Ark gets pitched as "the next Lightning." It solves a similar problem — instant, cheap, self-custodial Bitcoin payments — with an almost opposite design. Understanding the difference tells you what each one is actually for.
What if a thousand people needed each other's permission to spend their own money?
A normal Bitcoin transaction output belongs to one key. Spend it, and you need exactly that one signature — nobody else's. Ark breaks that rule on purpose. It lets a large group of people share a single on-chain output, while each person keeps a private, pre-signed way to claim their own slice, alone, whenever they want.
The trick is a tree of transactions, signed by everyone in advance, that mostly never gets broadcast. Bitcoin only ever sees the root of that tree — one output. Ark sees the whole tree, with one branch reserved for each participant.
Tap any dot in the bottom row to highlight one person's exit path back to the root.
More participants means a taller tree and a deeper branch. Each level is one more pre-signed transaction you'd have to broadcast, in order, to walk your own branch back to spendable bitcoin without anyone else's help. Cheaper to join, more expensive to leave alone — that tradeoff is the whole story of Ark, and it's worth remembering as the numbers get bigger below.
How do you get a thousand people onto one blockchain transaction without a thousand people showing up at once?
Ark calls this a round. An operator — the Ark Service Provider, or ASP — gathers a batch of participants, builds the shared tree together with them off-chain, then broadcasts only the root transaction on-chain. One transaction, however many people are inside it.
It's the same trick as a group shipment. A hundred people don't each pay for their own delivery truck. One truck goes out, the trip's fee gets split a hundred ways, and everyone still ends up with exactly their own package.
Round size assumes roughly 110 vB of overhead plus about 43 vB per participant output — a representative estimate, not one implementation's exact figure. Solo transfer assumed at 141 vB, typical for a single-input, single-output payment.
The more people in the round, the smaller everyone's slice of the one on-chain fee. That's Ark's core pitch: Lightning-speed payments, without needing a dedicated on-chain transaction for every balance change.
What does someone have to set up before you're able to pay them?
Getting bitcoin onto Ark for the first time — called boarding — means co-signing a funding transaction with the ASP and waiting for it to confirm, same as any on-chain deposit. But that's a one-time, sender-side step. A brand-new wallet, with nothing boarded and nothing configured, can receive a VTXO from someone else the instant it's created. No inbound liquidity to arrange, no channel to open first.
The catch: a VTXO doesn't live forever. Every branch in the tree carries an absolute deadline. A worked example published by the protocol's own documentation uses 14 days, though each ASP sets its own value — some deployments use a window closer to four weeks. Past that deadline, both you and the ASP can unilaterally spend it. Miss the window, and you're racing the operator, not just Bitcoin's mempool.
Under normal use this is invisible — wallets are built to refresh a VTXO into a new round well before the clock runs out. But "normal use" is doing some work in that sentence, and the next two sections are about what happens when it isn't.
If the ASP goes offline — or you simply stop trusting it — what does it actually cost to get out without its help?
Recall the tree from section one: a unilateral exit means broadcasting one on-chain transaction per level, in order, all the way down to your leaf. Ark's own published specification puts each of those transactions at roughly 150 vbytes, and O(log n) transactions per person for a batch of n participants. Normally almost nobody does this — the cooperative path, called an offboard, is cheaper and is what wallets use by default.
But the unilateral exit is the actual security promise. It's supposed to work even when the ASP refuses to cooperate. So it's worth asking what it costs if a lot of people need it at once — a bank run, in transaction form.
†Assumes ~3,000 transactions per block and one block every 10 minutes, with 100% of block space dedicated to these exits alone and zero competing traffic — a generous best case, not a prediction.
At small batch sizes this is nothing — a rounding error next to Bitcoin's daily transaction count. Slide it up.
Adjust the slider in section 04 and this comparison updates.
On May 8, 2023, Bitcoin's mempool held roughly 440,000 unconfirmed transactions at once — its largest backlog to date — after a single day set an all-time record of 682,281 transactions, driven by a wave of BRC-20 token and Ordinals inscription activity. Fees followed: by mid-December that year, a separate backlog of about 470,000 pending transactions pushed the highest fees above $40 each.
That wasn't an Ark event — Ark hadn't launched at any real scale yet. It was ordinary Bitcoin congestion. But it landed on Lightning users the same way Ark's own specification warns a mass exit would. A public bug report from that exact week shows a node operator who force-closed several channels with too-low fees during the spike; the closing transactions sat stuck for days, with some channels left stalled in a "waiting close" state until the mempool cleared.
Ark's specification is candid about the parallel: it names the congestion risk from a mass unilateral exit directly, notes the protocol still guarantees exits eventually, and lists ideas like staggered exit windows or paid exit priority as "future work" — meaning nobody has shipped a fix yet.
It's natural to reach for Lightning as the comparison — it's the only other Bitcoin layer running at real scale. But the mechanics share very little. Lightning routes a payment across a chain of bilateral channels; every hop needs its own pre-funded, balanced liquidity, and receiving means either staying online or paying a service to watch for you. Ark has no routing graph at all — a payment simply swaps a share of one pooled output for a share of a new one, cosigned by the ASP, done.
| Lightning | Ark | |
|---|---|---|
| Structure | Bilateral channel network | One shared UTXO tree |
| Moving a payment | Routed across hops | Swap share for share, cosigned by ASP |
| Receiving, first time | Needs inbound liquidity | No setup required |
| Staying safe offline | Watchtower recommended | Refresh before expiry |
| Full exit without cooperation | Force-close, 1 tx | O(log n) txs per person |
| Coordinator | None — permissionless | A named ASP |
The two aren't strictly rivals, either. ASPs commonly run a Lightning gateway, letting Ark users send and receive real Lightning payments without operating a Lightning node of their own. And in July 2026, Lightning Labs — the company behind Lightning's most widely used implementation — released an early, Ark-like settlement layer of its own, aimed at agent-to-agent payments. Even the Lightning side sees a use for shared-UTXO batching.
If the ASP can never spend your live VTXO, is Ark actually trustless?
Mostly, but not entirely. The ASP genuinely can't move funds inside a VTXO you haven't forfeited — every exit path is pre-signed to your key alone. What it can do is refuse to include you in a future round, freezing you out of the cheap cooperative path and pushing you toward the expensive unilateral exit from section four. And in a payment chain, a dishonest sender could in principle collude with the ASP to double-spend a VTXO before its new owner refreshes it — a risk the protocol's own documentation names directly, cleared by refreshing promptly rather than sitting on an unconfirmed chain.
And it's simply new. Arkade processed its first mainnet payments in August 2025 and opened publicly that October; Second's Bark implementation reached mainnet in June 2026, with a public server, SDK, and integrations spanning mobile wallets, an Umbrel app and a BTCPay Server plugin. A shared format for backing up and verifying VTXOs across different implementations, V-PACK, was still just a proposal as of March 2026. Lightning has been running in production, imperfectly, since 2018. Ark hasn't had that much time to find its own edge cases yet.
A pre-signed exit isn't a free exit. Know roughly what a unilateral exit would cost at the batch size your ASP actually runs before you rely on "I can always leave."
Refresh before the clock runs out. A VTXO you forget about doesn't just sit there safely — past its timelock, the ASP can claim it too.
No custody isn't the same as no power. An ASP can't steal a live VTXO, but it can freeze you out of future rounds. Know who operates the one you use.
Check whether your wallet bridges Lightning. Increasingly, Ark and Lightning are one stack, not two competing choices — don't assume you have to pick.