The hyperscalers do not sell you a server. They sell you a metered utility with several hundred billable dimensions, and outbound traffic is the one that surprises people. Here is what 5 TB of egress costs on each platform, and what it costs on RackWorks.

A Different Kind of Comparison

Part 1 of this series compared RackWorks against DigitalOcean, Vultr, and Akamai Cloud. That was a comparison between products that are roughly the same shape: a virtual machine, a monthly price, an included transfer allowance. The differences were real, but they were differences of degree.

Amazon Web Services, Microsoft Azure, and Google Cloud are not the same shape. They do not sell you a server. They sell you a metered utility with several hundred billable dimensions, and the server is one of them. Comparing a $40.00 monthly plan against that is not a pricing comparison, it is a comparison of two philosophies about who should carry the risk of a busy month.

So this article does something specific. It ignores everything the hyperscalers are genuinely good at, which is a lot, and focuses on the one number that surprises more people than any other: what it costs to send data out of the cloud.

The Egress Number

Every one of these platforms charges you to move data from their network to the public internet. Inbound is free. Outbound is not. Here are the current rates for standard internet egress from a compute instance in a US region:

ProviderFree per monthRate after that
Amazon Web Services100 GB$0.09 per GB to 10 TB, then $0.085 to 50 TB, $0.07 to 150 TB, $0.05 above
Microsoft Azure100 GB$0.087 per GB to 10 TB, then $0.083 to 50 TB, $0.07 to 150 TB, $0.05 above
Google Cloud, Premium tier1 GiB$0.12 per GiB to 1 TiB, $0.11 to 10 TiB, $0.08 above
Google Cloud, Standard tier200 GiB$0.085 per GiB to 10 TiB, then $0.065, then $0.045
RackWorks, unlimited planseverything$0.00

Read the Google Cloud rows twice. Premium tier is the default. Premium tier includes one gibibyte free per month. Not one hundred. One.

Note also that Google bills in gibibytes rather than gigabytes, so its effective rate is slightly higher than the headline number suggests when you convert to the units everyone else uses.

What That Means on a Real Invoice

Consider an application serving 5 TB of outbound traffic in a month. That is not an extraordinary number. A moderately popular podcast, a photo gallery, a software download mirror, a video-heavy marketing site, or a busy public API will all get there.

ProviderEgress cost for 5 TB out
Google Cloud, Premium tierabout $560.00
Amazon Web Servicesabout $441.00
Microsoft Azureabout $426.00
Google Cloud, Standard tierabout $408.00
RackWorks KVM Growth$0.00, and the whole machine is $20.00

Push that to 10 TB and the AWS egress line alone reaches roughly $891.00. Azure lands near $861.00. Google Cloud Premium tier passes $1,100.00. That is the transfer charge by itself, before the instance, before the storage, before the load balancer, before the address.

The RackWorks number does not move. KVM Growth is $20.00 per month at 1 TB of traffic and $20.00 per month at 10 TB. KVM Pro at $40.00 and KVM Heavy at $80.00 behave the same way, as does the entire dedicated core line.

A single month of 10 TB egress on AWS costs more than two years of KVM Pro.

That is arithmetic, not rhetoric. $891.00 divided by $40.00 is a little over twenty two months, and that compares one line item from one month against the entire cost of the alternative. Run it across a year of steady traffic and the comparison stops being interesting.

The Second Problem: Nobody Can Predict the Bill

Egress is the headline, but the deeper issue is that a hyperscaler invoice is a metering report, not a price. A single modest application can generate charges across compute hours, block storage capacity, block storage IOPS, snapshot storage, object storage capacity, object storage requests, egress, inter-zone transfer, NAT gateway hours, NAT gateway data processing, load balancer hours, load balancer capacity units, public address hours, DNS zones, DNS queries, log ingestion, log retention, metrics, alarms, key management operations, and support plan percentage.

