The Fallacy of Internet Speed

You don't need a Ferrari to buy milk

Say you need a car for the weekly grocery run. The store is six miles away, the speed limit is 45. So you buy a Ferrari, on the grounds that it'll do 200 mph.

The Volvo next to it is more comfortable, far easier to drive, copes better with potholes and snow, and was designed first and foremost to keep you safe. But its top speed is lower, so clearly it's inferior. Right?

Obviously ridiculous, yet this is almost exactly how we all buy internet. We compare the speed and the monthly bill, pick the biggest number we can justify, and find out about everything else during an important video call. That isn't because we're daft. The market has spent years teaching us that a bigger number means better internet - that we NEED the Ferrari.

Why internet is sold by the megabit

Speed tiers make commercial sense. Providers have networks to build and capacity to pay for, and faster plans do cost a bit more to deliver. But speed is also just the easiest way to slice internet into a menu. The same line into your house can be sold at several speeds, each with its own price, and a bigger number is an easy upsell.

Now, some people do need more bandwidth, and they should be able to buy it. The trouble comes when the entire conversation shrinks to two numbers: megabits-per-second (Mbps) and dollars.

But how often does the connection falter? How does it behave at 8 pm when the whole street is on Netflix? How stable is latency? Who fixes it when it stops working? What happens when someone crashes their Ferrari into your fiber junction and takes service out for 3 days? Those things are very hard to promise and harder to fit on a billboard, so they rarely make the sales pitch. They deserve at least as much attention as whether the premium plan comes with another 200 Mbps and a free streaming trial you’ll forget to cancel.

A speed test is a drag race

A speed test does something your household almost never does. It opens a handful of connections at once and floods your line with as much data as it can, for a few seconds, just to see how fast it goes.

Real traffic never behaves like this. A movie streams at a steady pace and uses no more than it needs. A video call needs a few megabits, delivered smoothly. A game sends tiny packets and only cares how quickly they get there and back (latency, not speed). None of them saturate your connection. Cloudflare, which runs one of the big speed tests and hosts much of the internet, makes the same point: day to day, most people use nowhere near their full bandwidth, and what they're doing cares far more about latency and stability.

The drag race isn't even consistent. Run five speed tests from different providers and you can get five different answers, because each test picks its own server, opens a different number of connections and boils its results down differently.

A speed test is still useful. It can expose a fault, or check that you're getting the capacity you pay for. But it can't tell you how your connection behaves while you're actually using it, or how often it falters over a month. And if your complaint is "my Zoom calls keep lagging," a big download number doesn’t really help.

How much speed do you actually need?

Less than you'd think. The number on your plan measures capacity: how much data the connection can move each second. For most households, about 150 Mbps down and 30 Mbps up is where extra capacity stops being noticeable.

Faster internet doesn't actually make anything run faster than it was built to. A gigabit line won't push your Netflix stream or Zoom call to a gigabit. Netflix recommends 15 Mbps for 4K and won't use much more than that. A 1080p Zoom call needs under 4 Mbps each way. An online game needs well under 1 Mbps. Even a big download is limited by the server sending it - on Steam, the average US download runs at about 200 Mbps.

So let's be generous and build a hypothetical hyper-online household. It's a weeknight, everyone's home, and everything is happening at once:

Every number is rounded up, and yet it still fits comfortably inside 150/30, with plenty of room left on the download side.

There are, of course, exceptions. If you upload video for a living or download enormous games, more speed does help. Under ideal conditions, a 100 GB game takes nearly 90 minutes at 150 Mbps and about 13 minutes on a gigabit line. If you do that every day, you'll notice.

But once a stream can play at its highest quality, more bandwidth won't make the picture any sharper. And it won't stop your call freezing.

What actually ruins your internet

We all know the feeling. Technically you're online, and some pages even load, but your call keeps freezing, or the download keeps failing.

That's a brownout. Several things decide whether your internet feels like that, and it’s almost never a shortage of bandwidth.

