What does a terabyte of VPS bandwidth actually cost?
Between $5 and $150 per TB, and at two providers nothing at all. Akamai/Linode $5.00 (core), Hetzner Singapore $8.49, DigitalOcean $9.31, Vultr $10.00, AWS Lightsail $90–150 by region. Hostinger and Contabo bill no overage — they throttle instead. Hetzner does not publish a per-TB rate for Europe or the US at all.
Key finding Contabo’s throttled port speeds are exactly its traffic packages expressed as a rate limit: 100 Mbit/s sustained for 30 days is 32.4 TB, against a 32 TB package. The same holds for all six packages.
Prices verified 29 August 2026 against each vendor’s own pricing page, linked in full below.
Every VPS pricing page prints an included-traffic figure. 20 TB. 4 TB. 2 TB. “Unlimited.” Those numbers get placed next to each other in every comparison article on the internet as if they measure the same thing.
They do not. What matters is what happens at 20.1 TB — and that ranges from $5 a terabyte to $150 a terabyte to no charge at all.
| Cheapest terabyte billed on this page | $5.00 |
| Most expensive | $150.00 |
| Providers that bill nothing for overage | 2 of 7 |
This page computes the marginal price of a terabyte at each provider, sets out the three incompatible ways the “included” allowance is pooled, and names the plans that ship with zero bundled transfer.
Every figure verified 29 August 2026 against the provider’s own pricing page or documentation, all linked below. Where a provider does not publish a number, this page says so instead of guessing.
The table: what one extra terabyte costs
Converted to a common unit of one decimal terabyte (1,000 GB), sorted cheapest first.
| $/TB | Provider | Published rate | Notes |
|---|---|---|---|
| $0.00 | Hostinger VPS | no overage charge | speed cut to 10 Mbps for the rest of the month |
| $0.00 | Contabo | no overage charge | throttled only after 10 consecutive days over package |
| $5.00 | Akamai / Linode (core) | $0.005 per GB | except Jakarta and São Paulo |
| $7.00 | Akamai / Linode (São Paulo) | $0.007 per GB | |
| $8.49 | Hetzner Cloud (ap-southeast) | $8.49 per TB | Singapore zone only |
| $9.31 | DigitalOcean | $0.01 per GiB | billed in GiB, so a decimal TB is 931 GiB |
| $10.00 | Vultr | $0.01 per GB | |
| $10.00 | Akamai / Linode (distributed regions) | $0.01 per GB | |
| $15.00 | Akamai / Linode (Jakarta) | $0.015 per GB | |
| $90.00 | AWS Lightsail (US, EU incl. Paris, Canada, Hong Kong) | $0.09 per GB | |
| $108–114 | AWS Lightsail (Malaysia, Sydney, Tokyo) | $0.108–0.114 per GB | |
| $120 | AWS Lightsail (Singapore, Spain) | $0.12 per GB | |
| $126 | AWS Lightsail (Seoul) | $0.126 per GB | |
| $150 | AWS Lightsail (São Paulo) | $0.15 per GB | |
| — | Hetzner Cloud (eu-central, us-east, us-west) | not published | see below |
Sources: Akamai network transfer costs, DigitalOcean bandwidth billing, Vultr overage rate, AWS Lightsail pricing, Hostinger bandwidth limits, Contabo traffic rules, Hetzner Singapore.
Between the two ends of the billed range there is a 30× difference for an identical terabyte. Restricted to US and European locations only — where Akamai charges $5 and Lightsail’s Spain region charges $120 — it is still 24×.
The equivalent spread in the price of memory is eleven-fold. Bandwidth is the wider variable, and almost nobody prices it in when choosing a host.
The included allowance is three different products
This is what makes cross-provider comparison of the headline TB figure meaningless. The allowance is pooled by mutually incompatible rules.
DigitalOcean — pooled at the team level. Their documentation is explicit: transfer “is pooled cumulatively across all Droplets at the team level, not individually per Droplet.” Ten small Droplets share one large pot. It does not roll over between months and cannot move between teams.
