128 Banks Shared a Single Point of Failure. Bitcoin Was Built Without One.

One shared computer center took 128 savings banks offline for five and a half hours. The same pattern runs through rail, air traffic, power grids and payments. Bitcoin's own stress tests show a network that keeps running.

AnalysisBlock · 975,39619 min read
AI-assisted

AI helped with research and writing. The editor chose topic and angle, decided what stays in and approved the article. Sources at the end.

How we work

This is analysis. It interprets events and their context, and it is not financial advice.

On November 6, 2026, 128 Japanese savings banks lost their branches, ATMs, online banking and transfers at the same moment, for five hours and 38 minutes. The cause was not 128 separate faults but a single point of failure, one shared computer center whose equipment broke. In our reading, that is how modern systems usually break, in banks as much as in rail networks, air traffic control and power grids: through one part that everything else quietly depends on.

Bitcoin was built without such a part. It has been through real stress tests, from a mining ban in what was then its largest mining country to a winter storm that switched off a large share of its miners, and the network has never stopped. It does not depend on any single place, and many independent participants keep it running. While the 128 banks were offline, the Bitcoin network added 38 blocks.

What happened in western Japan on November 6

At 1:27 p.m. local time, an equipment fault hit the Shinkin Kyodo Center, the shared computer center of Japan's shinkin banks, according to a joint notice by the operator, the National Association of Shinkin Banks and the Shinkin Central Bank. The center first gave the start as 1:35 p.m. and later corrected it. Full service returned at 7:05 p.m. The operators said the cause was a failure of system equipment and not a cyberattack.

The outage hit 128 shinkin banks from the Tokai and Hokuriku regions westward, Jiji Press reported. Branch counters, ATMs including those in convenience stores, online banking, apps and transfers to other banks all stopped. The payment app PayPay could not take top-ups from the affected accounts. Transfers that missed their window were credited on the next business day, November 9.

The joint system serves 236 of Japan's 254 shinkin banks and, according to the operator, manages deposits of around 124 trillion yen. It runs two centers, one for eastern and one for western Japan. The eastern center kept working normally. By our count, that left 108 banks in the east unaffected. Another 18 shinkin banks do not use the joint system. The eastern center did not take over the western banks. The second center halved the damage. It did not prevent it.

Grid of 254 squares, one per shinkin bank in Japan: 128 orange squares for the banks served by the West Japan Center, which went offline on November 6, 2026, 108 dark squares for the banks served by the East Japan Center, which kept working, and 18 outlined squares for the banks outside the joint system.
One shared center, 128 banks offline. Sources: joint notice of November 6, 2026, Jiji Press, Shinkin Kyodo Center, our calculation.

According to Kyodo, the 128 banks had renewed their system through the center in mid-October, and the operator considers a defect from that update a likely cause. As of November 11, 2026, the final cause had not been published.

Japan has seen this pattern before. On January 29, 2026, a fault at a shared center cut off 140 of 143 credit cooperatives. In October 2023, a software change at Zengin-Net, the network that connects Japanese banks for transfers, broke the link for 10 institutions for two days and affected more than 5 million payments. It was the network's first outage with an impact on customers since it started in 1973.

The money was never gone in any of these cases. Balances were intact, and delayed payments went through later. That matters, and we come back to it below.

The same failure in trains, planes, power and payments

Banks are not special. The same structure appears wherever many users depend on one shared component.

