# Ark Arrives: Bitcoin Payments After Lightning Plateau

> Lightning public capacity fell then recovered while volume grew ~266% YoY. Ark hit mainnet in 2026. A mechanism-level comparison.

- Source: https://www.steadyrain.xyz/en/articles/ark-vs-lightning-bitcoin-payments-2026/
- Publisher: Steady Rain
- Author: 0xYatha
- Published: 2026-09-02T03:14:07.327Z
- Categories: BTCFi
- Language: en

## Introduction

Lightning's public channel capacity bottomed near 4,200 BTC in August 2025, down roughly 20% from the \~5,400 BTC it carried in late 2023, according to Spark Money's 2026 state-of-the-network report. Four months later the same metric printed an all-time high around 5,637 BTC as Binance, OKX and Coinbase expanded their integrations, then relaxed to \~4,898 BTC by May 2026 across 41,080 channels and 17,438 nodes. Node count has been declining since a 2022 peak near 20,700. Over the same window, publicly measured routed volume grew roughly 266% year over year, crossing $1.17 billion in November 2025 across approximately 12 million monthly transactions.

Capacity is a stock and volume is a flow, so the two series are not required to move together. But the divergence is large enough to demand an explanation, and the explanations point in opposite directions. Either Lightning is consolidating into a smaller set of high-velocity enterprise nodes that recycle the same satoshis many times per month, or hobbyist routing operators are abandoning a network whose channel-management burden never got cheap enough, and exchange flow is papering over that retreat. Spark Money's report leans toward the first reading and calls it network restructuring. That is an interpretation, not a measurement, and it comes from a company that sells Bitcoin payment infrastructure.

Into this ambiguity arrives Ark. The Bark implementation went live on Bitcoin mainnet in 2026 per the Bitcoin Optech topic page, following an Arkade beta from Ark Labs and a formal treatment published on arXiv by Keer, Alexopoulos, Maffei, Argentieri, Camilleri and Avarikioti that defines the protocol and proves security properties. Ark takes the opposite structural bet from Lightning. Rather than giving each pair of participants a bilateral channel that must be funded before it can receive, Ark packs many users under one onchain UTXO and lets an operator batch their offchain transactions into a commitment the paper measures at roughly 200 virtual bytes regardless of how many users are inside.

That design achieves substantial onchain compression while adding coordination and trust constraints. Evaluating those constraints is essential because Ark and Lightning are already being stitched together at the gateway layer rather than competing directly for the same slot.

## What Lightning's Public Metrics Actually Count

*[Figure: img1]*

The headline capacity figure counts announced channels only. Mobile wallets, enterprise nodes and Lightning service providers routinely run unannounced channels that never appear in gossip. Spark Money estimates private capacity at two times or more the publicly visible amount, which would put real network capital well above the \~4,898 BTC figure. That estimate is not independently verified and the report does not describe its derivation, so it should be read as a directional correction rather than a number to build on.

Channel counts tell a similar story of compression. The report tracks roughly 39,000 channels in August 2025, \~43,000 in December, and 41,080 by May 2026. Capacity swung far more than channel count across the December spike, which is consistent with existing operators enlarging channels rather than a wave of new participants opening small ones. Exchange integrations behave that way: an exchange does not open ten thousand retail channels, it opens a modest number of very large ones to its liquidity partners.

The demand side is where the growth sits. Coinbase completed its Lightspark-based integration in mid-2024 and by mid-2025 more than 15% of its Bitcoin withdrawals moved over Lightning. CoinGate reported Lightning at over 16% of Bitcoin orders in 2024, up from roughly 6.5% two years earlier. Block began rolling Lightning out to approximately 4 million US point-of-sale merchants. Outside exchange rails, El Salvador's Chivo wallet processed 4.2 million Lightning transactions in 2025, Bitnob recorded 340% year-over-year volume growth across 23 African countries, Machankura provides USSD and SMS access in five African countries, and Nostr zaps passed 5 million cumulative transactions by May 2025. All of these figures come from the same Spark Money report and share its sourcing limitations.

The protocol did not stand still either. BOLT12 offers merged into the specification in September 2024 and have been adopted by Core Lightning, LDK and Eclair, though LND still lacks native support, which fragments what a wallet can assume about a payee. Splicing, which resizes a channel without closing it, became the default in Core Lightning in 2026 with a prototype in Eclair and splice-out support in LDK. Taproot channels reduced onchain fees by roughly 20% in Phoenix Wallet's implementation while improving privacy through Schnorr signatures and P2TR outputs. Taproot Assets shipped v0.6 in June 2025 and v0.7 in December, adding multi-asset capability, reusable addresses and auditable supplies, with Tether announcing USDT support.