Vultr — pooled per account. A flat 2 TB free per account per month, plus each instance’s own accrued allowance, added into a single cap. Vultr prorates against a 672-hour month, so an instance destroyed halfway through contributes half its allowance.
Hetzner — per server, with no documented pooling. Each server carries its own allowance. Hetzner’s billing documentation describes no mechanism for sharing one server’s allowance with another, so twenty servers with 20 TB each should not be assumed to be a 400 TB pot you can spend on one machine. (Hetzner does not state this either way; the absence is the finding.)
AWS Lightsail — pooled, but only within one bundle type in one Region. AWS states the allowance “is aggregated for instances of the same bundle (bundleId) in a Region.” Three $5 instances in Seoul pool together. A $5 and a $12 instance in the same Region do not.
Akamai / Linode — pooled globally, with holes. Most core data centres share one global pool. Newer data centres may be tracked in a region-specific pool, and “Network transfer isn’t pooled in distributed compute regions” at all.
The practical consequence: if you run several small instances, DigitalOcean’s and Akamai’s modest per-plan numbers are worth far more than they look, and Hetzner’s large per-server numbers are worth less than they look. A single row in a comparison table cannot express this, which is why every comparison table gets it wrong.
The plans that include no transfer at all
Akamai’s newer plan generations ship with no bundled outbound transfer whatsoever. The list is verbatim from their own documentation: “NVIDIA RTX PRO 6000™ Blackwell Server Edition GPU plans, NVIDIA RTX 4000™ GPU plans, NVIDIA Quadro RTX 6000™ plans, Accelerated plans, G8 Dedicated CPU plans, or for Linodes in distributed compute regions.”
Every outbound gigabyte on those plans is billed from the first one. (Inbound stays free — Akamai lists “all inbound network transfer” as free and unmetered.) On a GPU instance serving model inference, that is not a rounding error; it is a second bill.
Akamai does date the change, in a changelog entry of 10 December 2025: “G8 Dedicated plans differ from G6 and G7 Dedicated plans in that they use our newest network transfer model of usage-based billing, which allows you to pay for only the bandwidth you use.” The earlier generations are described elsewhere as including “bundled transfer with allowances that scale based on plan size.”
So the trap is specific and dateable: if you sized a G8 Dedicated CPU plan on the assumption that it behaves like the G6 or G7 you ran before, the transfer allowance is gone and the wording for it is “pay for only the bandwidth you use.”
The number Hetzner does not publish
Hetzner’s pricing pages contain the sentence “For each additional TB at EU-Central locations we charge” — and then nothing. The figure is injected by JavaScript and is absent from the served HTML on both the English and German pages, on both the cost-optimized and regular-performance lines. The billing FAQ explains the mechanism — “we will bill you for the over-usage in blocks of 100MB” — without ever stating a rate.
The only per-TB figure Hetzner hardcodes anywhere on its own site is on the Singapore landing page: $8.49 per TB for the ap-southeast zone.
So this page will not tell you what Hetzner charges per TB in Europe, because Hetzner does not tell you either without logging into the Console. Every article that quotes a euro figure for it took that figure from another article.
Hetzner’s documentation does confirm the zones are priced differently: “Prices are the same for all locations in the network zone eu-central. We charge different prices for the network zones us-east, us-west, and ap-southeast.”
Hetzner’s other 20× problem
The same server costs roughly the same money and includes twenty times less traffic depending on where you click.
| Location | Included traffic |
|---|---|
| EU (eu-central) | 20 TB on the cost-optimized line |
| US (us-east, us-west) | at least 1 TB |
| Singapore (ap-southeast) | 0.5 TB to 5 TB, depending on plan |
Sources: Hetzner Cloud cost-optimized, Hetzner Singapore.
Read the Singapore range again: the floor is 0.5 TB. Half a terabyte. Hetzner’s own words are that included traffic “depends on the cloud plan and varies from 0.5 TB to 5 TB.” A Hetzner server in Europe can include forty times the traffic of the smallest Singapore plan.
