AWS Global Accelerator pricing in 2026

AWS Global Accelerator charges $0.025 for every hour an accelerator exists, used or not, and its per-gigabyte fee is added to your normal data transfer bill rather than replacing it. Both facts are on AWS's own pricing page, one paragraph apart, and neither appears in the worked example below them. An accelerator nobody is using costs $25.20 a month. AWS's example bill of $128 is really about $675 once the data transfer the same page says is charged "in addition" is counted.

What does an idle AWS Global Accelerator cost per month?

$25.20, on the 24-hours-times-30-days convention the Global Accelerator pricing page uses for its own example. Two meters run with no traffic at all.

The first is the fixed fee: "For every full or partial hour when an accelerator runs in your account, you are charged $0.025 until it is deleted." That is $18.00 for a 720-hour month. The second is addresses. The Global Accelerator developer guide says "For IPv4, Global Accelerator provides two static IPv4 addresses", and the VPC pricing page puts them on the meter: "public IPv4 addresses assigned to AWS Global Accelerator and AWS Site-to-Site VPN tunnel endpoints are charged as in-use public IPv4 address", at $0.005 per address-hour. Two addresses for 720 hours is $7.20.

The $7.20 is our combination of two AWS pages rather than a figure AWS prints: one says how many addresses an accelerator gets, the other says they are billable. The mechanics are in what public IPv4 addresses cost. A dual-stack accelerator gets four addresses, two of them IPv6, and only IPv4 is charged.

Multiply by the account quota. The Global Accelerator quotas page allows 20 standard accelerators and 10 custom routing accelerators per account by default. Thirty idle accelerators is $756 a month, $9,072 a year, for nothing. That is a hypothetical ceiling, not a prediction.

Is the Global Accelerator fixed fee prorated, or billed by the full hour?

It is not prorated. The pricing page bills "every full or partial hour", so a five-minute accelerator costs a full $0.025. The two addresses attached to that same accelerator do not work that way: the VPC pricing page says "The bill is calculated in one-second increments, with a minimum of 60 seconds."

So one resource carries two rounding conventions. Over a full hour the fixed fee is 2.5 times the address charge. Over five minutes it is 30 times, because one meter rounds up to the hour and the other does not. That matters for pipelines that create accelerators: a rollout that spins one up and tears it down inside a build is billed a whole hour each time.

Does AWS charge for a Global Accelerator that is disabled?

Yes, and AWS states it plainly on two pages. The pricing page: "For every accelerator that is provisioned (both enabled and disabled), you are charged a fixed hourly fee and an incremental charge over your standard Data Transfer rates, also called a Data Transfer-Premium fee (DT-Premium)." The developer guide's Pricing topic says the same thing in its own words: "for each accelerator that is provisioned in your account (whether it's enabled or disabled)".

The addresses keep running too. The developer guide's standard accelerator page says "The static IP addresses remain assigned to your accelerator for as long as it exists, even if you disable the accelerator and it no longer accepts or routes traffic." A disabled accelerator is therefore the full $25.20 a month, not a reduced rate and not zero.

The trap is procedural. The delete procedure says "On the console, disable the accelerator, and then you can delete it", and notes that if you have not disabled it, "Delete" is greyed out. Disabling is a mandatory waypoint on the way to deletion, and it looks exactly like an off switch. Anyone who disables an accelerator to stop traffic during an incident and never comes back is paying $302.40 a year for a resource the console lists as disabled. Only deleting stops the bill, and deleting is irreversible: you lose the static addresses.

One AWS page reads oddly against this. The Global Accelerator FAQ opens with "Like other AWS services, AWS Global Accelerator is a self-service, pay-per-use offering, requiring no long term commitments or minimum fees." A charge that accrues on a provisioned resource until it is deleted, whether or not a packet arrives, is a minimum fee by any working definition. Treat the pricing page as authoritative and the FAQ sentence as marketing.

What is the Global Accelerator DT-Premium rate by source region?

Between $0.007 and $0.105 per GB, depending on which region serves the request and which edge location the response goes to. The pricing page defines it as a premium, not a price: "an incremental charge over your standard Data Transfer rates". It also scopes it widely: "The DT-Premium fee for AWS Global Accelerator applies to all traffic that flows through AWS Global Accelerator, regardless of whether the destination is to the internet, to another AWS Region, or to an Availability Zone."

Here is the pricing page's table, reproduced, with the source region on the left and the destination edge location across the top.

The spread is 15 times, from $0.007 for Australia serving Australian edges to $0.105 for a US region serving them. It is not symmetric: US and Canada to Asia Pacific is $0.035, while Asia Pacific to US and Canada is $0.012, 2.92 times cheaper in the other direction. And the cheapest cell in any row is usually the one on its own diagonal, so an accelerator bought to serve a distant continent pays the top of the table for exactly the traffic it exists for.

Does Global Accelerator DT-Premium replace data transfer out, or add to it?

It adds. The pricing page is unambiguous: "The DT-Premium fee is in addition to normal EC2 Data Transfer Out fees charged for your application endpoints running in AWS Region(s)." Every gigabyte leaving a region through an accelerator is billed twice, once by EC2 and once by Global Accelerator.

The VPC pricing page states the standard rate in the prose of its NAT Gateway example: "With the Data Transfer Out to Internet rate set at $0.09 per GB". The EC2 pricing page renders its own data transfer table in JavaScript, so that prose sentence is the citable source, and $0.09 is a US rate on the first tier. On that basis, here is the per-gigabyte comparison AWS's pricing page does not draw for you, for an application in a US region.

