Bitcoin Lightning Network: Instant & Cheap BTC Payments

Introduction

Forwarding 100,000 satoshis (sats) across Lightning costs about 10.5 sats per node at the median fee, River estimated $1.17 billion of Lightning volume for November 2025, and public capacity sits 45.3% below its December 2025 peak. All three statements are true on the same network. This guide explains how channels and hash time-locked contracts (HTLCs) move a payment without waiting for a block, why inbound liquidity decides who can receive, and when a fee beats the base layer. Capacity, node and fee data were pulled on 4 November 2026, with the caveat that private channels are invisible. It also compares custodial and self-custodial wallets, and it lists the limits that remain after the 2026 security incidents and the August channel closures.

What Is the Lightning Network, and What Does It Change About Bitcoin Payments?

The Lightning Network is a layer 2 payment network built on Bitcoin. Users lock coins into two-party channels, pay each other off-chain, and settle on the blockchain only when a channel opens or closes. On 4 November 2026 its public graph held 28,939 channels and 3,203 BTC (mempool.emzy.de, 2026).

The base-layer limit

Bitcoin adds a block about every ten minutes, and each block holds roughly 1,000,000 virtual bytes (vB) of transactions. A payment waits for a block and then competes with other senders for that space. Fees on 4 November 2026 sat at 1 to 2 satoshis per virtual byte (sat/vB), so a 141 vB payment cost about $0.11 to $0.22 (mempool.space, 2026). That price suits a salary transfer and strains a coffee. Joseph Poon and Thaddeus Dryja proposed the fix in their 14 January 2016 whitepaper. Repeated small payments move into channels, so thousands of payments need only two on-chain transactions (Poon and Dryja, 2016). Readers new to the base layer can start with how Bitcoin works ↗.

Where the network stands on 21 October 2026

The public network had 15,899 nodes holding 3,203 BTC, about $251 million at $78,269 per bitcoin on 4 November 2026. That price is 37.9% below the 6 October 2025 high of $126,173 (mempool.emzy.de; CoinPaprika API, 2026). That capacity is 45.3% below the peak of 5,851 BTC reached on 28 December 2025. Channels are 43.9% below 51,584, the June 2024 high of the series used here, and 65.1% below the all-time high of 82,825 in March 2022 in mempool.space's own history. Public channels lock about 0.016% of the 20.11 million bitcoin in circulation (CoinPaprika API, 2026). Growth in the number of channels has reversed. Later sections cover what these counts miss, what usage data exists, and what drove the 2026 decline.

Lightning's design choices begin with the channel itself, and the channel is where every payment starts.

How Does a Payment Channel Move Bitcoin Without Using the Blockchain?

A payment channel is a shared Bitcoin balance between two parties who change its split by exchanging signed messages. Only two events touch the blockchain: the transaction that opens the channel and the one that closes it. Every payment between them happens off-chain, so none waits for a block.

Opening: one on-chain funding transaction

Two nodes open a channel with a funding transaction that pays into a 2-of-2 multisignature output. Both parties must sign to spend it (BOLT 3). A funding transaction with one native SegWit input, the channel output and a change output weighs 152.5 vB. That costs 153 to 305 sats at 1 to 2 sat/vB. At the 4 November 2026 price that was about $0.12 to $0.24 (computed from Bitcoin Stack Exchange sizes and mempool.space rates, 2026). Nodes normally wait for the funding transaction to confirm before routing through the channel. Each side must also keep a channel reserve, which the specification suggests setting at 1% of the channel total (BOLT 2).

Updating: signed balances replace each other

Every payment creates a new commitment transaction, a signed spend of the funding output that splits the balance in the new proportions. Each party holds its own copy and does not broadcast it, because broadcasting closes the channel. After each update the parties revoke the previous state by handing over a secret. If one side later publishes an old state, the other can claim the whole channel with a penalty transaction. The reserve exists for this reason: each side keeps something to lose if it cheats. Outputs below a dust limit, which the specification floors at 354 sats, are left out of commitment transactions (BOLT 2).

Closing: cooperative or forced

A cooperative close is one transaction that both parties sign, paying each side its final balance. A force close happens when a peer is offline or uncooperative. One side then broadcasts its latest commitment transaction, and its own outputs stay locked for a number of blocks, the to_self_delay, which each channel sets. That delay gives the other party time to submit a penalty if the published state was old. A force close therefore returns funds to the closer later than a cooperative one, by the length of that delay. The party that opened the channel pays the commitment transaction's fee (BOLT 2).

