Lightning as a Retail Payment System
Lightning as a Retail Payment System

Lightning as a Retail Payment System


[This article has been reposted to the Galoy blog. It initially appeared on Medium on Jul 22, 2020]

To get insights about a potentially life-changing new technology, it is most useful to compare it with other technologies that already exist. Analogies involving things we already know help us make sense of something new. In this way, the comparison allows us to evaluate whether this new technology could have an impact on the world.

A fundamental insight is, in the past, walled gardens and closed networks ceased to be relevant, over a long enough period of time, when an open network, accessible to everyone, was alternatively available.

Our first attempt to make sense of Lightning compared it with Visa. These initial thoughts were previously outlined in this long document: Visa and Lightning, how do they compare?

After spending six months developing two alpha-stage wallets (one custodial, one non-custodial), we have developed an even stronger conviction about what Lightning will achieve¹.

Comparing Lightning and Visa was a good starting point. However, we now believe Lightning will have a much bigger impact than Visa alone. Lightning can replace most other “Layer 2” networks that facilitate USD payments. These include a variety of networks such as ACH, Mastercard, Star, Plus, Cirrus, PayPal, ApplePay, and many others.

One good analogy to help understand the impact of Lightning is to follow the evolution of information networks. A fundamental insight is, in the past, walled gardens and closed networks ceased to be relevant, over a long enough period of time, when an open network, accessible to everyone, was alternatively available.

Before diving in, we would like to thanks Max Webster at VersionOne for his valuable feedback. This article has been written by Nicolas Burtey and Chris Hunter from Galoy. We are building payment solutions using the Lightning network. We’ll cover more of what we do in future articles.

Current digital dollar system

To start, let’s look at how payment networks are built for the U.S. dollar (USD). The dollar is composed of a hierarchical payment system.

At the base layer, there is Fedwire. Fedwire is ultimately where all final settlement of US Dollars between various financial institutions occurs.

Above Fedwire, you have other networks. They include:

There are scores of payment networks. Each has specific technical complexities, participation requirements, and other unique attributes.

It always settles on a base layer

What does it mean for a network to be the base layer?

The base layer is where the final settlement eventually occurs. A base layer network is systemically important for a very specific reason. It is not only used to settle transactions between different parties who directly access this network. It also settles transactions that take place on other payment networks (typically after periodic netting occurs).

An example: Alice buys a coffee for $5 with an Ally card at “Bob’s coffee”, a merchant with a bank account at Citi. To settle this transaction, the money needs to move from Ally bank to Citibank.

Small banks typically have an account with larger banks. In the instance where Ally, the bank, has an open corporate account at Citi, this transaction is easily settled when Citibank simply updates its internal ledger.

But let’s say Ally does not have an open corporate account at Citi. Moreover, let’s presume that various customers with accounts at these two banks conduct many transactions during the course of any given day. In this instance, there are two basic scenarios:

  • The two banks, Ally and Citi, can settle each and every individual transaction. This is called gross settlement.
  • Alternatively, the two banks could simply “net out” the sum total of all transactions completed during a given day. In this instance, they would have only one transaction to settle all payments for that period of time. This is the concept of net settlement. This is what is practically used by most “Layer 2” networks.

Let’s presume that during the course of one day the following happens:

  • Ally customers owe a total of $500, across 14 different transactions, to Citi customers
  • Citi customers owe a total of $300, across 11 other transactions, to Ally customers

In this scenario, these 25 transactions in aggregate can be net settled with a single payment of $200 from Ally to Citibank.

It does not matter whether the customers of Ally Bank and Citibank used Visa, PayPal, or ApplePay to conduct their transactions. When the two banks settle amongst each other, they do not use any of these named networks. Instead, Ally Bank and Citibank conduct the net $200 final settlement on the base layer network. For all USD, this final settlement occurs at Fedwire².

Net settlement requires some trust. As until the net settlement transaction occurs, one bank owes money to the other bank.

Why is Fedwire the base layer? Because Fedwire, an arm of the US Federal Reserve, is the entity that settles all US dollars. As most US banks have an account at the Fed, a Fedwire transfer is essentially a ledger entry between two bank accounts at the Federal Reserve bank. Fedwire payments are instant and final.

For a bank, having a dollar at the Fed is also the best dollar possible. How can one dollar be more valuable than another dollar? The answer, in one word: risk.

