Ask three engineers which cloud is cheapest and you will get three confident, contradictory answers. All three can be right, because “cheapest” is the wrong question. AWS, Google Cloud, and Azure price commodity compute within a narrow band of each other. Your bill is decided by the pricing model wrapped around that compute: how discounts are earned, whether you have to remember to claim them, and what it costs when data leaves the building.
This comparison focuses on those mechanics instead of list prices that go stale every quarter. Cloud pricing changes constantly, so treat everything here as structural: the models are stable, the exact rates are not. Run your own workload through each provider’s official calculator before committing. What follows reflects how the three hyperscalers charge as of early 2026.
Why List Prices Tell You Almost Nothing
The per-hour price of a generic 4-vCPU virtual machine is similar on all three clouds. Engineers who stop at that comparison get surprised later, because the real money lives elsewhere: commitment mechanics, discount automation, and data transfer. Two teams can run identical workloads and pay very different bills depending on commitments, discount automation, and how much data crossed region or internet boundaries.
Three levers decide the outcome: how you pay for compute (on-demand, committed, or spare capacity), how discounts reach you (bought in advance or applied automatically), and what leaves the cloud (egress). The rest of this article walks through each lever, provider by provider.
AWS: Maximum Choice, Maximum Rope
AWS prices EC2 compute in three main ways. On-Demand means no commitment and the highest hourly rate. Savings Plans let you commit to a dollar-per-hour spend for one or three years in exchange for a lower rate, and the flexible plans apply that commitment across instance families and regions, so you are not locked to one machine shape. Spot Instances sell spare capacity at steep discounts, with the catch that AWS can reclaim the capacity with a two-minute warning.
Where AWS wins: the Savings Plans model is the most flexible commitment mechanism of the three clouds, because one plan can cover many workload shapes. The Spot market is the deepest, which matters if you run large batch or fault-tolerant jobs. And with the broadest service catalog, you can usually find the exact primitive you need instead of approximating it.
Where it bites: the catalog is enormous, and the billing reflects that. Data transfer shows up as multiple line items, and it is easy to leave expensive resources running. Cost Explorer and Compute Optimizer help, but the model rewards teams that actively manage commitments and clean up idle resources.
Google Cloud: The Discounts Find You
Google Cloud’s signature move is the sustained-use discount: run a Compute Engine VM for most of the month and the discount applies automatically, with no upfront commitment and no paperwork. For predictable workloads, committed use discounts offer deeper cuts in exchange for a one- or three-year commitment. Billing is per-second with a one-minute minimum, so short-lived jobs are not rounded up to full hours.
Where Google Cloud wins: automatic discounts mean less spreadsheet work and fewer calendar reminders to buy reservations. Billing is the simplest of the three to reason about, and the always-free tier covers real experiments and small projects.
Where it bites: sustained-use discounts do not stack with committed-use discounts on the same resources, so you still need to think about which mechanism covers which workload. Region selection affects both price and availability, and egress is billed here just like everywhere else, which surprises teams that assumed Google’s network would be the exception.
Azure: Built for the Microsoft Shop
Azure prices VMs through pay-as-you-go, Reserved VM Instances for one- or three-year commitments, and Spot VMs for spare capacity. Its standout lever is Azure Hybrid Benefit: if your company already owns Windows Server or SQL Server licenses, you can apply them to Azure VMs instead of paying the license-included rate. For enterprises deep in the Microsoft ecosystem, this single benefit can outweigh every other pricing difference covered here.
Where Azure wins: that license lever. A Windows-heavy enterprise comparing list prices without Hybrid Benefit is not comparing the price it would actually pay.
Where it bites: the pricing pages are famously hard to navigate, and the advantage concentrates in Microsoft-licensed workloads. For pure Linux and open-source stacks, Azure’s pricing story is ordinary: competitive, but with no equivalent of Hybrid Benefit to tip the scales.