Every one of those is defensible on its own. Collectively they mean that no engineer on the team can answer "what does this application cost per month" without opening a billing console, and the answer changes when traffic changes.

Two examples of charges that catch people specifically:

Public IPv4 addresses are billed by the hour. AWS charges $0.005 per hour for every public IPv4 address, in use or not. That is about $3.65 per month per address. If you run twenty small instances, you are paying roughly $73.00 per month purely for the privilege of having addresses. On RackWorks, one IPv4 address is included with every plan, plus a routed IPv6 /64 per machine, and additional addresses are a flat monthly add-on you can see on the pricing page before you buy.

Cross zone and NAT gateway traffic is billed like egress. Traffic between availability zones carries a per gigabyte charge in each direction, and traffic through a managed NAT gateway carries a data processing fee on top of an hourly charge, even though none of it ever leaves the provider's network. Architectures designed for resilience generate these charges by construction. A great deal of cloud cost engineering consists of restructuring applications to avoid paying for their own internal plumbing.

Support Costs Extra, and It Is a Percentage

On the major platforms, the ability to open a ticket with an engineer is a paid product priced as a percentage of your bill.

AWS restructured its support tiers in 2026. The entry paid plan is now Business Support+, priced as the greater of $29.00 per month per account or a sliding percentage of your monthly charges: nine percent of the first $10,000, seven percent from there to $80,000, five percent to $250,000, and three percent above that. Enterprise Support starts at $5,000.00 per month. The older Developer and Business plans are being retired on January 1, 2027.

The floor went down. The model did not change. Read it as a business model: the more you spend, the more you pay for the right to ask why you are spending it. And because the percentage is calculated per account, an organization running many accounts pays the minimum many times over.

RackWorks has one support level, included, for everyone, at every plan size. A $4.00 container customer and an $80.00 KVM Heavy customer reach the same place. There is no premium tier because there is nobody to upsell.

What About the Free Tiers and the Credits

Every hyperscaler will hand a new account a pile of credits. This is not generosity, it is customer acquisition, and it works because it is genuinely useful. If you are prototyping, take the credits.

The thing to understand is what happens at the end. The credits expire, the architecture is already built around managed services with no portable equivalent, the data is already sitting in object storage that charges you to remove it, and the team has already learned one vendor's console. The switching cost was created during the free period. That is the design.

Egress pricing is the load-bearing part of that design. It costs nothing to put data in and a lot to take data out, which means the cost of leaving grows in direct proportion to how much you have committed. Regulators in several jurisdictions have taken an interest in exactly this dynamic, and some providers have softened the rules for full account migrations. The day to day rate for ordinary traffic has not moved.

The RackWorks Model, Stated Plainly

There is one number. It is the plan price. Here is the whole catalog:

PlanResourcesTransferMonthly
LXC Nano 1 core, 1 GB, 20 GB SSD2 TB$4.00
LXC Start 2 cores, 2 GB, 30 GB SSD4 TB$7.00
LXC Growth 4 cores, 4 GB, 60 GB NVMeUnlimited$10.00
KVM Nano1 vCPU, 1 GB, 25 GB SSD2 TB$6.00
KVM Start2 vCPU, 2 GB, 40 GB SSD4 TB$10.00
KVM Growth4 vCPU, 4 GB, 80 GB NVMeUnlimited$20.00
KVM Pro6 vCPU, 8 GB, 160 GB NVMeUnlimited$40.00
KVM Heavy8 vCPU, 16 GB, 250 GB NVMeUnlimited$80.00
VDS-11 dedicated core, 2 GB, 40 GB NVMeUnlimited$25.00
VDS-22 dedicated cores, 4 GB, 80 GB NVMeUnlimited$35.00
VDS-44 dedicated cores, 8 GB, 160 GB NVMeUnlimited$55.00

Every one of those includes an IPv4 address, an IPv6 /64, portal-managed snapshots and backups, reverse DNS you set yourself and that applies instantly, a browser console, a portal firewall, live resource graphs, OS reinstall on demand, and access to all 50+ one-click applications. Annual billing takes 15 percent off. Two year billing takes 20 percent off. Payment by card, PayPal, or Bitcoin.