Date System What failed Impact
April 3, 2022 NS, the Dutch national railway The IT system that plans trains and staff No NS trains across the Netherlands from the early afternoon until the next day
July 8, 2022 Rogers, Canada A configuration error during an update of the core network Over 12 million customers without mobile and internet service, Interac debit cards down nationwide for a day
October 8, 2022 Deutsche Bahn, northern Germany Two fiber cables of the GSM-R train radio, cut at two sites Trains in northern Germany halted for almost three hours
August 28, 2023 NATS, UK air traffic control The flight-plan system and its backup, on the same flight plan Over 700,000 passengers affected
July 19, 2024 CrowdStrike One faulty update for Windows computers About 8.5 million devices crashed
March 20, 2025 Heathrow Airport, London A fire at the North Hyde substation, one of the airport's three power feeds Airport closed for most of the following day
April 28, 2025 Power grid of Spain and Portugal Voltage control, followed by a cascade of shutdowns Total blackout, 30,000 to 35,000 rail passengers evacuated in Spain
October 20, 2025 Amazon Web Services, US-EAST-1 A race condition in the DNS automation of one database service Banks such as Lloyds and Halifax and exchanges such as Coinbase disrupted
November 18, 2025 Cloudflare One configuration file that doubled in size Error pages on many websites for several hours
June 23, 2026 Deutsche Bahn, all of Germany The GSM-R train radio All trains in Germany stopped for about two hours
November 6, 2026 Shinkin Kyodo Center, Japan Equipment at the western computer center 128 banks offline for five hours and 38 minutes

Two details stand out. The Rogers case shows that a single point of failure can sit outside the system that fails. Interac, Canada's network for debit cards and e-transfers, had nothing wrong with it. It ran on one telecom provider, and when that provider's network went down from 4:58 a.m. on July 8 until 7:00 a.m. the next day, debit payments stopped across the country, regardless of which internet provider a shop used. Interac added a second provider afterward.

The German rail case shows how deep such a point can sit. GSM-R is the radio that connects trains with control centers, and without it trains are not allowed to run. In October 2022, cutting two cables at two distant sites was enough to stop trains across northern Germany. On June 23, 2026, from about 10:30 p.m., a fault in the system halted every train in the country, including commuter lines and private operators. According to the federal government, the component involved had never before been part of a GSM-R failure, and there had been no comparable outage in the five years before.

Four patterns behind the outages

The cases look different, but the point that failed falls into one of four kinds, which our explainer on decentralization describes in more detail.

  • One location. A building, a substation or a region on which everything depends: the western center in Japan, North Hyde for Heathrow, the AWS region in northern Virginia.
  • One update channel. A single change that reaches many machines at once: CrowdStrike, Cloudflare and, possibly, the October update in Japan.
  • One copy of the same logic. A backup that runs the same software on the same data and fails in the same way. At NATS in 2023, main and backup system stopped on the same flight plan within about 20 seconds. At Visa Europe on June 1, 2018, a rare partial fault in a switch prevented the backup switch from taking over, and 5.2 million of 51.2 million card payments failed.
  • One supplier. A service that relies on one provider underneath: Interac on Rogers, banks and exchanges on one cloud, trains on one radio network.

Redundancy only protects against the failures it does not share. When a fire hit a data center in Pangyo, South Korea, in October 2022, KakaoBank kept working because it ran its own data centers elsewhere, while the Kakao services that depended on the burning building went down. When a battery fire hit South Korea's national government data center in September 2025, 647 government systems were shut down, and a storage system without a backup lost an estimated 858 terabytes of data. A copy in the same building is not a copy against fire.

Bitcoin's real stress tests: China 2021 and Winter Storm Fern

Bitcoin has its own version of these events, and they are well documented, because every block carries a timestamp and every difficulty change is public.

The largest came in 2021, when China banned mining. Global hash rate fell by 38 percent in June 2021, according to the Cambridge Centre for Alternative Finance. We looked at the block data directly. In the 2,016 blocks before July 3, 2021, a block arrived on average every 13.9 minutes instead of every 10. At block 689,472, every node lowered the difficulty by 27.94 percent, the largest downward adjustment in Bitcoin's history. In the following 2,016 blocks, the average was 10.5 minutes again. No one had to decide anything. The rule in the code did it.