This is the reason “Hetzner is the cheapest” is only true in Europe, and it has nothing to do with the June 2026 price increases — the traffic split predates them, and Hetzner’s own price-adjustment notice covers monthly, hourly and setup fees without listing any traffic change.
Worth knowing, and free: Hetzner bills only outgoing traffic. “Incoming and internal traffic is free,” and internal includes other Hetzner Cloud servers over a private Network and servers in the same network zone reached by public IP. Traffic between two zones — Europe to the US — “will be billed just as any other normal internet traffic.”
The two providers that never send you a bandwidth bill
Both replace the bill with a speed limit. The interesting part is that both speed limits can be converted into a hard monthly ceiling.
Hostinger publishes plan allowances of 4 TB (KVM 1) through 32 TB (KVM 8) and then does something unusual with them: there is no overage charge. Exceed the allowance and “the connection speed will be reduced to 10 Mbps for the rest of the month,” with services remaining operational, resetting at the start of the next month.
A permanently saturated 10 Mbps link moves about 3.2 TB in a 30-day month. So the practical shape of a Hostinger overage is not a bill — it is a ceiling of roughly your allowance plus, at most, 3.2 TB of very slow traffic. For a website, that is a bad month. For a download host, it is an outage.
Contabo is the one most comparison articles get wrong, because Contabo’s marketing page and its knowledge base say different things. The VPS product page advertises “a generous 32TB of outgoing” traffic and, further down, “Outgoing Traffic: Unlimited – fair usage policy applies based on average server workloads.”
The knowledge base is where the actual rule lives, and it is neither unlimited nor discretionary. Contabo publishes traffic packages of 10, 32, 81, 162, 243 and 324 TB, and a deterministic trigger: “Only if the outgoing bandwidth exceeds your bandwidth package for 10 days in a row, the port speed of your instance will be throttled.” The resulting speed is published per package:
| Traffic package | Throttled port speed | Sustained ceiling after throttling |
|---|---|---|
| 10 TB and 32 TB | 100 Mbit/s | ~32.4 TB/month |
| 81 TB | 250 Mbit/s | ~81 TB/month |
| 162 TB | 500 Mbit/s | ~162 TB/month |
| 243 TB | 750 Mbit/s | ~243 TB/month |
| 324 TB | 1 Gbit/s | ~324 TB/month |
Source: Contabo, Rules for traffic. The right-hand column is computed here, not published by Contabo.
That last column is the finding. The throttled speed for each package is set almost exactly at the package size — 100 Mbit/s sustained for a 30-day month is 32.4 TB, against a 32 TB package. Contabo’s throttle is not a punishment; it is the package enforced as a rate limit rather than a bill. And you have to exceed it for ten consecutive days before it applies at all.
For a workload that runs at or under its package, Contabo is genuinely the cheapest bandwidth here. For one that spikes past it, the consequence is bounded and knowable, which is more than “unlimited” ever promised.
AWS Lightsail counts your inbound traffic
Every other provider that addresses the question states that ingress is free and unmetered. (Hostinger’s documentation does not address it.) Lightsail is the exception, and the wording is precise: “Both data transfer IN and data transfer OUT of your instance count toward your data transfer allowance” — but “You will not be charged for the excess data transfer IN to your Lightsail instance.”
So inbound is not billed, but it does consume the allowance that protects you from the $90/TB outbound rate. For an upload-heavy application, the included transfer therefore runs out sooner than the pricing page implies — how much sooner depends entirely on your inbound-to-outbound ratio, which the pricing page cannot know and this page will not guess.
One Lightsail detail worth flagging before you budget: the pricing page and the userguide FAQ publish different regional overage rates, and the FAQ is higher in at least four regions.
| Region | Pricing page | Userguide FAQ |
|---|---|---|
| Sydney | $0.114/GB | $0.17/GB |
| Tokyo | $0.114/GB | $0.14/GB |
| Mumbai | $0.1093/GB | $0.13/GB |
| Seoul | $0.126/GB | $0.13/GB |
Both pages are live, official, and undated. The table at the top of this article uses the lower pricing-page figures throughout. If you are sizing a budget in one of these four regions, confirm the rate in the Lightsail console before committing — and assume the higher number until you have.