Instead of settling at the Federal Reserve, a bank could have an account directly with other banks. However, there is always some risk of institutional failure (see: Lehman brothers). This risk of insolvency doesn’t exist with money held at the Fed. As Chairman Greenspan famously said, there is zero probability of default, as the Fed can always print more money.

Why are there so many dollar payment networks, and not just one?

TL/DR: Because the Fedwire doesn’t scale.

For people with Bitcoin experience, when thinking about scaling, they typically think about constraints related to the number of transactions per second.

Interestingly, the Fedwire has a similar constraint. It does not handle significantly more transactions per second than the bitcoin network! However, unlike Bitcoin, Fedwire’s inability to scale transaction throughput is not due to a technical limitation.

The Fedwire is a permissioned system. Only a subset of entities can access it. Accordingly, the Fedwire throughput limitations stem from controls on which entities and organizations can utilize the system directly.

As individuals, we can not have access to the Fed. As a company? we can’t either. Unless we’re a company with a banking license… ie: a bank.

Banking is the most highly regulated industry on earth. Receiving a bank license is time, capital and resource-intensive. And for good reason. It comes with a lot of responsibilities, as banks take custody of money, and also have a license to “print money” in a fractional reserve system.

Fedwire is the ultimate and final US dollar settlement system. But this network is only directly accessible by companies with a banking license.

To facilitate payments between individuals and companies, additional networks, or layers, have been built on top of Fedwire.

You can think of these payment networks as “Layer 2” solutions, built on the “Base Layer” of Fedwire

This is why other payment networks have blossomed in various forms during the past 50 years. The logos of just a few of such payment networks are shown below.

A non-extensive list of private USD payment networks.

With that in mind, there is one element that all of these payment networks have in common. Whether it is Visa, ACH, PayPal, SWIFT, or American Express, all US dollar payment networks rely upon Fedwire for final settlement. Moreover, you can think of these payment networks as “Layer 2” solutions, built on the “Base Layer” of Fedwire.

Regulation has created a barrier to access the best USD settlement system. As mentioned, only certain licensed banks can access Fedwire. Accordingly, other permissioned networks — from credit cards and ATM networks to physical checks and mobile payment apps — have emerged to mitigate this scalability issue.

Despite the maturity of the USD payment industry, the market remains extremely fragmented. Dozens of different “walled garden” payment platforms are widely used today.

Why is this?

There is a simple, fundamental answer to this question. All of the existing “Layer 2” USD payment networks are commercial in nature. They are designed to benefit one specific company, or a consortium of companies. Given this dynamic, there is little incentive to create compatibility amongst the various networks.

Additionally, each of these payment networks is closed. They all have very specific participation requirements. Oftentimes these requirements are not even publicly disclosed.

It is similar to private intranets. From an era where Internet had not yet reached mainstream adoption. This raises the fundamental question. If it was truly possible to have an open payment network, would these dynamics have been different?

How does BTC compare to Fedwire as a base settlement layer?

The graph below shows a comparison of the number of annual transactions for Bitcoin and Fedwire, from the inception of the Bitcoin network in 2009, through the end of 2019. We can see the transaction values for these two “Base Layer” networks are getting closer to each year³.

Critics of Bitcoin are swift to highlight its infamous “7 tx/sec” limitation. But do these critics know that the throughput of Fedwire is not materially different?

Comparison of annual transaction volume between Bitcoin and Fedwire.

Fedwire use case

Putting transaction volume aside, it may prove more instructive to explore the use cases for using the Fedwire base settlement layer.

These use cases almost always involve exchanging a large amount of money between two parties, typically companies or governments. Examples include:

  • A company paying for the goods of another company. You can think of Apple Inc paying supplier Foxconn for the manufacturing of its devices. This type of transaction involves billions of dollars every month.
  • It could also be a government, like the city of San Francisco, paying for a private company for an infrastructure project (e.g., renovating the Bay Bridge).
  • When a new owner is buying a house, a wire transfer, with its finality guarantee, would also be very appropriate.
  • For settling “Layer 2” networks. This is where things become interesting. Individual retail customers use payment networks (such as credit cards, or P2P platforms) for smaller transaction values. That said, the banks and owners of the payment networks (Visa, Venmo, etc) need to ultimately net settle payment at some stage. They do such on Fedwire.

