Why is my VPN slow?40% to 90% of the line survived, depending on the city

"My VPN is slow" is the most common complaint about VPNs and the least useful, because it does not say slow compared with what. We measured the comparison: the same line, in the same session, with and without the tunnel, across 14 cities.

The headline result is that the spread between cities is much wider than the spread between providers. In London about 90% of the line survived on both services we tested. In Tokyo it was roughly half. Same providers, same method, same week.

• 40% to 90% of the raw line survived, depending on the city
• The gap between cities is larger than the gap between providers
• Tokyo kept about half; London kept about 90%
• We only name a winner where we have 5+ runs and a 10% gap — 4 of 14 cities
• Distance and the exit route explain most of the difference

The five causes, in the order they actually occur

Distance to the server. This is the big one and the one people control without realising it. Every packet goes to the server before it goes anywhere else, so connecting to a server on another continent adds that round trip to everything. If you are in Japan and connected to a European exit for no reason, that is your answer.

Server load. A popular location at a popular hour is shared with more people. This is why the same server can be fine in the morning and poor in the evening, and why providers with more locations per country have an advantage that does not show up in a country count.

The protocol. WireGuard is generally faster than OpenVPN on the same route, and obfuscated modes cost more again because the traffic is being reshaped. If you turned obfuscation on for a trip and left it on, that is a standing cost you are paying for nothing.

Your own line. A tunnel cannot exceed what you started with, and it makes an already-marginal connection worse rather than adding to it. Measure the bare line first — this is the step almost every troubleshooting guide skips, and it is why we measure a baseline in every city we can.

The network you are on. Some hotel and campus networks shape or throttle traffic they identify as a VPN. The symptom is distinctive: fine for a minute, then collapsing, rather than uniformly slow.

What our numbers say about blame

Read the table again with one question in mind: how much of this is the provider's doing? In London the two services were within 1 Mbps of each other. In New York they were within 8%. The only cities where we could name a winner at all were Tokyo and Los Angeles, and even there the gap was 11% and 17% — real, but small next to the 50-point spread between Tokyo and London on the same service.

That is the practical conclusion. Changing provider to fix a slow connection is usually the wrong move, and changing server — or checking which one you are actually on — is usually the right one.

What to try, in order

A short list, ordered by how often it works rather than by how sophisticated it is:

• Check which country you are connected to. Apps stay where you last put them. This is the single most common cause and takes five seconds to rule out.
• Move to the nearest location that serves your purpose. If you need a Japanese address, take the Japanese city closest to you rather than the first in the list.
• Try another city in the same country if your provider has one. Four of the seven we track publish two or more Japanese cities; three publish exactly one.
• Switch protocol to WireGuard if the app offers it and you are not on it.
• Turn obfuscation off if you are not in a country that needs it.
• Measure the bare line before concluding anything. Half the reports we have seen turn out to be the underlying connection.

Why we publish the sample size

Most speed comparisons give you one number per provider and no indication of how it was produced. Ours gives the number of paired runs per city, because with two runs a 10% difference means nothing and with nine it starts to mean something.

We set the bar at five paired runs and a 10% gap before we will say one service beat another in a city. Four of our fourteen cities clear it. In the other ten we publish the numbers and say the sample is too small — which is less satisfying than a ranking and more honest than inventing one.

How much speed actually survives the tunnel

Every VPN site says a tunnel costs you speed. Almost none of them say how much. We measured it by hand rather than repeating a claim: 14 cities, 46 paired runs, comparing NordVPN and Surfshark against the same raw line in the same session. In the 11 cities where we also measured the bare line, the share that survived the tunnel ranged from 40% to 90%.

Measured by VPN LIFE. Raw-line figures are rounded to the nearest 10 Mbps. Percentages are the share of the raw line that survived.

• 14 cities, 46 paired runs, measured by hand
• 11 of those cities also have the bare line measured
• 40% to 90% of the raw line survived
• We only name a winner in 4 cities — the rest have too few runs

The spread between cities is larger than the spread between providers. In London roughly nine tenths of the line survived on both services; in Tokyo it was closer to half. That is worth knowing before you blame the provider: "this VPN is slow" is usually "this route is slow".