Data current as of November 2026.

A channel only connects two parties, so reaching anyone else requires the contract that chains channels together.

How Do Hash Time-Locked Contracts Route a Payment Across Several Channels?

A hash time-locked contract (HTLC) lets Alice pay Dave through Bob and Carol without trusting either. Each hop locks funds behind the same secret and a deadline, so the payment settles at every hop or reverts at every hop. Three mechanisms do the work: a hash lock, time locks and onion routing.

The hash lock: one secret pays every hop

Dave generates a random secret, the preimage, and gives Alice its SHA-256 hash inside an invoice. Alice offers Bob an HTLC that Bob can claim only by presenting the preimage. The amount covers Dave's 100,000 sats plus the fees of Bob and Carol, about 21 sats at the 4 November median. Bob offers Carol the same condition minus his fee, and Carol does the same toward Dave. Dave reveals the preimage to collect, and each node then uses it to claim from the node before it. A hash cannot be reversed, so no intermediary can claim funds before Dave reveals the secret (BOLT 2).

The time locks: each hop gets a shorter deadline

Every HTLC carries an expiry measured in blocks. If the preimage never arrives, the offering node reclaims its funds after the deadline. It cancels with its peer or broadcasts an HTLC-timeout transaction (BOLT 3). Expiries shrink toward the destination by the cltv_expiry_delta each forwarding node announces. A route's total expiry is the sum of every hop's delta plus the final node's minimum (BOLT 4). That gap gives Bob time to learn the preimage from Carol and claim from Alice before her offer expires. A delta set too low endangers the forwarding node, not the sender. A channel also caps its open HTLCs at 483 per side, so one penalty transaction can still spend the whole commitment (BOLT 2).

The onion: each node sees only its neighbours

The sender wraps the route in a 1,300-byte onion packet of encrypted layers. Each node peels one layer using a shared secret derived from the sender's ephemeral key. A hash-based message authentication code (HMAC) lets each node verify the packet was not altered. A forwarding node learns the amount, the deadline and the next hop. It does not see the whole route (BOLT 4). Invoices carry the destination, the amount and the payment hash; on mainnet a BOLT 11 invoice begins with the prefix lnbc.

A route only works when every hop has enough balance pointing in the right direction, which is the harder problem.

What Is Inbound Liquidity, and Why Do Some Lightning Payments Fail?

Inbound liquidity is the balance on the other side of a channel, and a node can receive only up to that amount. A channel funded by one party starts with every coin on that party's side. The node can pay out immediately and cannot receive until balance exists on the far side.

Outbound and inbound are two sides of one channel

Capacity is fixed when a channel opens, but the split moves with each payment. A user who opens a 1,000,000-sat channel holds 1,000,000 sats of outbound liquidity and no inbound liquidity. The suggested 1% reserve puts 10,000 of those sats out of reach (BOLT 2). Bitcoin developer Matt Corallo has noted that one-click node installs can produce invoices that cannot be paid because inbound capacity is zero (Spark, 2026). Merchants face the reverse: sales fill their side of each channel until they spend or rebalance. Payment flow tends to run one way, and channel researcher René Pickhardt argues most channels deplete over time (Protos, 2026).

How wallets obtain inbound liquidity

Wallet providers sell inbound liquidity by opening a channel to the user. The Lightning service provider (LSP) specifications LSPS1 and LSPS2 standardize how a wallet requests a channel. They also cover how a provider opens one on the first incoming payment, so the user sees a fee instead of a setup step (Spark, 2026). Eclair also prototyped liquidity ads, a market for inbound capacity. Phoenix, built by ACINQ, charges 1,000 sat to create the first channel. It then resizes that channel with splicing instead of opening new ones (bitcoinproducts.com, 2026; ACINQ, 2023). Someone always pays: the provider locks capital, and the user pays through fees or a creation charge.

Why payments still fail

A payment fails when no route has enough balance on every hop, or when a non-custodial recipient is offline. River reported a 99.7% success rate across 308,000 of its own payments in August 2023 and named those two causes for the failures (River, 2023). Nodes announce channel capacity to the network but not how each channel is split (BOLT 7), so senders retry along other routes. A failed payment cancels at every hop, and the sender loses time rather than funds. Pickhardt's analysis finds failure rates rise with payment size, because larger payments push a route toward its capacity limit (Protos, 2026).

