Seven internet cables were cut at once – Bitcoin went largely unnoticed, but researchers discover the real sticking point

When an undersea failure off the coast of Ivory Coast severed seven submarine cables in March 2024, the impact on the region’s internet caused the IODA severity score to exceed 11,000.

In the case of Bitcoin, the global impact was negligible. The affected region hosted approximately 5 nodes, representing approximately 0.03% of the network, and the impact was within the normal range of -2.5%.

There was no price movement after that. No consensus breaking occurred.

A new study from Cambridge, covering 11 years of Bitcoin network data and 68 verified cable failure events, finds that network disruptions due to submarine cable failures have historically been minimal.

In contrast, concerted pressure on a small number of hosting networks can disrupt visible nodes much more effectively than random infrastructure failures.

A surprising result: China’s mining crackdown and adoption of global censorship-resistant infrastructure may have inadvertently pushed Bitcoin toward a more robust topology.

While Tor has long been understood as a privacy tool, it now functions as a structural resiliency layer. And most Bitcoin nodes run on top of that.

Empirical record contradicts fear

Cambridge researchers Wenbing Wu and Alexander Neumuller collected a dataset spanning 2014 to 2025. 8 million Bitcoin node observations, 658 submarine cables, and 385 cable failure events cross-referenced with outage signs.

Of these 385 reports, 68 matched verifiable disruptions, and 87% of verified cable events resulted in less than 5% node change. The mean impact was -1.5% and the median impact was -0.4%.

The correlation between node disruptions and Bitcoin price was virtually zero (r = -0.02). Cable failures that dominate local headlines routinely go unrecorded on Bitcoin’s decentralized network.

In this study, we model Bitcoin as a multiplexed network. That is, physical connectivity through 354 undersea cable edges connecting 225 countries, infrastructure routing through autonomous systems, and a Bitcoin peer-to-peer overlay.

If cables are randomly removed, the critical failure threshold for disconnecting 10% or more of nodes is between 0.72 and 0.92. Most cables between countries would have to fail before Bitcoin would experience significant fragmentation.

Where are the real vulnerabilities?

Targeted attacks work differently. With random cable removal, 72% to 92% of the cables must be removed to reach the 10% node disconnection threshold. The target for high intermediate cables drops to 20%.

The most effective strategy, which targets autonomous systems higher in number of nodes, reaches a threshold when only 5% of the routing capacity is removed.

The authors frame this ASN-targeted scenario as “a hosting provider shutdown or coordinated regulatory action rather than a physical cable disconnection.” This model identifies the top networks: Hetzner, OVHcloud, Comcast, Amazon Web Services, and Google Cloud.

The March 2026 Bitnodes snapshot confirms this pattern. Of the 23,150 reachable nodes, Hetzner hosts 869, Comcast and OVH each host 348, Amazon 336, and Google 313.

This is not a claim that “5 providers can destroy Bitcoin.”

Since Tor hosts the majority of the network, most nodes will remain operational even if Clearnet is completely removed. However, the coordinated action identifies locations where connection shocks and propagation interruptions may occur that would not occur with random cable failures.

Recent cloud disruptions illustrate the category of risk. Amazon attributed the March 2026 outage to a software implementation failure. Another report described the chaos in the AWS Middle East after an attack on its data center.

Although these did not have a significant impact on Bitcoin, they show that correlated hosting failures are real rather than theoretical.

Tor as a structural resilience

Bitcoin’s network structure has changed dramatically.

Tor adoption grew from almost zero in 2014 to 2,478 nodes (23%) by 2021 and 7,617 nodes (52%) by 2022. March 2026 shows 14,602 Tor nodes out of 23,150 reachable nodes, which is 63%.
This surge coincided with censorship events such as the Iranian government shutdown in 2019, the Myanmar coup in 2021, and the Chinese mining ban in 2021.

Node operators migrated to a censorship-resistant infrastructure without coordination, suggesting adaptive self-organization.

Tor has its challenges. Most Bitcoin nodes are currently located in unobservable locations.

The authors address this issue by building a four-layer model that incorporates the Tor relay infrastructure as a separate network layer. A Tor relay is a physical server with a known location.

The authors used consensus weight data from 9,793 repeaters to model how a cable failure that disrupts connectivity between countries could take repeaters offline.

This discovery overturns expectations. The four-layer model consistently produces higher critical failure thresholds than clearnet alone, increasing from 0.02 to 0.10.

Most of the Tor relay consensus weight is concentrated in Germany, France, and the Netherlands, where cable connections are abundant. Even if a cable failure disrupts connectivity with surrounding countries, the relay capacity of these well-connected countries will not be reduced.

An attacker would need to remove even more infrastructure to disrupt both clearnet routing and Tor lines at the same time.

”)

china effect

Bitcoin’s resilience coincided with the peak of mining concentration in 2021, hitting a low of 0.72.

According to Cambridge data, in 2019, 74% of hashrate was in East Asia. Geographical concentration of nodes reduced Clearnet’s resilience by 22% from peak to trough from 2018 to 2021.

The recovery in 2022 was rapid. After China’s mining ban, the threshold jumped to 0.88 as infrastructure became more decentralized. Tor adoption also accelerated at the same time.

Although the authors avoid single-cause claims, regulatory pressures forced geographic redistribution and encouraged the adoption of censorship-resistant infrastructure, both of which improved robustness.

Some of the apparent concentration is a measurement artifact. As Tor’s adoption increased, Clearnet samples became concentrated in a few locations. The Herfindahl-Hirschman index rose from 166 to 4,163, but Hetzner’s actual share fell from 10% to 3.6%.

This consolidation reflects a change in sample composition rather than true centralization.

Clouds are the real risk

Concerns about the security of submarine cables will continue to grow. The Baltic Sea study, the European Commission’s security toolbox and reports on Russian infrastructure all show that geopolitical insecurity persists.

In the case of Bitcoin, historical data suggests that most cable events are noise.

A practical infrastructure question is whether policy adjustments, cloud outages, or hosting limitations can cause connectivity shocks at the autonomous system layer.

Scenarios targeting ASN operate at 5% of routing capacity. This is a threshold for significant disruption to reachable clearnet nodes, rather than a consensus failure.

Tor’s majority share provides a lower bound under extreme scenarios. Protocol-level mechanisms excluded in the study, such as block relay networks, compact block relays, and blockstream satellites, add a layer of resiliency that the model does not capture, resulting in conservative estimates.

Bitcoin is not as fragile as critics imagine, but it is also not disconnected from its infrastructure.

The network exhibits gradual degradation under stress rather than catastrophic collapse. Censorship pressure has driven the implementation of infrastructure that strengthens resilience to adjustment risks.

Threat models featuring cable-cutting submarines overlook the more familiar challenge of a small number of networks that can cause temporary disruption through coordinated action without dramatic undersea operations or acts of war.