The internet feels like magic until something breaks. Then it turns out to be a mess of cables, servers, and creaky old protocols held together by a few thousand engineers and a lot of duct tape.

Once you know what’s under the hood, the whole thing gets easier to reason about. Here’s the short tour: cables, data centers, proxies, routing, standards, the works.

Cables and Concrete

Almost all international traffic (roughly 99% of it) runs through undersea fiber cables. There are around 550 of these things sitting on the ocean floor right now, and they’re what makes a video call to Tokyo actually work. Satellites pick up the slack in remote areas, but they’re slower and pricier, so nobody wants to lean on them.

At the ends of those cables are data centers. The huge ones belong to Amazon, Microsoft, and Google, but there’s a whole ecosystem of colocation providers, like Equinix and Digital Realty, doing the boring but essential work of hooking everyone up to everyone else.

There are also internet exchange points, or IXPs, where different networks meet up to swap traffic. AMS-IX in Amsterdam and DE-CIX in Frankfurt each push more than 10 terabits per second at peak. That’s a lot of TikTok.

The Middlemen

Once traffic leaves a device, all sorts of intermediary services can get involved before it reaches its destination. Getting a handle on what is a residential proxy is a decent starting point: residential proxies route traffic through IP addresses tied to real homes with real ISP contracts, so from the outside it looks like a person on their couch rather than a server in a rack.

Companies use them for things like ad verification, price monitoring, and pulling geo-specific search results. Datacenter proxies do a similar job but faster and cheaper, with the tradeoff that they’re pretty easy for websites to spot and block.

Then there are CDNs. Cloudflare and Akamai park copies of content closer to users, which is why a video on YouTube loads in a second instead of ten. Load balancers pull the same trick inside a company’s own infrastructure, spreading incoming requests across servers so nothing melts.

Reverse proxies sit at the edge of most big web apps too. They handle SSL, block obvious junk traffic, and hide what’s going on behind them. NGINX and HAProxy have basically become the defaults here, and anyone using the web today is almost certainly touching one somewhere.

Routing, DNS, and the Power Bill

Data doesn’t take a straight line anywhere. Packets get shuffled through routers using BGP (Border Gateway Protocol), which is more or less the internet’s version of “somebody figure out how to get this there.” It works, mostly.

BGP was built in a friendlier era and doesn’t really care about security. That’s why route hijacks still happen every so often, where traffic ends up pulled through some random network for a few minutes before anyone notices. RPKI helps, but the fix has been slow going for over a decade.

None of this is free to run. According to theInternational Energy Agency, data centers ate up around 1.5% of the world’s electricity in 2024, and that number is heading up as AI training soaks up more capacity. Cooling alone is a serious chunk of the bill.

DNS deserves a nod. It’s the thing that turns “google.com” into an IP address, and when it goes down, half the internet appears to go with it. Ask anyone who was around for the 2021 Facebook outage.

The Rulebook

Somebody has to agree on how all of this talks to itself. TheInternet Engineering Task Force writes the RFCs that spell out how protocols behave, from old-school TCP/IP through HTTP/3 and QUIC. ICANN runs the top level of the domain name system, and Regional Internet Registries hand out IP blocks by region.

Everything above that is politics. Wikipedia’s overview of internet governance is a decent read if you want the full picture. It’s a lot messier than most people assume, and it’s what shapes things like data sovereignty laws and country-level site blocking.

Zoom Out

The internet isn’t a single thing. It’s a stack: physical hardware at the bottom, protocols in the middle, intermediaries and standards bodies layered on top. Each part has its own quirks, its own money, and its own set of people arguing about how it should work.

Knowing that layout pays off. When something breaks (and eventually something always does), it’s a lot faster to fix if you already know where to look.