Splicing matters more than its low profile suggests. A channel that can be resized in place weakens the link between capacity and participation: an operator can hold less idle capital and expand only when flow demands it. If splicing adoption is meaningful, some portion of the capacity decline is capital efficiency rather than abandonment. The available data cannot separate the two effects.

## Batching Users Under One UTXO

*[Figure: img2]*

Ark restructures the problem. Instead of a channel between two named parties, an **Ark** operator maintains a single onchain UTXO that anchors a tree of pre-signed offchain transactions. Each leaf of that tree is a **VTXO**, a virtual UTXO belonging to one user. The tree is constructed and refreshed in periodic interactive rounds: users who want to join, exit or refresh coordinate with the operator, everyone co-signs the new transaction set, and the operator commits the result onchain.

The arXiv paper describes Ark as the first Bitcoin-compatible commit-chain and identifies two properties that distinguish it from earlier payment-pool designs. First, a user can receive an offchain payment without having locked funds in advance. Second, updating offchain state requires only the users involved in that specific update rather than every participant in the pool. The first property removes Lightning's inbound-liquidity problem at the root. A Lightning wallet that has never received bitcoin cannot receive bitcoin until someone opens a channel toward it or a service provider fronts inbound capacity, which is why LSPs exist and why they charge for the service. An Ark receiver just needs a VTXO issued to it in a round.

The second property is what makes the design tractable at scale. In a naive shared pool, every state change requires every member to sign, which means one absent member freezes everyone. Ark's construction narrows the signing set to the parties to a given transfer plus the operator.

Exits come in two flavors with very different cost profiles. A cooperative exit, where the operator participates, adds one onchain output per exiting user. A unilateral exit, where the user broadcasts the branch and leaf transactions in sequence to claim their output without operator help, costs O(log n) transactions of roughly 150 vB each for a batch of n VTXOs, per the paper's experimental evaluation. Those figures are the authors' own measurements and have not been independently replicated in the material available here.

The asymmetry is the point. Cooperation is cheap and abandonment is survivable but expensive, and the expense scales with the log of how many people share the tree. That is structurally similar to a Lightning force-close, where the honest path is a mutual close producing two outputs and the adversarial path involves broadcasting a commitment transaction, waiting out a timelock and potentially sweeping HTLCs. Ark's adversarial path is worse in transaction count and better in the sense that it does not require the exiting user to have been watching the chain for a counterparty's cheat attempt.

## Where Ark's Trust Sits

*[Figure: img3]*

Ark operates under constrained trust rather than the trust model of a Lightning channel between two online, well-watched nodes.

### Expiry and the refresh obligation

Every VTXO carries an absolute timelock. After it expires, both the operator and the user can unilaterally spend the underlying output, which is how the operator recovers the liquidity it allocated to that branch of the tree. A user who lets a VTXO expire without refreshing it into a new round has, in effect, handed the operator the option to take those funds. The mechanism is the operator's capital-recovery instrument, and without it no operator would allocate liquidity to a tree whose members might disappear forever.

This converts self-custody into a recurring action. Lightning imposes a comparable but different burden: a channel does not expire, but a user who goes offline for a long stretch relies on a watchtower or on their counterparty's honesty to avoid an old-state broadcast. Ark's version is a deadline rather than a vigilance requirement, which is easier to automate and harder to forgive. A wallet that misses a watchtower window might still be fine if nobody attacked it. A wallet that misses an expiry loses funds if the operator acts opportunistically.

### Collusion across payment chains

Payments between Ark users are pre-signed extensions of an existing VTXO requiring co-signature from the transacting user and the operator. Until the receiver rolls that VTXO into a new round, the sender could in principle collude with the operator to double-spend it. The risk compounds along a chain: if A pays B, B pays C, and C pays D, all without an intervening round, then D's position depends on none of the upstream parties having cut a deal with the operator.

This is the most critical line in Ark's threat model, and it explains why the round cadence is a security parameter and not just a throughput parameter. Frequent rounds shorten the window in which chained payments carry accumulated counterparty risk. Infrequent rounds are cheaper onchain and leave receivers exposed longer. An operator optimizing purely for fee efficiency has an incentive to lengthen rounds, and users have an incentive to shorten them. Where that tension settles in practice is an open empirical question; no adoption or round-cadence data for mainnet Ark deployments appears in the sources reviewed here.