Finally, it is also worth mentioning that many Layer 2 networks, such as Visa, operate 24/7. On the other hand, Fedwire only operates during limited hours on weekdays. Given this, Layer 2 payment networks not only enable smaller transaction sizes, but they also allow spending when the Fed itself is closed.

Retail payment uses Layer 2

In brief, “Layer 1” or “Base Layer” payment protocols do not scale. With Fedwire, this stems from a complex regulatory framework. In the case of Bitcoin, the inability of the base layer to scale stems from a technical⁴ constraint: the block size limit.

These scaling limitations in the base layer do not prevent massive real-world adoption of these protocols. Even with the USD as the most commonly used currency worldwide, having Layer 1 limitations are manageable, so long as the Layer 2 solutions provide the scalability required.

Who currently interacts with Layer 1 for USD (e.g., Fedwire)? Banks. Companies. Governments. And, very infrequently, individuals. This correlates with the average transaction size on Fedwire, which is above $2 million.

Given these dynamics, consumers do not directly rely on the Layer 1 protocol for daily payments. They use Layer 2 solutions.

In the world of USD banking, this dynamic is easy to observe. Consumers do not have an account at Fedwire. Instead, they have an account at a commercial bank. This commercial bank account serves as a custodial wallet, at least for daily transactions.

If we apply the framework of the digital dollar to BTC, where does it lead us?

As the usage of the protocol takes off, fees on Layer 1 increase. To the point where the majority of retail consumers would not want to transact on it.

Let’s assume on-chain fees on the Bitcoin Base Layer are $100/transaction. In this instance, it will be impractically expensive for most consumers to individually manage a well-connected Lightning node with a dozen channels.

By way of contrast, let’s imagine a regional commercial bank managing a Lightning node. If we assume this bank is managing money for 10,000 people, then $100 per on-chain transaction is bearable: in aggregate, maintaining those channels will cost a penny per customer.

In much the same way retailers use Visa and ACH for daily dollar payments, they will soon use Lightning in a world where BTC adoption is taking off.

USD has a lot of Layer 2 Networks

There are dozens of Layer 2 networks enabling dollar transactions in the United States, and hundreds of different payment networks worldwide.

All of these payment networks are walled gardens. Each of them targets a specific use case. They are not extensible and evolve very slowly. Key considerations and feature include:

  • Fee Structures: Because these networks are walled gardens, fee structures often incentive a multi-year lock up. For a bank, it is not worth the cost of initial implementation and integration, unless they plan to access the network for an extended period of time.
  • Incompatibility: There is no compatibility between these various payment networks. Let’s take the credit card industry as an example. Mastercard is not compatible with Visa. Which is not compatible with American Express. All of these networks may use some version of the ISO8583 protocol (which is analogous to a binary XML). However, these card network operators do not necessarily agree on similar tags to use in classifying and identifying transactions.
  • Legacy Issues: The reason we have so many payment networks is also historic. In the early days of ATM deployments, each US State or Region had its own network. We can see this evidenced through legacy brands such as Star, Plus, Maestro, and Cirrus. Each of these local, independent networks are still being widely maintained today — even as brands like Visa and Mastercard have become predominant.
  • Counterparty risk: Some networks, such as CHIPS, require accounts to be pre-funded (similarly to Lightning). As a result, they are less subject to counterparty risk compared to other networks that must have trust in the other party. Many networks mitigate this issue with a clearinghouse, which shifts the counterparty risk to the network operator.

Have you wondered why ATM fees have risen to be typically between $3 and $5 per transaction, simply to obtain cash? All of those independent ATM networks need to be maintained. It costs a lot of dollars to do such.

List of networks an ATM might be connected to.

The US dollar payment network industry is quite robust. It began flourishing in the 1950’s and 1960’s. That said, there remains a tremendous amount of fragmentation and friction in this long-standing, mature industry.

Whether a consumer is using a credit card, accessing an ATM, or completing a wire transfer, there is always a cost involved going to the network operator and different third parties. And there is typically little-to-no competition on the fees. In fact, in the case of the card networks, there was a competition… to increase interchange fees!

By way of contrast, we now have a wholly new alternative available. An Internet-based network that provides true Money over IP (MOIP), with a virtually instantaneous settlement and near-zero cost per transaction. Irrespective of the value of the transaction itself. This is the Bitcoin Protocol (Layer 1), in concert with Lightning (Layer 2). Today, in the year 2020, virtually anyone worldwide can take part in this network.

