Why megabits tell you almost nothing about working remotely
A fat pipe with a long, unstable round trip is a worse working day than a thin one that is close and steady. Almost nobody measures the thing that breaks.
By Adam Simmons-Spaans · Updated 2026-09-23
What actually breaks a call
Bandwidth is the number every site quotes because it is the number every speed test leads with. But a video call is not a download: it is a continuous exchange of small packets, and what ruins it is delay and the variation in that delay. Beyond about 150 ms of latency people start talking over one another; jitter above roughly 30 ms is what produces the robotic artefacts.
Throughput barely enters into it. A call needs a few megabits. Almost everywhere clears that bar, which is precisely why ranking on megabits separates nothing.
Measured, the two rankings disagree
China measures around 31.4 Mbps down with 207.4 ms of latency. On a bandwidth table that reads as merely slow. In practice it is the hardest place on this list to hold a meeting from.
Singapore is the mirror image: about 198.4 Mbps down, which looks excellent, with 63.2 ms of latency, which is middling. It reads better on paper than it works on a call.
These are aggregated from real connections over a rolling window rather than self-reported, and a country with too little data is left out rather than shown as zero.
Where the number comes from matters too
Most published connectivity figures are self-reported speed tests, which carry an obvious selection problem: people run a speed test when they suspect something is wrong, or when they have just bought a faster line. Neither is a random sample of a country's connections.
Aggregating from real traffic avoids that particular bias and introduces others — it reflects the networks actually being used to reach a given service, which skews towards whoever uses it. No measurement here is neutral, and a figure without a stated method is not worth arguing with.
The practical consequence is to distrust precision. A country reported at 190 Mbps and one at 210 Mbps are the same country for your purposes; a country at 30 Mbps with 200 ms of latency is a different proposition entirely, and that is a gap wide enough to survive any methodological quibble.
How to read a connectivity claim
If a guide quotes one number for a country's internet, it is quoting bandwidth, and it is not answering your question. Ask for latency and jitter, and treat any figure without them as a rough signal about infrastructure rather than about your working day.
It is also worth separating the last mile from the backbone. How fast your connection reaches the local network and how far your traffic then has to travel to reach the servers you use are different problems with different fixes. A country can have excellent domestic infrastructure and still be a poor place to work from if every service you rely on is hosted eight thousand kilometres away, because distance sets a floor on latency that no amount of local fibre can lift.
That distinction has a practical consequence when choosing where to go: the relevant question is not how good the internet is, but how far you are from the things you connect to. Someone whose whole working life runs through servers on the US east coast will find western Europe workable and south-east Asia hard, and the local connection quality is nearly irrelevant to that verdict.
The last thing to check is variability rather than the headline. A connection that averages well and drops for ten minutes twice a day is worse for a job built on meetings than a slower one that never wavers, and an average by construction hides exactly that. If a place is being considered seriously, the only real test is a week of actually working from it.
Compare measured connection quality
Every figure above is computed from published data and recomputed on each build. Run it against your own numbers rather than taking the example: compare measured connection quality .
About the author
Adam Simmons-Spaans, Founder, Geo-Parity. Adam builds Geo-Parity single-handedly: the engines, the datasets and the fetchers that refresh them. Every figure on the site comes from a published source — World Bank, IMF, OECD, Eurostat, WHO, HUD, Cloudflare Radar — and where a number is modelled rather than measured, the page says so. More about the site.