For contrast, standard Amazon EC2 gives every account “100 GB of free data transfer out to the internet free each month, aggregated across all AWS Services and Regions (except China and GovCloud).” Beyond that, EC2’s tiered per-GB rates are rendered by a JavaScript region selector and could not be extracted from the served page, so this comparison does not quote them.
Two worked examples
A busy content site: 5 TB per month.
Almost nobody pays anything. Hetzner in the EU (20 TB), Hostinger KVM 2 (8 TB), Contabo on any package, and most mid-tier Akamai and DigitalOcean plans absorb this inside the allowance. The exception is Lightsail: the $12 bundle includes 3 TB, so the extra 2 TB costs $180 — fifteen times the price of the server. At this volume, bandwidth is the entire decision only if you chose AWS.
A media or download workload: 50 TB per month.
Now the marginal rate is the whole bill. Included allowances vary by plan, so this prices the same 45 TB of overage at each published rate rather than assuming a plan:
| Provider | Published rate | 45 TB of overage |
|---|---|---|
| Akamai / Linode (core) | $5.00 / TB | $225 |
| Hetzner (ap-southeast) | $8.49 / TB | $382 |
| DigitalOcean | $9.31 / TB | $419 |
| Vultr | $10.00 / TB | $450 |
| Hostinger | no charge | $0, throttled to 10 Mbps |
| Contabo | no charge | $0 if inside package; throttled after 10 days over |
| AWS Lightsail (US/EU) | $90.00 / TB | $4,050 |
| AWS Lightsail (São Paulo) | $150.00 / TB | $6,750 |
The server itself costs $10–50 in every row. At 50 TB, the machine is a rounding error and the transfer policy is the product.
So what should you actually do
If your traffic is under a terabyte or two a month, none of this matters and you should choose on price per GB of RAM instead. That is most people.
If you are in Europe and predictable, Hetzner’s 20 TB per server on the cost-optimized line is the most generous allowance here by a wide margin, and the unpublished overage rate is unlikely to ever apply to you.
If you are outside Europe, Hetzner’s advantage largely disappears — 1 TB in the US, as little as 0.5 TB in Singapore — and Akamai’s $0.005/GB becomes the cheapest published marginal rate in this comparison.
If you run many small instances, pooling is worth more than the headline number. DigitalOcean’s team-level pool and Akamai’s global pool both beat a larger per-server figure you cannot share.
If you move genuinely large volumes, Contabo’s published package-plus-throttle rule is the only policy here that caps your bandwidth cost at zero with a knowable consequence. Read the ten-consecutive-days trigger before you rely on it.
If you are on AWS Lightsail and moving real traffic, this is the clearest-cut case on the page. At $90 to $150 per TB, Lightsail’s outbound rate is between nine and thirty times the rate of the four providers here that publish a per-TB price — and higher still against the two that publish none.
Method and limits
Prices were read from each provider’s own pricing page or documentation on 29 August 2026 and converted to a single decimal terabyte (1,000 GB) for comparison. DigitalOcean bills in GiB; that row is converted accordingly and is therefore cheaper per decimal TB than its $0.01 headline suggests. The “sustained ceiling” column in the Contabo table is computed from the published port speeds, not published by Contabo.
Four things could not be verified on an official page and are deliberately absent rather than estimated: Hetzner’s per-TB rate for eu-central, us-east and us-west; Amazon EC2’s tiered data-transfer-out rates; whether Hostinger counts inbound traffic against the allowance; and which of the two conflicting AWS Lightsail rate tables is current.
Bandwidth pricing changes less often than server pricing, but it changes without announcement. This page will be re-verified when any provider on it publishes a change.
Last updated: 29 August 2026.