Liquidity also sets price, because every node that lends it charges a fee.

What Does a Lightning Payment Cost Compared With an On-Chain One?

Each forwarding node charges a base fee plus a proportional fee, set per channel. On 4 November 2026 the median channel advertised 100 parts per million (ppm) and a 0.5-sat base fee. One node forwarding 100,000 sats therefore earned 10.5 sats (mempool.emzy.de, 2026).

The fee formula and the measured median

The forwarding fee equals the base fee plus the amount times the ppm rate divided by 1,000,000. A rate of 1,000 ppm on 1,000,000 sats therefore costs 1,000 sats, or 0.1% (Core Lightning documentation, 2026). At the median policy the fee is 0.5 + 100,000 × 100 ÷ 1,000,000 = 10.5 sats, about $0.01. Three forwarding nodes charge 31.5 sats. The average advertised policy is higher, 779 ppm with a 0.891-sat base, which would charge about 78.8 sats per node on the same payment. Expensive channels pull the average up, and route search prefers cheaper ones. For scale, 100,000 sats equal $78.27 on 4 November 2026.

Where a percentage fee passes the on-chain fee

A 141 vB on-chain payment ↗ costs 141 sats at 1 sat/vB and 282 sats at 2 sat/vB (mempool.space, 2026). Phoenix charges 0.4% plus 4 sat per send (bitcoinproducts.com, 2026). That equals the 141-sat on-chain fee at 34,250 sats, about $27, and the 282-sat fee at 69,500 sats, about $54. Below those sizes the wallet's send is cheaper than the chain, and above them the chain is cheaper. The sender knows the total before paying, and the recipient gets the invoice amount. Provider fee schedules change, and none of these figures include the 1,000-sat charge for creating the first channel.

Data current as of November 2026.

Cheap routes depend on a network whose size is harder to pin down than its prices.

How Large Is the Public Lightning Network, and What Do the Numbers Hide?

On 4 November 2026 the public network had 28,939 channels, 15,899 nodes and 3,203 BTC of capacity (mempool.emzy.de, 2026). The figures count announced channels only. Private channels, which mobile wallets and service providers use, do not appear in them, and no public measure sizes them.

How the figures were pulled and what they miss

The numbers come from the mempool.space Lightning API, queried on 4 November 2026. The official endpoint still returned a newest record dated 30 August 2026, so the current figures come from mempool.emzy.de, a community-run instance of the same software. For 30 August the two sources agree within 0.33% on channels and 1.06% on capacity. The series used here starts on 1 June 2024. The first records change implausibly fast, from 1,609 BTC on 30 April 2024 to 4,812 BTC on 1 May 2024. Of the nodes, 7,739 are reachable only over Tor, a privacy network (48.7%), 4,706 use clearnet addresses, 2,021 are unannounced and 1,432 use both.

Capacity and channels since 2024

In this series channels peaked at 51,584 on 14 June 2024 and now stand 43.9% lower. That is not the all-time high: the official mempool.space history, which runs to 30 August 2026, counted 62,629 public channels on 22 October 2023 and a peak of 82,825 on 28 March 2022 (mempool.space, 2026). Capacity stayed between about 3,700 and 5,400 BTC from June 2024 to November 2025, with a low of 3,710.4 BTC on 8 August 2025. It then jumped 31.3% from 4,138.7 BTC on 1 November 2025 to 5,434.9 BTC on 1 December, while channels stayed flat near 42,000. That pattern fits large deposits into existing channels, which Protos attributes to Binance and OKX (Protos, 2026). Capacity reached 5,851.2 BTC on 28 December 2025. River cites 5,606 BTC for December 2025, so trackers differ by method (Bitcoin Magazine, 2026).

The 2026 contraction

From 1 July to 4 November 2026, channels fell 28.5%, from 40,456 to 28,939, and capacity fell 30.4%, from 4,602.7 to 3,203.1 BTC. The steepest drop came in August. Channels fell 13.6% between 1 and 16 August, from 38,151 to 32,980, and capacity fell 12.8%, from 4,271.8 to 3,725.0 BTC. Bitcoin Data Labs counted 2,563 channel closures by four large operators, including Kraken, Boltz and ACINQ, between 2 and 16 August. It tied the period to a wave of security incidents (Bitcoin Data Labs, 2026). Its tally for 1 July to 12 October shows channels down 25% and capacity down 22%. The series used here gives 25.1% and 20.9% for the same dates.