### Operator liveness

Rounds are interactive. A user who wants to refresh, exit cooperatively or transact needs the operator to be online and willing to sign. Lightning routing is asynchronous in a different way: a payment needs a path of online nodes, but no single privileged party gates every state transition. Ark's operator is a coordination bottleneck by construction. It cannot steal from a user who exits unilaterally in time, but it can refuse service, and refusal plus an approaching expiry is a coercive combination for a user facing an O(log n) exit bill during a fee spike.

## What the Formal Analysis Establishes

The arXiv paper does more than restate the design. Its authors formally defined the protocol, proved security properties, and in the course of that analysis found two attacks affecting the testnet implementation. They disclosed them responsibly and the fixes were integrated into the mainnet implementation. The abstract does not describe what the attacks were, so we cannot assess whether the fixed versions carry residual risk of the same class.

That sequence is still meaningful. A protocol that went to mainnet after a formal treatment caught live implementation bugs is in better shape than one that shipped on a whitepaper and a testnet. It is also a reminder of the gap between a proven protocol and a correct implementation of it, which is where most protocol losses have historically occurred.

Ark Labs is direct about the maturity question. The Arkade launch post describes the software as beta and states that it is not ready to handle a user's life savings, despite testing of core fundamentals. That is a vendor's own framing in an announcement, which reinforces its credibility. Modeling Ark as a drop-in Lightning replacement in 2026 contradicts the documentation from its builders.

The implementation landscape is also still forking. Bark reached 0.5.0 and then mainnet. V-PACK proposes a stateless VTXO verification standard. hArk is a hash-lock-based variant. Wavelength shipped an alpha with an Ark-like settlement layer. A design space that is still producing structural variants a year into deployment has not converged on what the production version looks like.

## Operator Economics Against Routing-Node Economics

*[Figure: img4]*

This comparison is where the sourcing thins out, and it is necessary to mark the boundary before crossing it. Neither Ark source publishes an operator fee schedule, revenue model, mainnet volume, user count or TVL. What follows reasons from the mechanics, not from disclosed financials.

A Lightning routing node earns fees on forwarded payments and pays for that income with locked capital, rebalancing costs and onchain fees for opening, splicing and closing channels. Its capital is committed per channel and per direction. Inbound capacity that sits unused earns nothing, and the operator cannot redeploy it without an onchain event. The falling node count against rising volume is consistent with this economics being hard for small operators: routing fee income scales with flow, flow concentrates on well-connected paths, and a node that is not on those paths earns close to nothing while still absorbing the capital cost. Splicing softens the resizing problem but does not change who gets the flow.

An Ark operator's capital works differently. It allocates liquidity into a tree that can hold many users under a single onchain footprint, and it recovers that liquidity automatically at VTXO expiry rather than through an onchain close. The constant \~200 vB commitment per batch means the operator's marginal onchain cost per additional user in a round trends toward zero, which is a distinct cost curve from Lightning's per-channel onchain cost. Against that, the operator carries a liveness obligation to every user in the tree simultaneously and must fund rounds on a schedule.

The revenue side is where the model is least legible. An operator plausibly charges for round participation, for providing liquidity to new VTXOs, or through a spread on swaps into and out of the Ark. Which of these dominates determines whether Ark operation is a thin infrastructure business or a fee-extractive one, and we have no data to say. It also determines the round cadence question raised earlier: an operator paid per round has different incentives from one paid on a spread.

One structural observation does survive the data gap. Lightning routing is a many-operator market where any node can attempt to earn on any path, and the evidence suggests that market is consolidating. Ark operation is closer to a service relationship, where a user picks an operator and inherits its liveness and cadence choices. Consolidation is inherent to Ark's architecture. Whether users end up with a competitive choice among operators, or with two or three that everyone routes through, will be visible in the operator count long before it shows up in anyone's fee schedule.

## What Covenants Would Change

Ark runs on Bitcoin today with no consensus changes. Every path through the transaction tree exists because the participants pre-signed it during a round. That is why rounds are interactive and why the signing set matters. The Optech topic page states that Ark would support significantly more users with better fee efficiency given covenant opcodes such as **OP_CHECKTEMPLATEVERIFY**, and references discussion of CTV combined with CSFS, but the available excerpt does not detail the mechanism.

The plausible reading, and we flag it as inference rather than sourced claim, is that a covenant lets a script commit to the shape of its spending transaction without anyone signing that transaction in advance. Applied to Ark, the tree structure could be enforced by script rather than reconstructed from a pre-signed set, which would compress the interactive work each round has to do and reduce the state each participant must retain to exit. The V-PACK stateless verification effort points in the same direction from the implementation side.

