Who Owns the Wires Your Country Doesn't

The internet has no center, except where it does

The idea that the internet has no headquarters is not myth. It was a design goal. The modern concept of an internet backbone traces to NSFNET, the academic network the US National Science Foundation funded and operated from 1985 to 1995 as the internet's original core transit system, engineered so traffic could route around the failure of any single link. When NSFNET was decommissioned in 1995, the routing job it had done did not pass to a government agency or to any single successor. It passed to a set of competing commercial carriers, and the internet has been routed that way, through private companies with no central coordinator, ever since.

What replaced a government-run backbone was not a backbone in the old sense at all, but a marketplace of connections between independent networks. Two kinds of arrangement carry traffic between them. Transit is a paid relationship: a smaller network buys the right to reach the rest of the internet through a larger one's connections, the way a shop pays a courier to reach customers it cannot deliver to directly. Peering is different: two networks agree to exchange traffic bound for each other's own customers directly, at a shared physical location built for exactly that purpose, usually with no money changing hands for the data itself. That location is an internet exchange point, an IXP, and it is where the internet's supposed formlessness turns out to have a street address after all.

There are hundreds of IXPs worldwide, but the traffic moving through them is not spread evenly among them. A handful, clustered in three European cities, carry an outsized share of the world's interconnection. DE-CIX in Frankfurt, AMS-IX in Amsterdam, and LINX in London have spent decades building the density of connected networks that makes an exchange point valuable in the first place: the more networks already present, the more useful it becomes for the next network to join, and that advantage compounds year over year. A financial center's worth of infrastructure investment, a set of favorable subsea cable landings, and forty years of incumbency did the rest, and no amount of internet decentralization elsewhere has displaced them.

The first exchange points of this kind emerged in the early 1990s, in the years around NSFNET's wind-down, for a specific and unglamorous reason. As the internet stopped being one government-funded network and became a collection of competing commercial providers, those providers needed a place to hand traffic to each other without each paying the other for transit. The exchange point solved a coordination problem, not a political one. It became a chokepoint anyway, because coordination points tend to accumulate the power of everyone who has to show up and be coordinated.

How an exchange point earns its keep

An IXP's underlying business is plainer than the geopolitics around it suggests. A network pays for a port, a physical connection into the exchange's switching fabric, and then sets up individual peering sessions with the other networks present at the same facility. Both the port and the peering connections are billed on a recurring basis, month after month, whether or not the traffic crossing them changes at all. That recurring fee is what turns operating a major exchange into a durable business rather than a one-time infrastructure build, and it is the commercial engine behind Frankfurt, Amsterdam, and London's continued dominance.

The traffic that actually crosses those connections once a network has joined is usually free between the two parties. Most peering arrangements at an IXP are settlement-free: no money changes hands for the data itself, only for the port and the connection. That detail changes what an exchange point is actually selling. It is not a toll booth charging by the byte. The commercial logic runs toward avoidance rather than extraction: a network joins an exchange specifically to stop paying transit fees to reach every other network gathered there, and the exchange operator charges for the privilege of being in the room where that avoidance becomes possible.

For a country, the same arithmetic scales up, and the consequences show up in latency and in hard cost rather than in abstraction. Before India built the National Internet Exchange of India, including the Mumbai exchange, two networks sitting in the same Indian city could still be forced to send traffic to each other by way of Singapore or the United States, because no domestic exchange point existed where they could hand traffic off directly. Every one of those detours meant a longer round trip for the data and a transit bill paid to a foreign carrier for the privilege of one Indian message reaching another Indian recipient. NIXI exists to close exactly that loop, so the country's own traffic can stay inside the country.

The businesses that feel a detoured route most acutely are rarely the large multinationals with private network capacity of their own to fall back on. A small exporter selling through an online marketplace, or a regional bank running its core systems from a domestic data center, pays the same detour tax as everyone else on the connection: a slower page load for a customer abroad, a higher latency reading on a payment gateway, a transit invoice passed down the chain until it lands on the smallest business least able to negotiate it away. A domestic exchange point does not only serve national pride. It serves the ordinary economics of every company whose customers, suppliers, or bank happen to sit inside the same border.

The Frankfurt, Amsterdam, London numbers

The scale now involved is not small. DE-CIX Frankfurt, widely reported as the world's largest internet exchange point by peak traffic, recorded throughput of 18.73 terabits a second in December 2025, a figure that would have read as science fiction at the exchange's founding in the 1990s and now represents a single evening's peak load on one exchange's switching fabric alone.

AMS-IX in Amsterdam is close behind it. The exchange carried 35.66 exabytes of traffic across its Amsterdam platform over the course of 2025 and opened that year with a fresh traffic peak of 14.2 terabits a second. LINX in London recorded peak traffic above 11 terabits a second as of November 2025. Industry reporting on the world's thirty largest exchange points found that the number of bilateral peering sessions between networks at those exchanges grew 18 percent between January 2024 and January 2026, a pace suggesting the pull toward a small number of major hubs is still gathering strength rather than leveling off.

What flows through those wires has itself become concentrated by ownership. Content and cloud computing traffic generated by Google, Meta, Microsoft, and Amazon's own networks made up 75 percent of used international internet bandwidth in 2025. So the exchange points in Frankfurt, Amsterdam, and London are not routing a broad cross section of global commerce in roughly even shares among thousands of small players. They are routing traffic dominated by a handful of American technology companies, which means the concentration compounds in two directions at once, geographically into three cities and commercially into a small number of owners.