We only call a winner in a city when we have five or more paired runs there and the gap is at least 10%. That threshold is met in 4 of the 14 cities — Tokyo and Los Angeles, where NordVPN averaged ahead, and London and New York, where the two were level. In the remaining ten cities we publish the numbers and say the sample is too small, rather than ranking on one run.

The apps behind our measurements

The providers, and what we can verify about each

We rank on what we can check. Each of the five below publishes a location in this use; the tables are our own counts and the providers' own published terms, not a summary of their marketing.

ExpressVPN: the widest spread of cities

ExpressVPN

ExpressVPN publishes more multi-city countries than anyone else in our set — 11 countries where you can pick between two or more locations, including 56 in the United States, 7 in the United Kingdom, 6 in Australia and 4 in Japan. If your problem is a blocked address rather than a missing country, that spread is the thing that gets you out of it. It is the most expensive option here, and it is the one we would pick for streaming.

Go to ExpressVPN

NordVPN: the largest country list, and the one we measured

NordVPN

NordVPN covers more countries than anything else we track. It is also one of the two services we measured ourselves, so the retention figures above are its actual numbers rather than a claim. It operates from Panama, has been audited by Deloitte, allows 10 simultaneous connections, and lists obfuscated servers and a kill switch among its features — the two things that matter where VPN traffic is filtered.

Go to NordVPN

Surfshark: unlimited devices, and level with NordVPN in half our tests

Surfshark

Surfshark is the other service in our own measurements. In London and New York it was level with NordVPN; in Tokyo and Los Angeles it trailed by 11% and 17%. It places no limit on simultaneous connections, which makes it the sensible choice for a household rather than a person. Audited by Deloitte, operating from the Netherlands, with Camouflage Mode for networks that filter VPN traffic.

Go to Surfshark

CyberGhost: the longest money-back window, with a catch

CyberGhost

CyberGhost's 45-day money-back guarantee is the longest here, and it is the reason it appears on most lists. Read the condition: 45 days applies to the six-month and two-year plans, and the monthly plan gets 14 days. It publishes 10 multi-city countries, with 11 locations in the United States, but only a single city in Japan — so it is a weaker choice if Japanese services are your target.

Go to CyberGhost

Private Internet Access: unlimited connections, US jurisdiction

Private Internet Access

PIA places no limit on devices and publishes an Advanced Kill Switch, obfuscation, multi-hop and port forwarding. Its no-logs claim has been audited by Deloitte Audit Romania. The trade-off is jurisdiction: it operates from the United States, which some readers will rule out on principle. We list it because the rest of the record is strong and because it is one of only four providers with two or more cities in Japan.

Go to Private Internet Access

Frequently asked questions

How much speed loss is normal?

On the routes we measured, anywhere from 10% to 60%. If you are losing half your line on a long route, that is within what we observed on paid services in good conditions. If you are losing 90%, something else is wrong — usually distance or the underlying connection.

Does a faster plan from my ISP help?

Only up to a point. The tunnel takes a proportion rather than a fixed amount, so a faster line gives you a faster tunnel — but if the bottleneck is distance or server load, a bigger pipe into the same congested route changes little.

Is WireGuard actually faster?

On the same route, generally yes, and the difference is largest on high-latency connections. We measured providers rather than protocols, so we are reporting the general finding rather than our own — and we say which is which.

Can I use a free VPN for this?

Usually not, and the reason is measurable rather than moral. Even on paid services we measured, between 10% and 60% of the raw line disappears into the tunnel — in Tokyo a 780 Mbps line came out at roughly half. Free tiers add a data cap and a much smaller server list on top of that. They also have to earn money somewhere, and where the product is free the revenue usually comes from the traffic itself.

Conclusion

If you take one number away: 40% to 90%, and the city matters more than the brand. Check which server you are on before you check anything else.

The raw measurements, including the cities where the sample is too small to conclude anything, are in our open data.

Every figure on this page comes from our own daily collection or our own measurements. The files are published as open data under CC BY 4.0, and the method — including what the numbers do not prove — is on how we test.