Nobody owns the internet. It is an agreement.
Roughly 80,000 separate networks — ISPs, tech companies, universities, governments — agreed to forward each other's traffic. There is no central router and no central database. Each network announces which addresses it can reach; every other network listens, remembers, and figures out the best path. The whole thing works through trust and BGP.
The protocol holding it all together is mostly announcements and reputation.
Imagine every network on Earth shouting into a global PA system, on repeat: “I can deliver mail to ZIP codes 10001 through 10099.” Every other network listens, writes it down, and remembers the path the announcement took. When you send a packet, the nearest network looks up which path is shortest and forwards it that way.
That's essentially how BGP works — except instead of ZIP codes it's IP prefixes, and instead of mail trucks the path is a list of intermediate networks. The whole internet is held together by these announcements, billions of them per day, between roughly 80,000 networks.
- Standard
- RFC 4271 (BGP-4)
- Transport
- TCP/179
- IPv4 prefixes in DFZ
- ≈ 950,000
- IPv6 prefixes in DFZ
- ≈ 200,000
- Active ASes
- ≈ 80,000+
- Path selection
- AS_PATH, LOCAL_PREF, MED
- Security overlay
- RPKI ROA + ROV
- Convergence after major change
- 1–10 min globally
The handshakes that turn 80,000 networks into one network.
Paid upstream
A smaller network buys access to everywhere from a larger one. Full routing table; about $0.30–$2 per Mbps committed in 2026. Typical providers: Lumen, Cogent, NTT, Tata, GTT.
Settlement-free swap
Two networks agree to deliver each other's traffic for free at an Internet Exchange (DE-CIX, AMS-IX, LINX, Equinix). Filters via IRR and RPKI. Cheap, low-latency, the workhorse of the modern internet.
Private interconnect
A dedicated cross-connect between an ISP and a hyperscaler (Google, Meta, Amazon). 100GE+ handoffs. Lowest latency, highest reliability, used for content with serious volume.
- Modulation
- PCS-64QAM (typical)
- Channels (C-band)
- 96
- Per-channel rate
- 400G / 800G ZR+
- Amplification
- EDFA every 80–100 km
- FEC overhead
- ≈ 25% (soft-decision)
- Total per-fiber capacity
- ~30–80 Tbps
- Span length record
- >10,000 km transoceanic
One pair of glass strands carries enough bandwidth for a country.
96 colors of light on a single fiber, each carrying 400 to 800 gigabits per second.
On the long-haul fiber that carries packets between cities, ISPs squeeze far more than one signal onto each strand. Lasers tuned to ninety-six slightly different wavelengths inject ninety-six streams of data into the same fiber, each one a different color of infrared. At the other end, prisms split them back out.
Every hundred kilometers or so, a doped section of fiber and a small pump laser boost all ninety-six signals at once — without ever turning them back into electricity. The light stays as light all the way across the continent.
The internet has no headquarters. No board of directors. No one is in charge. It runs because the alternative is for it to stop.
The backbone, in scale.
Questions about how the backbone actually works.
What is the internet backbone?
The internet backbone is the network of long-haul fiber, peering exchanges, and high-capacity routers that interconnect roughly 80,000 separate networks worldwide. It's not owned by anyone single entity — it's the cumulative result of every major ISP, hyperscaler, and tier-1 carrier agreeing (and contracting) to forward each other's traffic.
What is BGP?
BGP — the Border Gateway Protocol — is how networks tell each other which IP addresses they can reach. Each network announces its prefixes (IP ranges) to its neighbors; those neighbors propagate the announcements outward; eventually every BGP-speaking router on Earth has a path back to you. There's no central authority. It's just announcements all the way down.
What is an Autonomous System (AS)?
An Autonomous System is a network with its own external routing policy — usually an ISP, a content provider, or a large enterprise. Each AS has a globally-unique number (e.g., Google is AS15169, Cloudflare is AS13335). When you trace a route across the internet, you're really tracing a path through ASes.
What is an Internet Exchange Point (IXP)?
An IXP is a physical building where many networks bring their cables to meet and swap traffic with each other directly, instead of routing through paid upstream carriers. The largest exchanges (DE-CIX Frankfurt, AMS-IX Amsterdam, LINX London, Equinix Ashburn) carry terabits per second across thousands of cross-connects. Peering at an IXP is usually free or near-free between participants — it's a settlement-free agreement.
What is the difference between transit and peering?
Transit is when a smaller network pays a larger one to deliver traffic to anyone, anywhere on the internet. Peering is when two networks of comparable size agree to deliver each other's traffic for free, on the theory that the value is mutual. Most large ISPs run a mix: paid transit to a tier-1 carrier as a safety net, plus extensive free peering at IXPs to keep transit bills small.
What is a tier-1 network?
A tier-1 network is one that can reach every other network on the internet without paying anyone for transit. They exchange traffic with each other through settlement-free peering. There are only a handful: Lumen (formerly CenturyLink/Level 3), AT&T, NTT, Telia, Tata, Cogent, Verizon, and a few others, depending on who's counting.
How many submarine cables are there?
About 552 active submarine cables as of 2025, carrying roughly 99% of intercontinental internet traffic. The cables themselves are about as thick as a garden hose, contain 8–24 pairs of fiber each, and are laid by specialized cable-laying ships over months. Newer cables increasingly belong directly to Google, Meta, Microsoft, and Amazon — the hyperscalers now operate their own private subsea infrastructure.
How fast does data travel across the backbone?
Light in fiber moves at roughly 200,000 km/s (two-thirds the speed of light in vacuum). New York to London is about 5,500 km of fiber, so the one-way speed-of-light minimum is about 28 ms; in practice, transatlantic round-trip latency is 70–90 ms after routing, queuing, and protocol overhead. The fiber is the fastest part — every router along the way adds microseconds.
Can sharks bite submarine cables?
Yes, but rarely the cause of outages. The vast majority of submarine cable damage comes from fishing trawlers, anchors, and underwater landslides. Sharks have famously bitten cables — Google even armored some of theirs after viral footage in 2014 — but it's a small fraction of total faults. Repair ships handle 100+ subsea cable faults globally every year.