• Outages. The connection is simply gone. Annoying, but at least it's honest about it. Even on super-reliable fiber internet, a home connection goes completely dark about once a month on average, and the typical outage takes around 45 minutes to come back.
• Brownouts. You're technically online, but calls freeze, pages hang and video goes blocky. They're often more maddening than an outage, because you spend the first ten minutes wondering whose fault it is. They're also far more common. In FCC measurements, even the most reliable fiber service spent about 72 hours a year losing at least 1% of its data, the point where many apps start to struggle. A big cable provider: about 134 hours.
• Latency. The delay in getting data from one place to another and back. Under about 100 ms, most things feel fine, though gamers will want it well below that. Much higher, and conversations get awkward and games feel sluggish, however much speed you're paying for.
• Jitter. Even more important than latency - how much that latency jumps around. Live audio and video need a steady flow, so a connection that sits at 40 ms feels much better than one that averages 15 ms but spikes to 200. Your Zoom call and online game don’t care about the average. They care about the jitter.
• Packet loss. Data that never arrives and has to be sent again. A couple of percent is enough to make a video call wobble, and quickly eats up even the most impressive speed number.
• Bufferbloat. Someone starts a big upload or download, badly managed queues fill up, and everything time-sensitive waits behind it. Your laptop becomes unusable while another device gets on enthusiastically with its photo backup. More speed can make the line harder to fill, but doesn’t solve bufferbloat if your router and network are the problem. That's why it's worth measuring latency under load: how responsive your connection stays while it's busy. Our free bufferbloat test does exactly that.
• Backup and Support. Something will go wrong eventually. Does the connection come back on its own? Is there a backup? Does anyone actually investigate, or does support invite you to turn it off and on again for the fourth time?

How reliable is US internet, really?

Less reliable than you'd hope, and harder to find out than it should be. Providers don't publish these numbers, so the best independent data comes from two large academic studies.

Researchers at Northwestern and Princeton analyzed FCC measurements from about 8,000 homes between 2011 and 2015. They counted every hour in which a connection lost 10% or more of its data. That's enough to wreck a video call, online game, even a buffered stream, whether or not the line is technically up. Researchers at the University of Maryland and UC San Diego pinged 10.8 million US homes between 2011 and 2018 and counted how often each connection went completely silent for 11 minutes or more.

Even the most reliable service in the FCC data, a fiber plan, still lost about 18 hours a year. That's more than two full workdays. Paying for a business-class plan made little difference.

There are limited independent outage studies for 5G home internet or Starlink. What does exist points to brownouts more than blackouts. In a 2023 study by NetForecast, one major 5G home internet service averaged about 5% packet loss.

And even the biggest networks have bad days. On July 24, 2025, a software failure took Starlink offline worldwide for about two and a half hours.

The most useful finding is what fixed it. In the Northwestern and Princeton study, most failures (68%) happened on the stretch between the home and the provider. So the researchers simulated giving homes a second connection from a different provider. Reliability jumped from roughly 99% to 99.992%, which is under an hour of bad time a year.

What to shop for instead

If speed is the wrong number to shop on, here's what to look at, roughly in order:

1. Enough speed. For most homes, about 150 Mbps down and 30 up covers everything at once. Pay for more only if you regularly move big files around: uploading video, hosting a server at home or downloading enormous games.
2. A second, backup connection. The biggest reliability upgrade in the research wasn't a faster line. It was a second, independent connection on a different network, so a single failure can't take you offline. Backup internet is its own rabbit hole, read our guide here.
3. How it behaves when it's busy. Run a bufferbloat test and look at latency under load, not just the download number. If it balloons, a more capable router with smart queue management can fix a lot of it.
4. How often it lets you down. It's the hardest thing to know before you buy, so ask the neighbors. They'll tell you how often it drops or crawls at 8 pm.
5. Who fixes it. Find out what happens when it breaks: how quickly someone responds, and whether they actually investigate or just ask you to restart the router.

Nobody buys a car on top speed alone. There's no reason to buy internet that way either.

What we did about it

We sell Waveform Internet, a reliability-first internet service in a market obsessed with speed. It's still fast 5G, with two connections on separate networks, bonded for seamless failover and to smooth out jitter, and a team that monitors it 24/7 so you don't have to. I run it at my own house too, as Waveform Backup Internet on the Smart tier alongside my fast but unreliable ISP.

If you're tired of “fast” internet that lets you down and would rather have reliable internet that just works, take a look.