A smaller test came with Winter Storm Fern in January 2026. US miners powered down while the grid was under strain, and the largest pool, Foundry, briefly lost about 60 percent of its hash rate. In the period that ended on February 7, blocks took 11.3 minutes on average, and the difficulty fell by 11.16 percent. When the miners came back, blocks sped up to 8.7 minutes, and on February 19 the difficulty rose by 14.73 percent. A storm in one country moved the whole network, and the network corrected itself in both directions within four weeks.

Two bar charts of the average minutes between Bitcoin blocks in each difficulty period. In 2021 the average rose to 11.9, 10.6 and 13.9 minutes during China's mining ban, the difficulty then fell by 27.9 percent, and the average dropped to 10.5 and then below 10. In early 2026 it rose to 11.3 minutes during Winter Storm Fern, the difficulty fell by 11.2 percent, and after an average of 8.7 minutes it rose again by 14.7 percent.
Average time between blocks per difficulty period, 2021 and 2025 to 2026. Source: mempool.space, our calculation.

The ordinary outages of the outside world barely register. By the block timestamps, Bitcoin added 38 blocks during the five hours and 38 minutes of the Shinkin outage, 64 blocks during the first ten hours of the Iberian blackout and 86 blocks during the AWS outage of October 2025.

Bar chart of new Bitcoin blocks during three outages: 38 blocks during the 5 hours and 38 minutes of the Shinkin outage, 64 blocks in the first 10 hours of the Iberian blackout and 86 blocks during the 14 hours and 32 minutes of the AWS outage, each close to the number expected at one block every 10 minutes.
New Bitcoin blocks during three outages, by block timestamp. Source: mempool.space, our calculation.

What if two-thirds of miners vanished overnight?

Blocks would keep coming, at first about one every 30 minutes, and the network would return to its 10-minute pace on its own. That holds even in the harshest version of the scenario, in which the mining power disappears for good, and that version is very unlikely. In 2021, miners who lost their sites in China packed up and moved. The US share of the hash rate rose from 16.8 percent at the end of April to 35.4 percent at the end of August, and Kazakhstan's from 8.2 to 18.1 percent, according to Cambridge. In 2026, the miners switched back on once the storm had passed. Power usually comes back, or the machines go to where it is.

To test the design against a sudden and permanent shock anyway, we ran our own calculation on the rule in Bitcoin Core. Every 2,016 blocks, each node compares the actual time those blocks took with the target of 14 days and adjusts the difficulty by that ratio, capped at a factor of four in either direction. Our assumption: two-thirds of all hash rate disappear at one moment and never return. Blocks then arrive every 30 minutes on average instead of every 10.

Shock arrives Blocks at 30 minutes Adjustment Pace afterward Back to 10 minutes after
right after an adjustment 42 days factor 3 10 minutes 42 days
a quarter into the period 31.5 days factor 2.5 12 minutes for 16.8 days 48 days
halfway through 21 days factor 2 15 minutes for 21 days 42 days
three quarters in 10.5 days factor 1.5 20 minutes for 28 days 38.5 days
right before an adjustment 42 days after a first, unchanged step factor 1, then 3 10 minutes 42 days

Blocks never stop. In the worst timing, they arrive about every 30 minutes for six weeks, and whatever the timing, the network is back at its normal pace within at most eight weeks, without anyone having to decide anything. Even in this scenario, the 338 minutes in which the 128 banks could not book a single transfer would have seen about 11 Bitcoin blocks.

The calculation is a model. It assumes that block times follow a random process and leaves out fees, the economics of the miners who stay and, above all, the miners who come back, which is what happened in both real cases.

Where this argument is weakest

The case for traditional systems is stronger than an outage list suggests, and it deserves to be stated properly. So do the answers.

Redundancy usually works, and nobody lost money. Japan's eastern center ran through the outage, KakaoBank kept running through the Pangyo fire, and most of the outages above lasted hours, not days. In none of the bank cases did balances disappear. Payments were delayed and then completed, in Japan on the next business day.

