Could the Internet Collapse Overnight?

Every few years, a major outage happens — AWS goes dark for a few hours, Cloudflare has a bad day, a submarine cable gets cut — and the same question floods search engines: could the whole internet just… stop? It’s an understandable fear. But it’s also the wrong question, and asking the wrong question gets you the wrong preparation.

The internet will not collapse overnight in the way a power grid can. It wasn’t designed to have an off switch — redundancy is baked into the protocol layer going back to the 1970s. TCP/IP was literally designed to route around damage, and it’s genuinely good at this. What can happen — and what has happened, repeatedly, with increasing scale — is cascading regional failure: a sequence of events that doesn’t kill the internet globally but effectively kills it for specific geographies, platforms, or economic sectors, sometimes for days.

That distinction matters. A lot. If you’re preparing for the wrong threat, you build the wrong resilience. So let’s be precise about what the failure modes actually are.

The Concentration Paradox: Why “Redundancy” Is Your False Comfort

Here’s the thing that keeps infrastructure engineers up at night, and that almost no mainstream piece on this topic addresses correctly: the redundancy that protects the internet is increasingly routing through the same three buildings.

AWS, Google Cloud, and Microsoft Azure collectively control somewhere between 65–67% of the global cloud infrastructure market, per Synergy Research Group’s 2025 data. That’s not a bug in the system — it’s a consequence of the economics of scale. But here’s the paradox it creates: a company that “diversifies” by using both AWS and Azure hasn’t actually diversified its infrastructure risk. Both data centers run on the same undersea cables, route through the same internet exchange points, and share the same BGP peering relationships with the same tier-1 providers.

This is what I’d call the redundancy illusion. The physical layer underneath all that digital diversity is far more concentrated than the vendor count implies. When the Cloudflare BGP incident in November 2023 propagated, it didn’t care which cloud provider’s applications were calling its APIs — they all went down together. About 10% of global internet traffic was affected within minutes.

This is the structural moat of the real risk — and it’s the thing every “internet collapse” piece misses by focusing on the headline threat (cyberattack, solar storm) rather than the architecture underneath. A targeted attack against a tier-1 provider’s BGP routing tables would cascade faster, and further, than a cable cut, precisely because BGP was not designed for adversarial conditions. It was designed in a world where every router on the network was assumed to be telling the truth.

The Actual Threat Landscape, Ranked by Real-World Evidence

Let’s go through the failure modes that have actually caused significant outages, ordered by how many people they affected and how fast damage propagated. Not by which ones make better headlines.

Three Realistic Scenarios and What Actually Happens

Instead of “could the internet collapse,” it’s more useful to ask: what does a serious-but-plausible disruption actually look like for the people living through it? Here are three scenarios ranked by realism, not by drama.

What Society Actually Loses (and What It Doesn’t)

The “society without internet” thought experiment tends toward apocalypse in online discourse. The reality of past major outages is both more mundane and more instructive. The 2021 Facebook outage — six hours, 3.5 billion users, $100 million in lost ad revenue — produced a wave of think-pieces about digital dependency. It also produced almost zero deaths, grocery shortages, or infrastructure failures. People called their families on the phone. The world kept turning.

Longer outages would look different. Here’s what the impact chain actually looks like at each duration:

The sectors that break fastest aren’t the ones you’d guess. Financial markets have extensive offline backup systems precisely because they anticipated this threat. The real fragility is in healthcare — specifically in the quiet shift from paper medical records to cloud-based EHR systems that require constant connectivity. And in logistics, where just-in-time supply chains have stripped out the inventory buffers that would have provided breathing room in a pre-internet era.

What Actually Works — For Organizations and Individuals

The advice you’ll find in most “internet outage preparation” guides is too generic to be useful: “use offline backups,” “have cash,” “diversify cloud providers.” All true. None of it addresses the specific failure modes we’ve just walked through. Here’s preparation that maps to the actual risks.

For organizations

• 01
  Audit your BGP exposure. If your primary connectivity runs through a single upstream provider, you’re one misconfiguration away from a full outage regardless of how many cloud vendors you use. Multi-homing — maintaining BGP sessions with two or more transit providers in different autonomous systems — is the single highest-leverage mitigation for the cascading-failure scenario. Most organizations with under $100M revenue have never done this.
• 02
  Treat cloud redundancy and network redundancy as separate problems. AWS + Azure isn’t resilience if both route through the same IXP. Map your actual network path — not your vendor dashboard — to identify where concentration actually lives. Tools like BGPView and CAIDA’s traceroute tools will show you things your cloud console won’t.
• 03
  Run a “DNS failure” tabletop at least annually. Ask: if your authoritative DNS went dark for 4 hours, what breaks first? Most organizations don’t know. The answer is usually surprising — and the fix is usually cheap (secondary DNS provider, DoH fallback configuration).
• 04
  For geographically exposed operations (Australia, New Zealand, Pacific islands, East Africa): submarine cable redundancy is a real strategic variable, not an abstract risk. Know which cables your upstream providers use. Know the repair SLAs. Some organizations in high-exposure regions have added Starlink or OneWeb as a legitimate backup layer — not for performance, but for the cable-cut scenario.

For individuals

• 01
  The realistic preparation isn’t survivalist — it’s boring. Cash in hand (the payment ecosystem breaks before banking does). Downloaded offline maps. A battery-powered or hand-crank radio. A printed list of emergency contact numbers. These address 95% of plausible disruption scenarios.
• 02
  Learn what’s on your local network versus what requires internet. Your router’s admin panel, local NAS, Raspberry Pi home server — all accessible when the internet is down. Understanding which of your tools are purely local vs. cloud-dependent changes your outage experience dramatically.
• 03
  For high-risk regions: goTenna and similar mesh radio devices allow local communication without any internet infrastructure. During the 2024 Red Sea cable cuts, users in affected regions found that local mesh networks maintained community communication when international traffic collapsed.

What the Evidence Points Toward After 2026

Read together, the BGP incident history (20+ major propagation failures in the last decade, per BGPMon’s archives), the cloud concentration data from Synergy Research Group, and the undersea cable fault rate from ISCPC point toward a structural tension that resolves in one of two ways — and we won’t know which for a few more years.

Path one: the concentration continues, AI-managed routing reduces human configuration errors (BGP misconfiguration is almost always human error), and the internet gets quietly more resilient at the routing layer while getting more fragile at the provider layer. A world where the internet almost never has major outages — right up until a single dominant provider has a catastrophic day.

Path two: the geopolitical fragmentation accelerates. China’s great firewall is the proof of concept; Russia’s RUNET isolation project is the replication attempt; Brazil, India, and the EU are all developing varying degrees of “network sovereignty” infrastructure. By 2028–2030, the organizations best positioned won’t be the ones with the most cloud vendors. They’ll be the ones that built genuine network-layer resilience — multi-homed BGP, regional failover, local processing where possible — before the fracture lines hardened into walls.

The internet won’t collapse overnight. But the internet as a single, unified global network? That’s already eroding. The question worth asking isn’t “what if it all goes dark” — it’s “when the network splits along political lines, which side of each split will your operations be on?”