Putting an accelerator in front of a US application adds between 16.7% and 116.7% to the per-gigabyte cost of serving it. Australian users more than double it. The general shape of the underlying charge is in AWS data transfer out pricing, and the regional variant that catches multi-AZ designs is in cross-AZ data transfer cost.

Now run AWS's own example through it. The pricing page describes 10,000 GB a month, 60% outbound, and concludes: "Your monthly DT-Premium charge is $110 and, including the $18 monthly fixed fee, your total AWS Global Accelerator monthly bill is $128." The arithmetic checks out, and the heading is honest: it is the total Global Accelerator bill. It is not the total bill for the traffic. The 6,000 billable gigabytes also cost $540 of standard data transfer out at $0.09, and the two addresses cost $7.20.

The number on AWS's page is 19.0% of the number on the invoice. DT-Premium, the thing the page is about, is 16.3% of it. That gap is hypothetical arithmetic on a US rate, but the structure is not hypothetical: the pricing page tells you the premium is additive and then never adds it.

How does Global Accelerator decide the dominant direction it bills?

Hourly, not monthly, and the distinction can cost you. The pricing page says "DT-Premium is calculated every hour on the dominant direction of your traffic" and "You will only be charged DT-Premium in the dominant data transfer direction."

Its example then reasons from a monthly ratio: 60% of the month's traffic is outbound, so "you are charged for 6,000 GB a month and not all 10,000 GB". That holds only if outbound wins in every single hour. Take a hypothetical month with a nightly ingest window: 600 hours at 9 GB out and 2 GB in, then 120 hours at 5 GB out and 25 GB in. That is 6,000 GB out and 4,200 GB in, 58.8% outbound overall, which looks like AWS's example. But the night hours bill inbound, so the charged total is 8,400 GB, not 6,000. At $0.015 per GB that is $126 rather than $90, 40% more than the monthly ratio predicts.

So a workload with a heavy upload window cannot be estimated from its monthly in/out split. You need the hourly shape, which is why an accelerator in front of a backup target behaves nothing like one in front of a web tier.

Are Global Accelerator custom routing accelerators billed differently?

AWS does not say they are, and does not say they are not. Neither the pricing page nor the developer guide's Pricing topic mentions custom routing at all. Both describe the charge as applying to "each accelerator that is provisioned", and the single rate card on the pricing page carries no second column. On the evidence AWS publishes, the same $0.025 hourly fee and the same DT-Premium table apply to both types. We are not going to assert a rate AWS has not printed.

What AWS does document is that the two types differ everywhere except the bill. The FAQ says "Custom routing accelerators support only VPC subnet endpoints, each containing one or more EC2 instances that are running your application". The custom routing guidelines add two restrictions with cost consequences: "Global Accelerator does not perform health checks for custom routing accelerators and does not failover to healthy endpoints", and "CloudFormation is not supported for custom routing accelerators." No CloudFormation means no stack deletion, which means the cleanest way an idle accelerator normally disappears does not exist for this type. The custom routing accelerator page confirms they get the same two static IPv4 addresses, so the $25.20 floor is the same.

Does Shield Advanced add a third per-GB charge to accelerator traffic?

Yes. The Shield pricing page says a Shield Advanced subscription "charges a monthly fee, plus a usage fee based on data transfer out from Amazon CloudFront, Elastic Load Balancing (ELB), Amazon Elastic Compute (EC2), and AWS Global Accelerator", and its table puts Global Accelerator at $0.025 per GB for the first 100 TB.

Stack that on a US region. Standard data transfer out at $0.090, DT-Premium at $0.015 for a nearby edge, Shield Advanced at $0.025: $0.130 per GB, 1.44 times the bare rate. Serving Australia from the same region, it is $0.220 per GB, 2.44 times. Three per-gigabyte meters on one gigabyte, billed under three service names. The subscription side of that, including when it pays for itself, is in what AWS Shield Advanced costs.

Is AWS Global Accelerator or CloudFront cheaper for the same workload?

CloudFront, by a wide margin, whenever CloudFront can do the job. The CloudFront pricing page now sells flat-rate plans: Free at $0 a month with 100 GB of data transfer, Pro at $15 with 50 TB, Business at $200, Premium at $1,000, all described as "a monthly price with no overage charges". It also removes the origin side of the bill: "Data transfer between CloudFront and your AWS origins is automatically waived when serving traffic through CloudFront."

That is the opposite billing shape. Global Accelerator's fixed fee alone, $18.00, is 1.2 times the entire CloudFront Pro plan. DT-Premium at its cheapest rate reaches the whole Pro monthly price after 1,000 GB. Running AWS's own 6,000 GB example through both, a US accelerator is about $655 against $15 for a Pro distribution that also covers the data transfer. CloudFront's page adds "There are no additional overage charges or usage calculations, even during traffic spikes or attacks", which Global Accelerator does not offer at any price. The plan detail is in CloudFront pricing.

That comparison is only valid where the two services overlap, and often they do not. The FAQ draws the line: "Global Accelerator is a good fit for non-HTTP use cases, such as gaming (UDP), IoT (MQTT), or Voice over IP, as well as for HTTP use cases that specifically require static IP addresses or deterministic, fast regional failover." UDP, a firewall allow-list needing fixed addresses, and sub-minute regional failover are things CloudFront does not do, and the honest test is whether you are buying one of those three. If you chose Global Accelerator for speed in front of an HTTP application, you are paying a per-gigabyte premium over a service that would have waived your origin transfer. If you need static anycast addresses in front of a TCP or UDP listener, price the accelerator alongside what the Network Load Balancer under it costs, because that stack carries its own hourly fee, its own capacity units and its own public addresses before a single byte moves.