VPS Tutorials
Why Outbound Traffic on Overseas VPS Is Expensive (2026): Understanding Bandwidth, Traffic Packages, and 95th Percentile Billing
Many people only look at CPU, memory, and disk when buying an overseas VPS, and only after going live do they realize that what balloons most easily isn't the monthly machine fee, but outbound traffic costs. This article breaks down bandwidth, traffic packages, outbound traffic, and 95th percentile billing, explaining why some platforms look cheap but end up being crushed by downloads, origin pulls, cross-region sync, and peak bandwidth settlement.
Many people buying an overseas VPS for the first time stare only at these numbers:
- 2 cores, 4G
- 80GB SSD
- 1Gbps bandwidth
- $5 per month
Looks great, right?
But once you actually start running a business, problems appear:
- As soon as a video site's traffic grows, the bill suddenly balloons
- As soon as object storage origin pulls increase, the cost exceeds the machine itself
- The bandwidth clearly says
1Gbps, so why are you still charged extra - Some vendors write "10TB traffic," some write "100Mbps unmetered," and some just throw out a "95th percentile"
That's when many people first realize: what's really hard to understand about overseas VPS isn't CPU and memory, but network billing.
This article is dedicated to clarifying the 3 most easily confused things:
- What bandwidth is
- What traffic packages / outbound traffic are
- What 95th percentile billing is actually calculating
Once you understand these three things, you'll see why some VPS look cheap but may end up being crushed by traffic fees.
1. The conclusion first: what's expensive usually isn't "network speed," but "data flowing out of your data center"
If you only remember one sentence, remember this:
The most likely hidden cost of an overseas VPS isn't peak bandwidth, but outbound traffic (egress) and the bandwidth settlement method.
First, separate the three terms:
| Concept | How you can understand it |
|---|---|
| Bandwidth | How thick the water pipe is, e.g. 100Mbps, 1Gbps |
| Traffic | How much water actually flows out of this pipe in a month, e.g. 2TB, 10TB |
| Outbound traffic | Data sent from your server to public internet users |
Why is outbound traffic expensive? Because for cloud vendors, what really keeps burning money is often not how fast you can run, but how much data you send outward over the long term.
Data uploaded to the server (inbound) is barely charged by many platforms; what's really easy to charge for is:
- Users downloading files
- Images / videos being accessed
- Object storage origin pulls
- Cross-region sync
- Large API response output
So the billing trap usually isn't at the moment of "buying the machine," but in "continuous outbound transmission."
2. Bandwidth and traffic packages are not the same thing; many people get it wrong at the first step
This is the most common misconception.
For example, a vendor writes:
100Mbps1Gbps10Gbps
This refers to peak throughput capability, i.e. how high your network ceiling is.
You can think of it as the thickness of a water pipe:
- The thicker the pipe, the more water can pass through instantly
- But it doesn't mean you'll necessarily use a lot of water in a month
For example, a vendor writes:
2TB transfer10TB outbound traffic20TB included
This refers to the monthly total.
You can think of it as a water meter quota:
- The pipe can be very thick
- But you're only allowed to release 2TB in total per month
Because many VPS pages write both at the same time:
1Gbpsbandwidth3TBmonthly traffic
Beginners easily mistake 1Gbps as meaning "run as much as you want." Actually it doesn't.
The real situation may be:
- You can indeed run fast instantaneously
- But once the monthly cumulative outbound exceeds 3TB, overage fees start immediately
Because what cloud vendors sell you isn't "a network port," but a whole set of global network capabilities:
- Data center egress bandwidth
- Carrier interconnection
- Cross-border links
- Peak-period capacity reservation
- Routing quality and stability
Especially in the following scenarios, outbound costs are usually higher:
- Many international exits, wide regional coverage
- Heavy intercontinental distribution
- Obvious user-download-type business
- High bandwidth peaks, long peak duration
- Very large cloud vendors charging by fine-grained metering
Simply put, CPU and disk are more like "you using them inside your own data center," while outbound traffic is more like "you occupying the cloud vendor's highway for outward delivery over the long term."
The bigger the cloud vendor, the more regions, the finer the billing, the easier this part of the cost is split out and charged separately.
This is exactly one of the cost differences most worth being wary of in 2026.
According to externally compiled 2026 outbound traffic data, the market can roughly be divided into several tiers:
Some budget providers, independent cloud vendors, or platforms that bundle traffic directly into packages make you feel like there's "almost no separate egress charge."
These platforms usually have these characteristics:
- Packages directly include a relatively large monthly traffic allowance
- By default they don't emphasize fine-grained egress billing
- More suitable for budget-sensitive sites, download-type sites, experimental environments
This is a common model for many VPS and some cloud platforms.
Typical wording is:
- Includes
2TB / 5TB / 10TBoutbound traffic - After exceeding, billed at
每 GB / 每 TB
These platforms are best for:
- Predictable traffic
- Non-exaggerated peaks
- Businesses that can accept "overage costs extra, but normally won't explode"
This is where many people get burned on AWS, Azure, and GCP.
The problem isn't necessarily that they're "unreasonably expensive," but that:
- Billing granularity is fine
- Unit prices are usually significantly higher than budget VPS
- Once downloads, origin pulls, and cross-region sync scale up, bills rise quickly
That is to say:
Some platforms sell "cheap compute, network billed separately"; some platforms sell "network bundled first, overage discussed later."
If you don't distinguish these two logics, you're most likely to get crushed in content distribution, backup, and file download businesses.
There's no fixed answer to this question; it depends on what your business is.
