government roles; Google Cloud is strongest in data and machine-learning roles. Learn one well and the others transfer.

Which is better for Windows workloads, AWS or Azure?

Usually Azure, because Azure Hybrid Benefit lets you reuse existing Windows Server and SQL Server licenses, and identity integrates with Entra ID. AWS runs Windows perfectly well and has a bring-your-own-license option with more conditions.

Which cloud is best for data analytics?

Google Cloud, on the strength of BigQuery, which includes 1 TiB of queries and 10 GiB of storage free every month. AWS (Redshift, Athena) and Azure (Synapse, Fabric) are capable but their free allowances are smaller.

Can I use all three free tiers at the same time?

Yes. They are separate companies with separate accounts. Each needs a card for identity verification; none charges you during its free trial unless you upgrade.

Do I need a credit card for the free tier?

Yes, on all three, for identity verification. AWS's Free Plan does not charge the card unless you upgrade to the Paid Plan; Azure and Google do not charge during the trial and require an explicit upgrade.

What is the Google Cloud Always Free tier?

A set of monthly allowances with no expiry: one e2-micro VM (in Oregon, Iowa or South Carolina) with 30 GB disk, 5 GB of regional Cloud Storage, 1 TiB of BigQuery queries, 2 million Cloud Run requests, 2 million Cloud Run functions invocations, and 100 GB of North American egress, among others.

Is Azure or AWS better for a small business?

For a small business already on Microsoft 365, Azure's identity and licensing integration make it simpler. For a small business with a developer building something new, AWS's breadth and community are hard to beat. For a very small site or app, either vendor's fixed-price tier (AWS Lightsail, Azure App Service basic) removes most of the bill risk.

Which cloud is easiest to clean up so I don't get billed?

Google (delete the project) and Azure (delete the resource group) let you remove everything in one action. AWS requires deleting resources individually, which is why forgotten volumes, IPs and snapshots are the most common AWS surprise bills.

Are the clouds' market shares relevant to my choice?

Only as a proxy for jobs and community size. AWS leads, Azure is second, Google Cloud third, and the order has been stable for years. It tells you where the most answers and job ads are; it does not tell you which one fits your project.

Should I learn Terraform instead of one cloud's native tools?

Learn the cloud first, then Terraform. Terraform works on all three and is the most requested infrastructure-as-code skill, but it describes resources you need to understand already. Our infrastructure as code explainer covers why you stop clicking in the console.

Can I move from one cloud to another later?

The basics (VMs, SQL databases, object storage, DNS) move in an afternoon. Code built around vendor-specific services needs redesign. Keep a first project boring and it stays portable.

If you have been putting off opening a cloud account because every comparison ends in "it depends", I hope this one ended somewhere more useful: with the clocks written down, the prices side by side, and one row in the decision table that is yours. None of the three is a mistake. The only real mistake is the one Jake made, which was not reading the clock. If a price or a limit here has moved by the time you read it, tell me through the contact page and I'll correct it; these three pages change often, and this one stays accurate because readers write in.

Revision note. Written September 26, 2026; credits, clocks, always-free allowances and the VM and egress prices are current as of that date, and the job-market statements reflect the consistent picture across the major job boards rather than a single survey. Next check: when any vendor changes its trial length or credit, or when AWS revises the Free Plan. Open the account with the clock you can live with, and set the budget alert before you launch anything.