Lightning is being developed in an open environment, relying on open-source code. Given this, there are already thousands of developers experimenting with this exciting new technology. Unlike closed garden networks, compounding innovation will contribute to the general network effect of lightning.

Comparison with the early days of the Internet

We like to compare payment networks today with information networks 25 years ago. This video with Bill Gates is a helpful reminder of the state of the Internet in the 90s. No one will challenge the fact that the Internet has prevailed today. But it was certainly a contrarian view in 1995!

The Internet has succeeded over the alternative — “secured” private Intranets and AOL-style closed information networks. Why did the Internet “win” over closed networks? Because it is open to anyone.

During the past 25 years, millions of developers have made contributions to make the Internet better. By launching new products. By making the underlying protocols more scalable and secure. This compounding effect of innovation leading to the success of the Internet was simply not possible on private networks, where the layers of corporate and government bureaucracy limit innovation.

We witnessed a massive consolidation effect from many intranets and private networks to virtually every application running on the public Internet. A similar consolidation will soon happen with payment systems.

This system is so balkanized, financial institutions need, on top of those payment networks, a paid messaging system (SWIFT) — simply to be able to allow these disparate networks to be able to “talk” to each other!

This begs a question: does it make sense for a bank to run dozens of payment networks?

How long will the existing walled-garden payment networks subsist? Particularly when there are hundreds of such networks worldwide, each incompatible between the other? This system is so balkanized, financial institutions need, on top of those payment networks, a paid messaging system (SWIFT) — simply to be able to allow these disparate networks to be able to “talk” to each other!

Fun fact: Visa (a $420B company) still relies on mainframe computers, and a messaging system with a standard defined in the eighties! (ISO8583:1987). Because so few developers really know about this standard, there is a huge cost to access the network. This is a fact that Visa is monetizing, and payment companies like Stripe use as an edge.

Compare this to lnd or c-lightning using REST api and GRPC. Millions of developers know how to interact with these endpoints instantly. Additionally, the code is free and open-source, so no approval from a bank or network provider is needed to create a proof of concept or even launch a product.

The adoption of lightning will soon grow rapidly. We can envision this growth by highlighting just one company: Square. The company has been a vocal proponent of both Bitcoin and Lightning.

What do you believe will happen when Square enables their millions of merchant terminals to accept Lightning payments? Jack, the CEO of Square (and Twitter), has been vocal saying this is not a question of if, but a question of when.

Which paradigm will thrive over the long-term — closed payment networks, or open payment networks?

Bitcoin is an open settlement network. Lightning is a low cost, instant open payment network. We believe it is a given that massive consolidation will take place during the next 10 years… with 10x the pace of innovation.

Current metrics

How does Lightning currently compare with existing fiat payment systems or even Bitcoin?

Let’s first look at how the Layer 2 system like ACH and Visa compared to Fedwire and Bitcoin.

Annual transactional volume between Bitcoin, Fedwire, Visa, ACH. Y-axis is logarithmic.

We see that ACH has about two orders of magnitude more transactions than Fedwire. Visa has about three orders of magnitude more.

Where does Lightning fit into this picture?

Privacy is at the root of Lightning which is a requirement for a retail payment network to succeed.

It probably has a lower transaction volume than Bitcoin does thus far.

Bitcoin has been running for about 11 years now and has reached a state of maturity and stability that is actually backing up its value.

Comparatively, Lightning is much more nascent. While the white paper was released in 2016, the first alpha version was only deployed to mainnet about two years ago. Two Bitcoin soft forks had to happen before Lightning transactions were possible: the addition of CSV and CLTV.

Lightning is now at a “beta” stage. But it is progressing very rapidly.

Yet, what is even more insightful, is the fact that nobody knows what the lightning throughput is. And this is a feature, not a bug! The only metrics a lightning node can know for sure is:

  • The value of payments the node is sending and receiving
  • The value of payments being routed through to other nodes

Privacy is at the root of Lightning which is a requirement for a retail payment network to succeed. It builds on Bitcoin which does not optimize for privacy⁵ but transparency and robustness.

Although it is not top-of-mind for average citizens just yet, it is likely that people will become increasingly concerned with both companies and governments surveilling their everyday financial transactions. We think the option of a digital alternative that offers a good UX will help.