For example:
- Download sites
- Video distribution
- Mirror sites
- Large file sync
For these scenarios, you should look first at:
- Total traffic allowance
- Overage unit price
- Whether there are bandwidth peak limits or shaping
Sometimes 100Mbps 不限流量 is actually more stable, because you know where your bill ceiling is.
For example:
- Campaign pages
- Image sites
- Occasional downloads
- SaaS API burst traffic
For these scenarios, 1Gbps + 10TB is sometimes more cost-effective:
- High bandwidth ceiling normally
- As long as the monthly total doesn't blow up, the experience is better
- Stable continuous business: look first at traffic and total cost
- Burst-peak business: look first at peak bandwidth and overage rules
Don't get led by the number 1Gbps, because it's often just a "ceiling," not a "monthly cost guarantee."
95th percentile billing is most easily misunderstood, because it doesn't calculate "how much traffic you transmitted in total in a month," but "how high your bandwidth usage was for most of the month."
The carrier or provider samples your bandwidth usage at fixed intervals, e.g. once every 5 minutes.
Over a month you get many sample points, then:
- Sort these samples from high to low
- Remove the highest 5%
- The highest remaining value becomes your billed bandwidth
This is the core logic of 95th percentile billing.
Because what it wants to express is:
- Occasional spikes of one or two peaks don't necessarily get you penalized
- But if you occupy egress bandwidth high for a long time, you should pay according to this high water mark
This method appears more often in:
- Dedicated servers / racks / colocation networks
- High-bandwidth businesses
- Carrier settlement
- Some enterprise-grade network services
| Model | What it mainly looks at |
|---|---|
| Traffic package | How much monthly total was used |
| 95th percentile billing | How much long-term peak bandwidth was occupied |
So you may encounter this situation:
- Total traffic doesn't look particularly large
- But because you stably output high bandwidth every evening peak
- The final 95th percentile bill is still very high
This is also why download sites, video distribution, proxy egress, and mirror sync businesses need to pay special attention to 95th percentile billing.
These are high-risk areas:
Every user click is your outbound burning money.
If the caching strategy is poor and CDN hit rate is low, origin pull traffic will be astonishing.
Database backups, cross-region object storage replication, and disaster recovery sync all continuously eat outbound allowance.
Many people only stare at request counts, ignoring that the returned content itself also goes through outbound.
These businesses are naturally "throughput machines"; if you don't watch bandwidth settlement, you'll almost certainly step in a trap.
8. Why do many people's bills spiral out of control not because traffic is large, but because "the architecture isn't layered"
Many sites could have been very cheap, but end up with outbound costs forced high because the architecture is too crude.
The 4 most common reasons:
Images, JS, CSS, and attachments all sent from the VPS itself, which means letting the most expensive network layer bear all downloads.
This causes a large number of user requests to hit the origin directly, and outbound costs naturally rise.
Backups are right in themselves, but without compression, deduplication, and incrementals, they only raise egress costs.
Some platforms are suitable for running compute, some for running content distribution. Choose the wrong scenario and the more you do, the more you lose.
If your business is naturally high-download, high-distribution, prioritize:
- Platforms with large built-in traffic packages
- Platforms with cheap egress
- Platforms with lower combined object storage / CDN costs
This is the most direct and most common way to reduce costs.
Especially:
- Images
- Video covers
- CSS / JS
- Installer packages
- Download files
Whatever can go into object storage shouldn't stay on the application machine's local disk served directly outward.
Whether logs, backups, or static distribution, a higher compression ratio means lower outbound.
Especially databases, media files, and object storage archives; without incrementals it's very wasteful.
If you use bandwidth billing or 95th percentile billing, peak control matters more than total control.
This one is the most critical.
When comparing VPS, put at least the following items in the same table:
- Monthly fee
- Renewal price
- Included traffic allowance
- Overage unit price
- Whether 95th percentile billing applies
- CDN / object storage pairing cost
Comparing only CPU and memory will almost certainly lead you wrong.
Before ordering, check at least these 6 items:
- Does it say bandwidth or traffic package?
- Are outbound and inbound billed separately?
- How is overage charged? By GB, by TB, or throttled?
- Are there extra fees for cross-region or cross-availability-zone?
- Does it involve 95th percentile billing or other peak bandwidth settlement?
- If you attach CDN / object storage, who bears the origin pull cost?
If there are 2 items here you don't understand, don't rush to order.
Because 1Gbps is only peak bandwidth; it doesn't mean monthly transmission is unlimited, nor that outbound is free.
Not necessarily. For businesses with many short-term peaks and low long-term occupancy, 95th percentile billing may be friendlier than a fixed large-traffic package; but for businesses with stable high-bandwidth output over the long term, it may be very expensive.
Not exactly the same. Some platforms bundle large traffic directly into packages, some charge finely by egress, some barely charge separately for object storage / specific products. The key is not to assume, but to read the terms.
Yes, but don't over-worry. Small blogs usually aren't a big problem early on; once you start having lots of images, attachments, backups, and CDN origin pulls, outbound costs are worth calculating seriously.
Why do so many people find overseas VPS outbound traffic "expensive and complicated"? Not because the network itself is mystical, but because you're facing three completely different billing logics at the same time:
- Bandwidth: how fast you can run
- Traffic package: how much you send out in total per month
- 95th percentile billing: how high an egress resource you occupy most of the time
What really makes your bill spiral out of control is often not the VPS itself, but that you put high-distribution, high-download, high-origin-pull businesses directly on an architecture with expensive egress and no CDN or layering.
So when choosing a VPS, don't just look at CPU, memory, and disk. What will really teach you a lesson later is often that most inconspicuous small print: Outbound Traffic / Data Transfer / 95th Percentile Billing.