Who holds the capacity

In the node ranking of 21 October 2026, ten nodes sit on one end of channels worth 2,463.2 BTC. Every channel has two ends, so all channel ends add up to 7,072.4 BTC, and the ten nodes account for 34.8% of them. Binance leads with 380.7 BTC across 159 channels, an average of 2.39 BTC each. ACINQ, the company behind Phoenix, follows with 348.6 BTC across 1,823 channels. Exchange and payment-service aliases fill the list, including two Bitfinex-style nodes, OKX, Strike at 149.1 BTC and Kraken at 138.7 BTC (mempool.space, 2026). On 4 November 2026 the average public channel held 0.1107 BTC, about $8,664, and the median 0.02 BTC, about $1,565, so a few very large channels dominate (mempool.emzy.de, 2026). Node aliases are self-declared and unverified.

Capacity measures locked bitcoin, not use, so payment data needs a different source.

How Much Is Lightning Actually Used, and by Whom?

Public capacity says little about payments. River, a Bitcoin financial-services firm, estimated $1.17 billion of Lightning volume across 5.22 million transactions in November 2025. It used data from node operators covering over 50% of network capacity (Bitcoin Magazine, 2026).

River's estimate for November 2025

River's average transaction was $223, up from $118 a year earlier, which fits transfers between exchanges and apps more than small purchases (Bitcoin Magazine, 2026). The count of 5.22 million sits below River's own estimate of 6.6 million routed transactions for August 2023. River tied that peak to micropayment experiments in gaming and messaging (River, 2023; bitbo, 2026). Dollar volume rose while the count fell. The figure is an extrapolation from operators covering about half of capacity, against about 80% a year earlier (Stephan Livera Podcast, 2026). River states that direct or private transactions cannot be estimated (River, 2023).

What exchanges and wallets report

Lightspark reported on 28 April 2025 that 15% of Bitcoin transactions on Coinbase moved over Lightning. Coinbase announced the integration on 3 April 2024 and it went live four weeks later (Lightspark, 2025). River's marketing director Sam Wouters said the growth came largely from exchanges and a rising number of businesses accepting bitcoin (bitbo, 2026). River also cited a $1 million Lightning transfer from Secure Digital Markets to Kraken in February 2026 (bitbo, 2026). Exchange and app flows are the measured growth. Whether everyday retail use grew cannot be separated in public data.

Data current as of November 2026.

Payments land in a wallet, and the choice of who holds its keys sets the risk.

Who Holds the Keys in a Lightning Wallet, and What Does Each Choice Cost?

A Lightning wallet is custodial when a company holds the keys ↗ and self-custodial when the user's recovery phrase does. A third option is a node the user runs. Each type trades convenience against control and channel work. Reviewers treat all three as spending tools rather than cold storage (Coin Bureau, 2026).

Custodial wallets: the provider holds the keys

In a custodial wallet the provider controls the private keys, the channels and account recovery. Strike is the usual example: a regulated financial app that holds bitcoin until withdrawal and requires identity verification (Knowing Bitcoin, 2026). Reviewers name provider failure, account restrictions, withdrawal limits and regional bans as the risks (Coin Bureau, 2026). Wallet of Satoshi's disclosure lists a 1.95% fee on Bitcoin top-ups in both of its models (Wallet of Satoshi, 2026). No deposit insurance backs these balances (FINRA, 2025). The user never sees inbound liquidity, because the provider manages it.

Self-custodial wallets backed by a provider

Phoenix is the worked example. A 12-word recovery phrase controls the funds, and ACINQ never holds the coins (ACINQ, 2026). Each user has one channel, resized by splicing. Sends cost 0.4% plus 4 sat. Receiving is free when liquidity exists, a channel operation costs 1% plus mining fees, and the first channel costs 1,000 sat to create (bitcoinproducts.com, 2026). ACINQ pulled Phoenix from US app stores in April 2024 and announced its return on 8 April 2025 (bitcoin.fr, 2025). The phrase restores funds, not payment history, unless the history was exported (bitcoinproducts.com, 2026).