This means that, at some point, lightning could surpass the throughput of Bitcoin, ACH and Visa combined — without data nerds being able to notice it!

Does the price of BTC need to become stable, before Lightning adoption can take off?

It doesn’t. It is possible to use Lightning solely as a means of payment. Irrespective of whether the price of Bitcoin is stable or volatile.

Lightning is ready to be used as a means of payment. Today.

Lightning can succeed, even if people do not necessarily view BTC as a proper store of value. Why is this so? There are several reasons:

  • Lightning is instant
  • There are many derivatives readily available for Bitcoin
  • Lightning offers more competitive fees than transactions on the Bitcoin base layer

In other words, Lightning is ready to be used as a means of payment. Today.

When transacting with BTC at the base layer, it is more difficult to hedge efficiently than it is for Lightning. This is because transactions on the base layer can take anywhere from 10 minutes to 1 hour (or more) before they can be considered to be confirmed.

From Rene Pickhardt Twitter feed. How long can it take to confirm the next block? Sometimes more than an hour.

In a similar vein, exchanges often require 3–6 blocks of confirmations, before allowing clients to trade newly deposited bitcoin on their platform. In a market that trades 24×7, and where the market can move at any time, minutes of extra delay can have a material impact.

Lastly, the higher fees associated with on-chain BTC transactions create a natural incentive to only conduct higher value transactions on the base layer.

Compared to Lightning, it is harder to use Bitcoin solely as a means of payment. This stems largely from latency issues and fee structures. All of which make it difficult to hedge bitcoin dynamically.

For instance, a bank might be waiting for a payment worth $10m in BTC, and wants to hedge against the price fluctuations of BTC.

If the bank sells a short $10m worth of BTC on a future market such as the CME, there are a number of issues to consider. First, there is a risk of slippage, because there is a large monetary value involved. Additionally, if the BTC price is going up, but the transaction did not go through⁶, then the short position will have a negative P&L. There would be no opposite position to offset the loss.

For the bank, it is possible to see that the UTXO has been sent. However, because it is not known when a UTXO will be confirmed, it is harder to hedge.

Paying higher fees to make sure the transition will be included in the next block helps. But a race to higher fees is not beneficial for a payment protocol targeted to individual consumers.

With Lightning, it’s a lot easier to properly manage a delta-neutral position.

First, because instead of receiving large payments, Lightning incentivizes small payments. And as those small payments can sum up to large payments, because the large payment would be divided into small payments, it’s easy to gradually hedge the position, with little to no slippage.

Second, unlike with Bitcoin, when a Lightning node receives a payment⁷, the transaction is immediately settled. There is no inherent timing variability on when a transaction settles⁸.

Non-custodial lightning is not for the unbanked/underbanked

Let’s play out a thought exercise. Assume that Bitcoin is taking off, and has an equivalent role as the Fedwire for USD. In this instance, being able to interact with Layer 1 Bitcoin, and therefore Lightning, as an individual, will be reserved for a limited group of people. This might be a contrarian view for hard-core Bitcoiners. But we believe it is the most likely outcome.

In this scenario, transacting at Layer 1 would become costly over time. Even if this cost may not be prohibitive for many US citizens, it would be for most people in developing countries. We can assume that having just a dozen lightning channels open will cost hundreds or thousands of dollars per year at some point.

Despite the anticipated Layer 1 costs and restrictions, access to an open, global network would still benefit individuals tremendously, especially the ones in developing countries.

Because a bank will provide financial services for many users, their ability to offer an economy of scale in managing liquidity efficiently will make this option appealing for most individuals. Most would prefer to interact with a bank they know and have features like social recovery if they forget their passwords, or provide a lending and borrowing offering within their community.

Peter McCormack has done an interview that highlights the use of a custodial wallet over non-custodial in El Salvador with BlueWallet and Wallet of Satoshi. In a country where the average salary is $250/month, the cost of an on-chain transaction is too high, even now, in a low fee environment

Open Node data also shows that Blue Wallet and Wallet of Satoshi, the two easiest to use lightning custodial (non-regulated) wallets are currently the top lightning wallet destinations for Lightning payment.