The double concentration also means a single policy decision, outage, or ownership dispute at any one of the three hubs carries outsized global consequences, because so much of the world's interconnection now runs through so few physical rooms. A regulatory order, a fire, or a change of control at DE-CIX, AMS-IX, or LINX would not stay a local story. It would ripple through every network that peers there, in every country whose traffic happens to transit Frankfurt, Amsterdam, or London on its way somewhere else entirely.

Whoever routes it can read it

Tier 1 status is a specific and narrow technical claim, not a marketing term. A Tier 1 network is one that can reach every other network on the internet through settlement-free peering alone, without ever paying anyone for transit. Only a small club qualifies, roughly a dozen carriers worldwide. Membership is not symbolic. A Tier 1 carrier's decision to peer, or to stop peering, with a given network can functionally sever an entire region's ordinary route to the rest of the internet, forcing its traffic onto slower or costlier detours until an alternative path is negotiated.

That is the plain fact underneath the geopolitics: physical control of a wire is control, however lightly it is normally exercised, and however neutral the operator running it claims to be. A state that hosts a major exchange point, or the core infrastructure of a Tier 1 carrier, holds a standing ability to observe traffic that has no real reason to be inside its jurisdiction at all beyond the accident of an efficient route, and in principle to interrupt it. A business in one country sending data to a customer in a third country may still have that data cross a router sitting under German, Dutch, or British law, simply because the shortest efficient path happens to run through Frankfurt, Amsterdam, or London.

Russia turned that anxiety into statute. Its so-called sovereign internet law, which took effect in November 2019, required Russian internet service providers to build the technical capacity to route all domestic Russian traffic through exchange points the state itself controls. The stated purpose was explicit rather than implied: to cut the country's dependence on internet infrastructure that Western-controlled networks could reach, monitor, or, in a crisis, withhold.

None of this is only hypothetical. Peering relationships between major carriers have broken down into deliberate disconnection before, over commercial disputes rather than politics, and state authorities have, in moments of unrest, ordered domestic providers to sever international links entirely. Rerouting around a single blocked path is technically possible in most cases, because the internet's design allows for alternate routes. It is rarely free, fast, or guaranteed, which is exactly why the position of being the blockage, rather than the network scrambling to route around one, is worth building toward.

Building your own wires

India made a version of the same decision years earlier, for a mix of the same underlying reasons and its own commercial ones. NIXI was built so domestic Indian traffic would not need to cross an international boundary, historically often to Singapore or the United States, purely to move between two points that were both already inside India. The immediate case for building it was cost and latency, not confrontation, but the shift in exposure is the same one Russia later legislated for more openly: a country's messages to itself stop needing to pass through infrastructure that another government could, in principle, watch or cut.

Neither country's move eliminated its dependence on the wider internet, and neither claimed to. India and Russia both still need the rest of the world's networks in order to reach the rest of the world, and both remain plugged into an interconnection system whose largest, densest hubs still sit in Frankfurt, Amsterdam, and London. What a domestic exchange point buys is narrower and more specific than full independence: control over the traffic that starts and ends at home, which happens to be the traffic a state is best positioned, and most motivated, to protect.

Whether this actually delivers the security these laws promise is not settled, and network engineers and policy analysts read the same infrastructure in opposite directions. One argument holds that domestic routing genuinely reduces a state's exposure to foreign interception and disruption, moving traffic out of a jurisdiction with no accountability to the country whose data it is. The opposing argument holds that a state which requires all its traffic to detour through domestically monitored infrastructure has simply built its own government a permanent viewing platform over ordinary citizens' communications, trading concentration by a distant government for concentration by a familiar one. Both descriptions fit the same wires. They differ only in which government the reader trusts less with the same technical capability.

What concentration liberates, and what it costs

The account of this system cuts both ways, and the two-sided version is the more useful one to hold. The same concentration that exposes traffic to a handful of jurisdictions is also what makes the modern internet as cheap as it is. Because Frankfurt, Amsterdam, and London hold such dense collections of networks already connected to one another, a new internet service provider almost anywhere in the world can join a single exchange and reach thousands of other networks through peering, rather than negotiating and paying for transit to each one individually. That density is a genuine public good. It is a real part of why global bandwidth has grown steadily cheaper even as the volume moving across it has climbed.

The cost sits alongside the benefit rather than canceling it out. A router does not know, or care, whether the packets crossing it belong to a hospital, a newsroom, or a dissident, and the state under whose law that router happens to sit is under no technical obligation to look away either. The internet's celebrated neutral core stays neutral because the companies operating it, mostly private firms with a commercial interest in staying out of politics, choose not to exploit the position they hold. That is a policy choice, upheld by industry norms and by each exchange's own interest in appearing trustworthy to every network that might otherwise route around it, not a structural guarantee built into the wires themselves.

The same shape is now showing up one layer higher in the stack, in the systems that decide which businesses get named in an answer rather than which packets get routed to a destination. A small number of AI answer engines increasingly stand where the exchange points stand, positioned in the middle, deciding what passes through legibly and what does not, for reasons a business on the other end rarely gets to see or contest. The wires taught this lesson first. Whoever sits in the middle of an exchange controls the crossing for everyone who has no other way through, whether what is crossing is a data packet routed at the speed of light or a business trying to appear in the answer a buyer was just given.

None of this argues for tearing the system down and starting over, and nothing in the traffic data suggests anyone could if they tried. It argues for treating physical concentration as a fact to plan around rather than a detail to ignore. A country's domestic exchange point, a business's redundancy across more than one network provider, and a firm's independence from any single machine's judgment about who deserves to be found are variations on the same insurance policy, purchased against the same underlying risk: that whoever sits in the middle of a chokepoint eventually acts like it.