Egress: The Line Item That Eats Budgets
All three clouds follow the same rule: inbound data transfer is free, outbound is metered. Traffic between services in the same region is typically free or cheap. Traffic that crosses regions or leaves for the internet is billed per gigabyte in volume tiers, where the per-GB rate steps down as your volume grows. The exact tier boundaries and rates differ by provider and region, and they change, so check the current tables.
Why egress dominates so many bills: every cross-region backup, every multi-cloud hop, every user download, and every CDN origin fetch counts as outbound. Teams running video, file delivery, or analytics pipelines routinely discover that data movement, not compute, is their largest line item. It is the cost most often modeled last and regretted first.
Four mitigations actually work. Keep data and compute in the same region so traffic stays on the free or cheap path. Use each provider’s private interconnect options for hybrid traffic instead of paying internet egress rates to your own data center. Cache aggressively at the edge so repeated requests never reach origin. And if you run multi-cloud, model the transfer costs between providers up front, because the architecture that looks cheapest on compute alone often loses on data movement.
Cost Drivers, Side by Side
| Cost driver | AWS | Google Cloud | Azure |
|---|---|---|---|
| On-demand compute | EC2 On-Demand, per-second billing | Compute Engine, per-second (1-min minimum) | VMs, per-second (1-min minimum) |
| 1- or 3-year commitment | Savings Plans (flexible $/hr commit) | Committed use discounts | Reserved VM Instances |
| Automatic discounts | No, must be purchased | Yes, sustained-use | No, must be purchased |
| Spare capacity | Spot Instances (2-min reclaim notice) | Spot VMs (30-sec notice) | Spot VMs (30-sec notice) |
| Standout lever | Commitment flexibility | Discounts apply themselves | Hybrid Benefit for Microsoft licenses |
| Egress model | Tiered per-GB internet egress | Tiered per-GB internet egress | Tiered per-GB internet egress |
| Free tier | 12 months plus always-free | Always-free tier | 12 months plus new-account credit |
Choosing for Your Workload
Match the model to the workload, not the brand to the hype:
- Steady 24/7 baseline: buy commitments and compare Savings Plans against committed use discounts and Reserved Instances for your exact machine shapes.
- Microsoft-licensed enterprise: price Azure with Hybrid Benefit applied. It is usually not close.
- Spiky, batch, or fault-tolerant: Spot and preemptible capacity on any of the three. Price the reclaim risk into your architecture, not just the discount.
- Bandwidth-heavy (video, downloads, multi-cloud): model egress first. It can exceed your compute spend, and it reverses naive rankings.
- Small team, new project: use the free tiers and favor the provider whose discount model needs the least babysitting.
The honest method: list your vCPU-hours, storage, and egress gigabytes by destination, run the numbers through all three official calculators, and revisit yearly. Anyone selling a universal “cheapest cloud” ranking is selling something.
FAQ
Which cloud is cheapest in 2026?
There is no universal winner. Commodity compute prices sit in a narrow band across AWS, Google Cloud, and Azure. Commitments, license benefits, and egress decide the bill, so the cheapest cloud is the one whose pricing model fits your workload.
Do AWS, Google Cloud, and Azure all charge egress fees?
Yes. Inbound transfer is free on all three. Data leaving the cloud, whether to the internet or another region, is metered per gigabyte in volume tiers. Traffic between services in the same region is typically free or cheap.
What causes the biggest surprise on cloud bills?
Data transfer, followed by idle resources: VMs, disks, and load balancers nobody turned off. Set up billing alerts on day one; every provider offers them free, and they are the cheapest insurance in cloud computing.
Is it worth switching clouds to save money?
Rarely for cost alone. The engineering time a migration consumes usually exceeds a year of potential savings. Negotiate pricing, buy commitments, and delete idle resources first. Switch clouds for capability or architecture reasons, with cost as a secondary benefit.
The Short Version
Do not compare list prices. Compare models: how discounts are earned, whether they apply automatically, what your existing licenses are worth, and what it costs when data leaves. Egress first, commitments second, official calculators always. Revisit yearly, because the only constant in cloud pricing is that it changes.