There is something quietly strange about the internet. We use it constantly, we depend on it for almost everything, and yet most of us have no idea what it actually is — not the physical reality of it, the cables running along ocean floors, the data centers humming in the desert, the radio waves bouncing between towers and satellites and the small rectangles in our pockets. We interact with a surface, a kind of polished interface on top of an enormous, aging, and deeply complicated machine. And that machine is changing in ways that are hard to fully grasp, even for the people building it.
When you try to think about the future of internet networks, you immediately run into the problem of speed — and not network speed, but the speed at which assumptions become obsolete. Twenty years ago, the dominant worry was bandwidth. Not enough of it, too expensive, too unevenly distributed. That problem has not been solved uniformly, but it has shifted. In many parts of the world, gigabit connections are becoming mundane. Starlink is putting satellite internet into fishing boats off the coast of Norway and into schools in rural Kenya. The constraint is no longer purely physical throughput. The questions are becoming stranger and more interesting.
One thing that seems increasingly clear is that the network of the near future will be radically more distributed than the one we have now. The current architecture of the internet — despite its original design philosophy of decentralization — has in practice consolidated into something that flows through a surprisingly small number of chokepoints. A handful of cloud providers host a vast proportion of global web traffic. A few submarine cable routes carry most of the data crossing the Atlantic or the Pacific. DNS, the humble system that translates domain names into addresses, is controlled by a small cluster of root servers. This concentration creates both fragility and power, and both of those things make people uncomfortable for different reasons.
The push toward edge computing is in part a response to this. The idea is straightforward enough: move computation physically closer to where data is generated and consumed, rather than routing everything through centralized data centers. This matters enormously for applications that are sensitive to latency — autonomous vehicles, surgical robotics, augmented reality overlays that have to respond faster than human perception. You cannot drive a car from a data center in Virginia if the car is in Seoul. But edge computing is also about something broader than low latency. It is about rethinking the architecture of where intelligence lives in a network, and that has implications that reach well beyond any specific application.
The rise of 5G, and eventually 6G, is part of this story too, though perhaps not in the way it was originally marketed. The transformative potential of millimeter-wave 5G is not primarily about downloading movies faster on a phone. It is about enabling dense networks of connected devices — sensors, cameras, industrial machines, medical equipment — that can communicate with each other and with edge infrastructure in something approaching real time. The industrial internet, the smart city, the fully instrumented supply chain: these are visions that depend not just on faster wireless links but on a different kind of network topology altogether, one that is woven into physical environments rather than accessed through discrete devices.
And then there is the question of what happens to the network’s underlying protocols. TCP/IP, the fundamental protocol suite that has governed internet communication for decades, was designed in an era of wired connections, moderate packet loss, and relatively modest scale. It has been stretched and patched and extended to accommodate a world it was never designed for, and those seams are showing. HTTP/3, running over QUIC rather than TCP, is one sign of this evolution — an attempt to build something better suited to the actual conditions of modern networks, including mobile connections that change IP addresses as users move between cell towers. But these changes are incremental. The more radical question is whether the growth of satellite constellations, delay-tolerant networking in space, and eventually communication infrastructure on the Moon or Mars will push us toward something genuinely different at the protocol level — networks designed from the start for discontinuous, high-latency, physically distributed environments.
There is also a geopolitical layer to all of this that is difficult to separate from the technical one. The internet was built largely by and for the English-speaking West, and its governance structures still reflect that history. But the infrastructure itself has become genuinely global, and different nations have different ideas about what a network should be — who controls it, who can be excluded from it, what data can flow across which borders. The fragmentation of the internet into something more like a collection of national or regional internets is not a hypothetical; it is already happening in various forms. How far it goes, and whether technical standards can hold together a network that political pressures are pulling apart, is one of the genuinely open questions of the next decade.
What is harder to predict, and perhaps more interesting to think about, is the way AI is beginning to change the network itself rather than just running on top of it. Networks are already using machine learning to manage routing, predict failures, optimize traffic, and detect anomalies. But as inference moves to the edge, as models become embedded in devices and infrastructure rather than living in data centers, the network and the intelligence it carries start to blur together in ways that are conceptually new. A network that is partly thinking, partly routing, partly sensing — a substrate that is not just carrying messages between points but actively participating in what those messages mean and where they go — is a different kind of thing than what we have now.
None of this follows a straight line. Technologies that seem inevitable get delayed or displaced. Problems that seem intractable get solved by approaches nobody anticipated. The history of networking is full of confident predictions that turned out to be wrong, and there is no particular reason to think the next twenty years will be different in that regard. But the direction of pressure is clear enough: toward more distribution, more intelligence embedded in infrastructure, more physical density of connection, and a deeper entanglement between the network and the physical world it runs through. Whatever emerges from that will be something we are only beginning to have the vocabulary to describe.
The internet is not a cloud. It never was. It is infrastructure — physical, fragile, political, and evolving — and how it evolves will shape almost everything else about the world we are building.