As, in all likelihood, the cost of participating in Lightning will still be an order of magnitude less expensive than the traditional banking system, even as Layer 1 Bitcoin costs continue to increase. Lightning and Bitcoin are leveling the playing field. We believe the centralization that has happened in the banking industry⁹ could reverse. Union banks would not be as much as a disadvantage compared to big banks if they could just as easily be part of this network. A group of friends, a small village, or a community could create their own bank. This is a model of a modern Union Bank.

At this time, it is technically difficult to deploy a Lightning node that can act as a bank for a community. Even with Blue Wallet open-source LnbHub as a multi-user backend, the features set is still lacking compared to what a typical bank would provide to its customers. But we expect this open-source infrastructure to blossom in the upcoming years.

From Union Bank in emerging markets to Amazon

The massive benefits will not be exclusive to developing nations. As an example, let’s turn our attention to Amazon. Per the most recently filed 10K, Amazon had $280B in sales during 2019. Although this is not disclosed, it’s fair to assume that the majority of payments were handled via debit or credit cards. With 2%+ of interchange for every payment processed by card on its platform, Amazon incurred multiple billions of dollars of fees, simply to accept payment from customers.

Imagine an innovative, new payment solution. One as international as the Internet itself. And with a fraction of the fees of debit or credit cards. This new solution would certainly become of interest to all merchants — from Amazon all the way down to your local coffee shop.

This innovative new system will consequently enable novel participants in the network to earn fees, in exchange for providing valuable services.

The text above highlights solely the cost benefits associated with this new system: Lightning. However, the network affords many additional benefits that we have not even touched upon — from privacy and censorship resistance to finality of financial settlement on the Base Layer in minutes, not days. Novel participants in this new network will capture massive value for providing services beyond simple fees for processing transactions.


During the past 50 years, a multitude of US dollar payment networks has blossomed — from credit cards and ATM networks to peer-to-peer platforms and ACH. All of these networks are closed walled gardens. Moreover, one can think of them as Layer 2 solutions. Built on top of Fedwire, the base settlement layer for all dollars.

Similarly, we expect that Layer 2 solutions for Bitcoin — most notably Lightning — will flourish in the coming years. However, unlike traditional USD payment networks, Lighting is open and permissionless. Accordingly, we anticipate that the pace of innovation and adoption will be staggering during the next decade.

The post is the first in a series. In our subsequent posts, we will explore many more detailed issues. This will include Lightning in a multi-currency environment, an analysis of key factors supporting the growth of the network, and an exploration of how the protocol could be a valuable bridge between fiat currencies.

Thanks to Hunter.

[1] Ping us if you would like to get the TestFlight or Android alpha of the Lightning wallet created.

[2] The concept of net settlement expanded for more than 2 parties. If Alice owes $2 to Bob, and Bob owes $2 to Charlie, then Alice could pay $2 to Charlie to settle both transactions.

[3] Comparing transactions from Fedwire to Bitcoin is not an Apple to Apple comparison. Unlike a Fedwire transaction, which is bilateral (A pays B), a bitcoin transaction may involve more than 2 parties. It’s hard to know if a particular output is a change address, or involves another party, but the point to have in mind here is that bitcoin transactions show the lower bond of parties involved, where it’s an exact number for Fedwire. So in reality, the Bitcoin throughput may be even closer than the Fedwire throughput.

[4] This technical constraint for decentralization (only need limited bandwidth + CPU to participate in the network) has also economic impacts by incentivizing a healthy transaction fees market.

[5] A network, such as Ethereum with an account-based system, where there is no privacy really possible will not make a good payment system. (note: this doesn’t mean it can’t be good for something else).

[6] Reasons for a failed transaction could include: (i) the sender created another transaction with the same UTXO, whether intentionally or not, and (ii) the transaction was rejected from the mempool, typically following a large fee spike.

[7] Technically, the receiver has certainty about the settlement of a payment, in a non-revocable way, when it received the preimage of the hash of the invoice associated with the invoice.

[8] At the current stage of development, payment can still be temporarily stuck for a period of minutes to days if a node in the path has shut down while it was routing a payment. Fortunately, features like probing, or stuck-less payment will provide robustness that will make payment succeed with the same or higher ratio than card payment today.

[9] In the US, where the banking industry might be more distributed than most other countries, there were, per FDIC data, 9904 banks in 2000. Now there is 5116. This is a 49% reduction in 20 years.

You may also like