There is no metering console, because there is nothing being metered.

Where the Hyperscalers Are Genuinely the Right Answer

This would be a dishonest article if it stopped there, because the reason these companies are enormous is that they are extremely good at things RackWorks cannot do at all.

Managed data services. A multi-region managed database with automatic failover, point in time recovery, and read replicas is a serious piece of engineering. If your business depends on one, buy it from someone who runs thousands of them.

Elastic scale. Traffic that varies by an order of magnitude between Tuesday and Black Friday is exactly the problem elastic infrastructure was invented to solve. Paying per hour for capacity you use for six hours a year is the correct decision.

Global distribution. Dozens of regions, integrated content delivery, and anycast routing at the platform level are not things a single facility can replicate.

Specialized hardware. GPU fleets for training, custom accelerators, and confidential computing enclaves are available on demand and nowhere else.

Compliance and procurement. Some contracts require certifications that only large providers maintain. That is a business constraint, not a technical one, and it is decisive when it applies.

Deep ecosystems. Infrastructure as code, event-driven glue, managed identity, and a vast third-party tooling market are real productivity gains for large teams.

If your workload needs those, use them. Nothing in this article argues otherwise.

The Workloads That Should Not Be There

The argument is narrower and, we think, harder to dispute: an enormous number of applications running on hyperscalers today are steady state workloads that never scale elastically, never leave one region, and never touch a managed service beyond a database that could run perfectly well on the same box.

For those, the hyperscaler is charging a substantial premium for optionality that will never be exercised, and metering a bill that could have been a fixed number.

Concretely, these belong on a fixed-price server:

  • A company website, a marketing site, or a documentation site
  • An internal tool used by a known number of employees
  • A self-hosted stack: file sync, password vault, VPN endpoint, chat, wiki, CRM, invoicing, analytics
  • A media server or a personal photo library
  • A staging or development environment that mirrors production
  • A monitoring and log aggregation host
  • A build agent or a repository server
  • An API with predictable traffic
  • Anything that pushes real outbound volume with a shape you can forecast

Most of that list is available on the RackWorks marketplace as a one-click install, which means the migration is often shorter than the meeting where you decide to do it.

A Realistic Way to Test the Claim

Do not migrate anything important on the strength of a blog post. Do this instead:

  1. Open your current cloud bill and find the total for one specific application, including its egress, its addresses, its storage, and its share of support.
  2. Write down the resources that application actually uses: cores, memory, disk, and monthly outbound traffic.
  3. Find the row in the table above that covers it. For most single-application workloads that row costs between $10.00 and $55.00.
  4. Move a non-critical copy first. A staging environment or a monitoring host is a good candidate because breaking it costs nothing.
  5. Run it for a month and compare the invoices.

If the numbers do not work, you have lost one month at $10.00 and learned something concrete. If they do work, the same logic applies to every other steady workload you are running.

What We Are Not Claiming

RackWorks operates from a single facility in Southfield, Michigan, on a 1 Gbps uplink with Detroit-IX connected upstream routing, and publishes a 99 percent uptime commitment with an externally monitored status page. That is a smaller promise than a global platform makes, stated in a smaller number, and it is stated that way because it is the one that can be kept.

Anyone whose application genuinely cannot tolerate a maintenance window needs redundancy across facilities and probably across providers. No single region from any vendor, at any price, changes that.

What RackWorks does claim is narrow and testable. The bill is fixed. The transfer on most plans has no ceiling. Support is included. The hardware is named publicly. The one-click marketplace is real. And nothing in the pricing model is designed to make leaving expensive.

Start with the virtual machine plans, the container plans, or the dedicated core line, and compare it against the invoice you already have.

Back to Blog