Running a node of one's own

A user running LND or Core Lightning, two node implementations, controls every key and sets routing fees. In Core Lightning the setchannel command changes a channel's base fee and ppm rate. Many peers limit how often they accept such changes, and Core Lightning allows two a day (Core Lightning documentation, 2026). The same user manages inbound liquidity, rebalancing, uptime and upgrades. Spark reports that force closes lock funds and cost on-chain fees, and that many operators have lost money to them (Spark, 2026).

Data current as of November 2026.

Each wallet type carries its own failure modes, and 2026 supplied several.

What Are the Limits and Risks of Lightning in 2026?

Lightning suits payments, not savings. Liquidity, uptime, channel size and force-close costs limit what a channel handles, and 2026 brought security incidents in both LND and Core Lightning software. A Lightning balance is a spending balance, and operators of large balances carry software risk on top of market risk.

Operational limits

A self-custodial recipient must be reachable for a payment to complete, and a payment fails when no route holds enough balance. Larger payments fail more often (Protos, 2026). A single channel holds at most 16,777,215 sats, about $13,131, unless both nodes support large channels (BOLT 2). A force close pays an on-chain fee at the rate the channel last agreed, and the closer's own outputs stay locked for the to_self_delay. Public capacity has fallen 45.3% from its December 2025 peak, and some of the 2026 decline came from closures by exchanges and service providers.

Security incidents of 2026

On 7 August 2026 BTCPay Server warned of a flaw that let an unauthenticated attacker obtain the credential files, called macaroons, that control an LND node. Attackers exploited the flaw and funds were stolen, though only deployments running LND were affected and the total was not disclosed. Hardware-wallet maker Foundation said its Lightning node was drained while its on-chain hot wallet stayed untouched (CoinDesk, 2026; BTCPay Server, 2026). On 1 October 2026 Core Lightning warned that attackers were targeting nodes on version 26.06.7 or earlier. Maintainers had released a patch on 22 September, and no stolen funds were confirmed (Bitcoin.com News, 2026). A node signs updates continuously, so its keys stay on an internet-connected machine.

FAQ

?Does a Lightning channel expire?

No. A standard channel stays open until one party closes it, and the specifications set no channel-wide expiry. Individual HTLCs inside it do expire, counted in blocks (BOLT 2, 2026). Funds in a channel cannot be spent on-chain until the channel closes, so a channel left open for years is also a balance locked for years.

?What does a Lightning invoice look like, and can it be reused?

A BOLT 11 invoice is a long text string, shown as text or a QR code, that begins with lnbc on mainnet and carries the amount, a payment hash and an expiry. Each invoice is tied to one payment hash, which makes it a single-use request in practice. BOLT 12 offers are a newer, reusable format that only some wallets support (BOLT 11 and BOLT 12, 2026).

?Can a Lightning payment be reversed after it is sent?

No. Once the recipient reveals the secret, every hop settles and the payment is final. A payment that fails or times out reverts instead, because the HTLCs are cancelled or expire and the funds return to the sender (BOLT 2, 2026). A refund after a completed payment depends on the recipient sending a new payment back.

?Is Lightning a separate cryptocurrency or token?

No. Lightning moves bitcoin only. Channels are funded and settled in BTC, and the network has no token of its own. A node's balance is a claim on an on-chain bitcoin output that either party can settle on the blockchain (Poon and Dryja, 2016).

?Is a Lightning payment private?

Partly. Onion routing hides the full route from each forwarding node (BOLT 4, 2026), but public channels and their capacity are announced to the whole network. A custodial provider sees every payment its users make. Privacy therefore depends on the wallet choice more than on the protocol.

?What can a recovery phrase restore in a self-custodial Lightning wallet?

That depends on the wallet. Phoenix restores funds from its 12-word phrase but not payment history, unless the history was exported (bitcoinproducts.com, 2026). A node operator needs a separate channel backup, because the commitment state lives in the node's database, and broadcasting an outdated state invites the penalty transaction.

?How does Lightning differ from a sidechain?

A sidechain is a separate blockchain with its own consensus rules ↗ and a two-way peg to Bitcoin. Lightning has no separate chain. Channel funds sit in a Bitcoin output, payments are signed balance updates, and a dispute is settled on the Bitcoin blockchain itself (Poon and Dryja, 2016).

References / Sources