Every search, stream, and payment depends on a layer of technology almost nobody sees: submarine cables under the ocean, data centers the size of small towns, and protocols quietly negotiating handshakes in milliseconds. This is the invisible technology behind everything online the physical and logical infrastructure that makes the internet function, largely without a name attached to it in most people’s minds. Here’s what it actually consists of, who runs it, and why it’s becoming more expensive to keep running.
What “Invisible Technology” Actually Means Here
The internet is generally described as a single thing, but it’s really a stack of separate systems working together: physical infrastructure that moves electrical or light signals, protocols that structure how devices talk to each other, routing systems that decide which path data takes, and governance bodies that keep all of it interoperable without any single owner. Each layer is largely invisible to a person opening a browser, but a failure at any one of them is usually what makes the internet visible through an outage, a slowdown, or a security incident.
The Physical Layer: Cables, Data Centers, and Towers
The physical infrastructure of the internet moves data across oceans, stores it in data centers, and connects it to homes and businesses.
Submarine Cables Carry Almost Everything
Despite the common image of satellites beaming data across continents, the overwhelming majority of international internet traffic — generally cited at around 99% — travels through fiber-optic cables laid across ocean floors. There are several hundred of these systems in service globally, connecting continents through a relatively small number of coastal landing stations. Traditionally, these cables were funded by consortiums of telecom carriers, but large technology companies now finance and build private cable systems specifically to connect their own data centers and cloud regions, giving them direct control over latency and capacity rather than leasing it from a third party.
Data Centers Are the Internet’s Warehouses
If cables are the roads, data centers are the warehouses. They house the servers and storage systems where websites, apps, and cloud services actually live, and they range from small regional facilities to hyperscale campuses covering millions of square feet. What most public discussion misses is how much energy this warehousing now requires — a point worth returning to in more detail below.
The Last Mile
The final stretch of infrastructure — the connection between an internet service provider and an individual home or office — is often the part that actually determines a person’s day-to-day experience of “the internet,” more than any transcontinental cable does. This last mile can run over fiber, coaxial cable, DSL, or fixed wireless, and the choice of technology explains much of the speed and reliability variation between neighborhoods, let alone countries.
The Protocol Layer: The Rules That Let Devices Talk
Underneath the physical wires, a set of agreed-upon rules — protocols — define how any two devices exchange data regardless of who built them or what software they run.
TCP/IP is the foundational suite: IP handles addressing and routes each packet of data toward its destination, while TCP breaks data into packets, tracks whether they arrived, and reassembles them in order at the other end. Nearly everything else on the internet is built on top of this pairing.
DNS, the Domain Name System, functions as the internet’s address book, translating a human-readable address like a website name into the numerical IP address a machine actually needs to route a request. Security add-ons like DNSSEC and DNS over HTTPS have improved the integrity and privacy of this lookup process over the past several years, though adoption across the internet as a whole remains uneven.
TLS, the encryption protocol behind HTTPS, is what establishes a secure, encrypted connection between a browser and a server before any content is exchanged — the mechanism behind the padlock icon in a browser’s address bar.
Who Actually Runs This System
There is no single company or government that owns the internet, which surprises people when they first think about it seriously. Instead, a mix of technical and nonprofit bodies coordinate the parts that need global agreement:
- ICANN (Internet Corporation for Assigned Names and Numbers) oversees the DNS root zone and allocates top-level domains.
- IANA (Internet Assigned Numbers Authority), operating under ICANN, manages the global pools of IP address space and autonomous system numbers.
- Regional Internet Registries – ARIN, RIPE NCC, APNIC, LACNIC, and AFRINIC then distribute those address blocks to internet service providers and organizations within their regions.
- IETF (Internet Engineering Task Force), an open volunteer technical body, designs and publishes the RFC documents that define core protocols like TCP/IP, HTTP, and DNS.
This decentralized structure is generally credited with letting the internet scale globally over three decades without a central point of control or failure — though it also means no single entity can unilaterally fix a global problem when one occurs.
Routing and Resilience: BGP, IXPs, and What Happens When a Cable Breaks
Data doesn’t take a fixed path between two points online. BGP (Border Gateway Protocol) is the routing protocol that lets independent networks, known as autonomous systems, exchange information about the best available path for traffic and adjust in real time. Internet Exchange Points (IXPs) are physical facilities, typically inside data centers, where multiple networks connect directly to each other rather than routing local traffic across an ocean and back. This local peering both reduces latency and adds resilience: if a transcontinental cable is damaged, an IXP can often keep regional traffic flowing over alternate paths while international routes are rerouted.
Cable damage itself is not rare. Anchors dragging along the seabed, fishing equipment, and occasional natural events sever cables on a fairly regular basis somewhere in the world. Because most systems have redundant routes and most regions are served by more than one cable, a single break rarely causes total outages — but the rerouted detour typically increases latency and can create noticeable slowdowns until repair ships, a specialized and limited global fleet, can reach and fix the site.
The Layer Most People Never Think About: Content Delivery Networks
Even with fast cables and efficient routing, physical distance still adds delay. Content delivery networks solve this by caching copies of frequently requested content — images, video, static pages — on servers positioned close to where users actually are, so a request doesn’t need to travel back to a single origin server every time. Cloudflare, Akamai, Amazon CloudFront, Google Cloud CDN, and Fastly are among the largest providers, and market share estimates vary by measurement method: some trackers show Akamai holding the largest overall share of CDN infrastructure by revenue, while usage-based trackers of live websites show Cloudflare with a substantially larger footprint. Either way, CDNs have become close to mandatory infrastructure for any site or app expecting meaningful global traffic, since the alternative is every single request traveling the full physical distance to one origin server.