For a train, that is the end of the story. A railway that stops for two hours is a nuisance. Savings that cannot be reached are a different matter, because they are what people fall back on when something else goes wrong. A single point of failure is also a single point of attack. When attackers break into one, the funds held there can be taken in one stroke, as the hacks of Bitget and Binance showed. And when the institution itself fails, savings above the deposit guarantee depend on whatever is left, as customers of Ueno Bank in Paraguay learned when their bank was shut and the guarantee began paying out in steps.

Mining is concentrated in a few countries and pools. As of late September 2026, an estimated 65 percent of the hash rate ran in the United States, Russia and China, according to Hashrate Index, and six pools found more than 95 percent of all blocks in 2025, according to the researcher b10c.

These three countries are rivals, not partners, and a ban in one does not stop the others. When China banned mining in 2021, the network kept going. Miners go where electricity and taxes suit them, and where they settle, they become a large buyer of power that can switch off when others need it. In Ethiopia, the state utility earned 55 million dollars from miners in ten months, about 18 percent of its revenue, selling hydropower that its grid could not yet deliver to households. Its marketing director said the utility needed that money to build out the grid. When water ran short, the utility cut power to the miners first to protect households and industry. Pools work the same way. Home miners and large farms join the pool that suits them and leave when it no longer does. When GHash.io briefly passed half of the hash rate in June 2014, it pledged within weeks to stay below 40 percent.

The software is a shared copy. Most nodes run Bitcoin Core, the same "one copy of the same logic" pattern as at NATS. The record answers this better than theory. A bug in August 2010 that let one transaction create about 184 billion bitcoin was fixed within hours. In March 2013, two versions of the reference software disagreed about one large block at height 225,430 and split the chain for about six hours, until the largest pools switched back. A bug found in 2018 was fixed before anyone exploited it. And unlike a pushed update, no fix and no rule change reaches a node unless its operator installs it.

Bitcoin needs power and a connection. During the Iberian blackout, people in Spain without a battery and a signal could not send a transaction, while the network as a whole added 64 blocks in the first ten hours. The connection does not have to be the internet, which is simply the most convenient route. Blockstream has broadcast the blockchain by satellite since 2017, the TxTenna app began relaying signed transactions over radio mesh networks in 2018, and the Bluetooth app Bitchat passes them from phone to phone until one reaches a device that can hand them to the network.

Access often runs through exchanges. During the AWS outage, some Coinbase customers could not log in or trade for 3 hours and 17 minutes, according to the exchange, while their bitcoin stayed safe. That dependency is a choice. Bitcoin can be bought directly from other people and held in self-custody, where no exchange stands between the owner and the network.

Lost keys are lost. No one can freeze bitcoin in self-custody, and no one can restore it either. That is the price of not depending on anyone, and it is why owning bitcoin comes with responsibility, the theme of our essay on bank failures and self-custody.

In our reading, each of these points is real, and each has an answer built into how Bitcoin works. None of them is a part whose failure stops the network or changes the rules for everyone.

What would prove this reading wrong

A reading that cannot fail is not worth much, so here is the test for ours. It would no longer hold if a single event, one country, one pool, one data center or one software update, stopped Bitcoin's block production for a sustained period or forced a change to the rules against the will of the nodes, the opposite of how Bitcoin changes by consensus.

The closest real test came in 2021, when what was then the largest mining country banned mining outright. Blocks kept coming, the difficulty adjusted, and the miners moved. Because miners, from home operators to large farms, can relocate and switch pools whenever they choose, the bar for this reading to fail is high. Since the network started in 2009, nothing has cleared it.

What it means for someone holding money

A bank deposit is a claim on a bank, and the bank runs on systems like the ones above. That is usually fine, and on the bad days it means waiting.

Bitcoin removes the single point at the level of the protocol. Whether a person benefits from that depends on how they hold it. Bitcoin on an exchange inherits the exchange's points of failure, from outages like the one at AWS to security breaches like the Binance hack. Bitcoin in self-custody depends on the owner's keys and on a network that, so far, has never stopped. Self-custody best practices and the role of miners explain the two halves of that sentence. For the energy side of mining, including what a storm like Fern means for miners and the grid, see our analysis of wind-powered mining.

This is not a recommendation to buy, sell or move anything. It is a description of where the switches are.

Frequently Asked Questions

It is one part of a system whose breakdown stops the whole system. On November 6, 2026, that part was a shared computer center: when equipment there failed, 128 Japanese savings banks lost branches, ATMs, online banking and transfers at the same moment.

Because they all run on the same shared system. The 128 banks in western Japan are served by one center of the Shinkin Kyodo Center, and an equipment fault there stopped all of them from 1:27 p.m. to 7:05 p.m. local time. The operators say it was not a cyberattack.

Yes. Block timestamps show 38 new blocks, from block 974,760 to block 974,797, between the start and the end of the outage. Bitcoin also kept producing blocks during the Amazon Web Services outage of October 2025 and the Iberian blackout of April 2025.

No. Our calculation shows that even if two-thirds of all miners vanished overnight and never came back, blocks would keep coming, at first about every 30 minutes, and the difficulty adjustment would restore the 10-minute pace within at most about eight weeks. In practice, miners switch back on or move to new sites much sooner, as they did in 2021 and 2026.

No. Bitcoin on an exchange is a claim against that exchange, and it is only as available as the exchange's systems. During the AWS outage in October 2025, some Coinbase customers could not log in or trade for over three hours, while the Bitcoin network itself kept running.