Core Invisible Technologies at a Glance
The internet relies on several interconnected layers, each handling a different part of how data moves and services stay connected.
| Layer | What It Does | Examples |
| Physical | Moves the actual signal | Submarine cables, data centers, last-mile fiber/DSL |
| Protocol | Defines how devices communicate | TCP/IP, DNS, TLS/HTTPS |
| Routing | Decides the path data takes | BGP, Internet Exchange Points |
| Delivery | Reduces distance-based delay | CDNs, edge caching |
| Governance | Coordinates global agreement | ICANN, IANA, IETF, Regional Internet Registries |
The Growing Cost of the Invisible: Data Center Energy Demand
One of the least visible costs of “the cloud” is how much electricity it now consumes. According to the International Energy Agency, global data center electricity consumption reached roughly 415 TWh in 2024, about 1.5% of global electricity use, and had been growing at around 12% annually over the preceding five years. The IEA’s base-case projection has that figure roughly doubling to around 945 TWh by 2030, representing just under 3% of global electricity demand at that point — driven largely by the buildout of AI-specific “accelerated” servers, whose electricity consumption is projected to grow around 30% a year, compared to roughly 9% annual growth for conventional servers.
This demand is heavily concentrated geographically rather than spread evenly. The United States and China together account for the large majority of global growth in data center electricity use, and in some regional markets the effect on the local grid is already substantial: Ireland’s data centers already account for more than a fifth of the country’s national electricity demand, a share the IEA has suggested could approach a third by 2026. This is one of the more concrete ways that internet infrastructure, despite being invisible in daily use, has a measurable and growing footprint on physical energy systems.
What Happens in the Time It Takes to Load a Page
A page load that feels instantaneous actually involves several discrete, sequential steps, each depending on the infrastructure layers above:
- DNS lookup – the browser resolves the site’s domain name into an IP address, often via a cached response from a nearby DNS resolver rather than a fresh query to the root system.
- TCP handshake – the browser and server exchange a short sequence of packets to establish a reliable connection.
- TLS handshake – for an HTTPS site, both sides negotiate encryption keys before any content is transmitted.
- HTTP request and response – the browser requests the page’s resources, and the server (often via a CDN edge node rather than the origin server itself) returns them.
- Rendering – the browser assembles the returned data into the page a person actually sees.
Under typical conditions, this entire sequence completes in well under a second, which is part of why the infrastructure behind it goes largely unnoticed — it’s only when one of these steps is delayed, by a DNS misconfiguration, a routing problem, or a distant origin server, that the underlying system becomes visible to an ordinary user as a slow-loading page.
Security Risks Built Into an Open System
The same openness that let the internet scale globally also creates specific, well-documented attack surfaces. Distributed denial-of-service (DDoS) attacks overwhelm a target with traffic to knock it offline. BGP hijacking exploits the trust-based nature of internet routing to redirect traffic through an unauthorized network, sometimes for surveillance or interception. DNS attacks can redirect users toward malicious servers by corrupting the address-lookup process itself. Most large infrastructure providers now build defenses against these specific threats directly into their CDN and routing products, which is part of why outages caused by deliberate attacks are generally less common in practice than outages caused by misconfiguration or physical damage.
Why This Layer Matters More as AI Expands
AI services depend on this same invisible stack, just under heavier load. Training and running large models requires data centers with far higher power density than traditional web hosting, which is a direct driver of the electricity demand growth described above. It also increases pressure on network capacity, since AI-powered features increasingly route requests through additional data-processing steps before returning a response. None of this changes the fundamental architecture — DNS, TCP/IP, BGP, and CDNs still do the same jobs they always have — but it does mean the physical and energy constraints of this infrastructure are becoming a more direct factor in how quickly new online services can scale, rather than a background concern.
Frequently Asked Questions
Why does a website sometimes load fast at home but slowly on mobile data in the same location?
The two connections generally take different physical and routing paths — home fiber or cable typically runs through a more direct last-mile connection, while mobile data often adds an extra hop through a cellular network before reaching the same backbone infrastructure.
If a submarine cable is cut, does the internet actually go down?
Rarely in full, because most regions are served by more than one cable system and traffic can generally be rerouted through alternate paths and Internet Exchange Points. The typical effect is increased latency or reduced capacity on affected routes rather than a complete outage, though repair can take days to weeks depending on the cable’s location and depth.
Is there a difference between “the internet” and “the cloud”?
Yes, the internet is the underlying network and protocol infrastructure described above, while “the cloud” generally refers to computing and storage services run on data centers that are accessed over that same internet infrastructure. Cloud services depend entirely on the internet layer; the internet does not depend on any particular cloud provider.
Does using a VPN change which invisible infrastructure your traffic passes through?
It changes the routing and encryption path, a VPN typically sends traffic through its own server before it reaches the destination and encrypts it along the way, but the underlying physical cables, DNS system, and BGP routing are still the same global infrastructure everyone else’s traffic uses.
Why do some countries have much slower average internet speeds than others?
This is generally explained by a combination of last-mile infrastructure investment, geographic distance from major cables and data centers, and the level of competition among local internet service providers, rather than any single global factor.
Conclusion
The technology that makes “going online” feel instantaneous is anything but simple: it’s a layered stack of undersea cables, data centers, routing protocols, and decentralized governance bodies, none of which most users ever interact with directly. That invisibility is, in a sense, the system working as intended but it comes with real physical costs, from vulnerable cable landing points to a rapidly growing electricity footprint, that are becoming harder to ignore as AI and cloud services push more load through the same underlying infrastructure.