Sources

  1. 1.National Association of Shinkin Banks, Shinkin Kyodo Center, Shinkin Central Bank — Joint notice on the restoration of the Shinkin Kyodo Center system failure and the resumption of shinkin bank services, November 6, 2026 (Japanese)
  2. 2.Jiji Press via Nippon.com — System failure hits 128 shinkin banks in western Japan
  3. 3.Jiji Press via Nippon.com — Shinkin bank system restored
  4. 4.Kyodo via OANDA — Shinkin system update in October and the 236 banks on the joint system (Japanese)
  5. 5.SBBIT — West Japan Center affected, East Japan Center operating normally (Japanese)
  6. 6.Shinkin Kyodo Center — Company profile (Japanese)
  7. 7.Kurashin — Notice on delayed transfers of November 6, 2026 (Japanese)
  8. 8.Jiji Press via Adnkronos — 140 Japanese credit cooperatives hit by system glitch (January 29, 2026)
  9. 9.paymentnavi — Zengin-Net final report on the October 2023 outage (Japanese)
  10. 10.The Japan Times — Payment network disruption restored (October 12, 2023)
  11. 11.Computer Weekly — Visa reveals rare datacentre switch fault as root cause of June 2018 outage
  12. 12.Asia Economy — KakaoBank on the Pangyo data center fire (October 17, 2022)
  13. 13.Korea Herald — 647 government systems down after the NIRS fire
  14. 14.DatacenterDynamics — 858TB of government data may be lost for good after South Korea data center fire
  15. 15.Tweede Kamer — Letter to parliament on the NS disruption of April 3, 2022 (Dutch)
  16. 16.CRTC — Assessment of Rogers Networks for Resiliency and Reliability Following the 8 July 2022 Outage
  17. 17.CBC News — Rogers service restored to vast majority of customers, Interac back online
  18. 18.Vixio — Rogers outage met with ministerial action, Interac adds backup supplier
  19. 19.echo24 — Deutsche Bahn statement on the cable sabotage of October 8, 2022 (German)
  20. 20.Landtag Nordrhein-Westfalen — Report on the attack on the GSM-R communication system on October 8, 2022 (German)
  21. 21.UK Civil Aviation Authority — CAP2993, Independent Review of NATS (En Route) Plc's Flight Planning System Failure on 28 August 2023
  22. 22.TechCrunch — Microsoft says 8.5M Windows devices were affected by CrowdStrike outage
  23. 23.NESO — North Hyde Review, Final Report
  24. 24.Passenger Terminal Today — NESO final report on the North Hyde outage that closed Heathrow
  25. 25.ENTSO-E — 28 April 2025 blackout in Spain and Portugal, Expert Panel final report
  26. 26.EFE via Prensa — More than 30,000 people evacuated from trains stranded by the blackout in Spain (Spanish)
  27. 27.Amazon Web Services — Summary of the Amazon DynamoDB service disruption in the Northern Virginia (US-EAST-1) Region
  28. 28.Coinbase — Retrospective: AWS outage impact and Coinbase next steps
  29. 29.TechInformed — Major AWS outage in us-east-1 disrupts Canva, Lloyds Bank, Coinbase
  30. 30.Cloudflare — Cloudflare outage on November 18, 2025
  31. 31.it-daily — IT failure halts German rail traffic for hours (June 2026, German)
  32. 32.Verkehrsrundschau — Radio failure brings rail traffic to a complete halt (June 24, 2026, German)
  33. 33.Deutscher Bundestag — hib 715/2026, rail traffic halted nationwide after GSM-R failure (September 16, 2026, German)
  34. 34.Bitcoin Core — pow.cpp, difficulty adjustment every 2,016 blocks
  35. 35.mempool.space — Blocks 974,760 to 974,797, mined during the Shinkin outage of November 6, 2026
  36. 36.mempool.space — Blocks 894,302 to 894,365, mined during the Iberian blackout of April 28, 2025
  37. 37.mempool.space — Blocks 919,913 to 919,998, mined during the AWS outage of October 20, 2025
  38. 38.mempool.space — Block 689,472, difficulty adjustment of July 3, 2021
  39. 39.mempool.space — Block 935,424, difficulty adjustment of February 7, 2026
  40. 40.mempool.space — Block 937,440, difficulty adjustment of February 19, 2026
  41. 41.The Block — Bitcoin records its largest difficulty drop ever
  42. 42.Cointelegraph — Bitcoin difficulty drops 11 percent during Winter Storm Fern
  43. 43.Bitcoin.com News — Bitcoin difficulty logs 11.16 percent reduction, largest drop since China's 2021 mining crackdown
  44. 44.Hashrate Index — Global hashrate heatmap update, Q4 2026
  45. 45.b10c — Bitcoin mining centralization in 2025
  46. 46.The Guardian — Bitcoin currency could have been destroyed by '51%' attack (June 16, 2014)
  47. 47.Ars Technica — Bitcoin pool GHash.io commits to 40% hashrate limit after its 51% breach (July 16, 2014)
  48. 48.Bitcoin Wiki — BIP 50, the March 2013 chain fork
  49. 49.Bitcoin Magazine — Bitcoin network shaken by blockchain fork (March 2013)
  50. 50.Cointelegraph Magazine — The 2010 value overflow bug and its fix
  51. 51.Bitcoin Core — CVE-2018-17144 full disclosure
  52. 52.Cambridge Centre for Alternative Finance — Bitcoin mining map update: US takes the lead after China's crackdown (data to the end of August 2021)
  53. 53.NTU Centre for African Studies — Cheap electricity sees crypto mining converge on Ethiopia (January 2025)
  54. 54.ForkLog — Bitcoin miners account for 18% of Ethiopia's electricity sales
  55. 55.Atlas21 — Ethiopia transforms hydroelectric power into Bitcoin (January 7, 2025)
  56. 56.Cointelegraph — Ethiopia cuts power to bitcoin miners amid hydro shortage
  57. 57.Hackaday — Radio free blockchain: Bitcoin from space (Blockstream Satellite)
  58. 58.Coin Center — Decentralizing Bitcoin's last mile with mobile mesh networks (TxTenna)
  59. 59.BeInCrypto — Bitchat allows Bitcoin transfers over Bluetooth