What matters for the trust model is whether covenants would remove the operator's co-signature from the payment path or only from the tree construction. Removing it from tree construction improves scale and cost. Removing it from transfers would eliminate the sender-operator collusion vector that currently makes chained payments risky between rounds. Those are very different upgrades and the sources do not distinguish them. Claims that a soft fork makes Ark trustless must specify which mechanism is altered.

The dependency runs the other way too. If mainnet Ark hits real capacity or fee ceilings without covenants, Ark becomes a constituency in Bitcoin's activation politics, with a working deployment and users as evidence rather than a proposal. That is a materially stronger position than CTV advocates have held before, and it may be the most consequential second-order effect of Bark shipping.

## Gateways Instead of a Handoff

*[Figure: img5]*

Ark Labs frames Arkade's origin as an attempt to fix Lightning's shortcomings that grew into a programmable-Bitcoin platform. The tooling it shipped does not read like a replacement effort. Arkade launched with SDKs in TypeScript, Golang and Rust, a BTCPayServer plugin, a reference wallet, and Fulmine, a wallet daemon aimed at swap providers and payment hubs that is explicitly designed to optimize Lightning channel liquidity while minimizing onchain fees through Arkade integration. Named partners include Breez, LayerZ Wallet, BTCPayServer, Boltz, BullBitcoin and Lendasat. Boltz and Breez are Lightning businesses.

Fulmine describes the emerging topology well. A payment hub holds Lightning channels that require inbound and outbound capacity and periodic onchain rebalancing. If that hub parks its float in an Ark and pulls it into channels as flow demands, its onchain cost per rebalance falls and its idle capital falls with it. Lightning does the routing and the last-mile retail interaction; Ark does the settlement and the capital storage. Neither replaces the other, and the user may never learn which one held their balance at any given moment.

That framing comes from Ark Labs' own announcement, so it warrants discounting. Two systems that both want developer attention, wallet integrations and liquidity capital are competing for scarce resources whatever the architecture says. But the gateway pattern is not merely rhetorical: swap providers like Boltz already move value between Lightning and onchain Bitcoin, and Ark is a third endpoint for the same machinery rather than a new category of problem.

Where each fits follows from the mechanics. Lightning wins where a payment must complete in under a second between two parties who have no prior relationship and no shared operator, which describes retail point-of-sale, exchange withdrawals and cross-network routing. Ark's shape suits populations of many users transacting infrequently against a common service, where the per-user onchain cost of channel management would dominate the payments themselves, and where a receiver needs to be able to receive before it has ever held funds. The programmable-Bitcoin ambitions in the Arkade announcement, including Arkade Assets and gestures toward lending and AMM constructions, are a bet that the second population is larger than the first. That bet is unproven and we have no adoption data to test it.

## The Measurements That Would Settle This

The comparison currently rests on asymmetric evidence. Lightning has years of public gossip data, imperfect and undercounting private capacity but continuous. Ark has a formal security paper, a mainnet implementation and no published volume, user count, operator count or TVL. Until Ark deployments publish something measurable, claims about which design is winning reflect architectural preferences rather than operational scale.

Three observable conditions would change the analysis. The first is Ark round cadence in production, because it prices the collusion window that chained payments carry. An Ark whose operators settle every few seconds has a very different risk profile from one settling hourly, and the difference is invisible from outside until someone publishes it. The second is the operator count. A handful of operators serving most VTXOs reproduces the concentration that Lightning's node attrition is already showing, without the consolation that anyone can spin up a competing node and earn on the same paths. The third is the fate of expiring VTXOs. Every expiry that an operator claims because a user missed a refresh window is a data point on whether the refresh obligation is an automatable background task or a real loss channel, and it surfaces as user complaints long before it surfaces as research.

On the Lightning side, the question that the 2025 capacity decline leaves open is whether splicing adoption explains part of it. If channels are being resized rather than closed, and if operators are holding thinner float against the same flow, then capacity is a deteriorating proxy for network health and the industry should stop citing it. Nobody in the available sources has done that decomposition, and the Spark Money report's restructuring thesis asserts the conclusion without the evidence that would support it.

The covenant question sits underneath all of it. Ark's current mainnet form is the version that fits inside Bitcoin's existing script, and its authors say a covenant version would serve significantly more users more cheaply. If mainnet usage grows into the limits of the non-covenant design, the argument for CTV stops being theoretical and starts being a report on a live system's ceiling.

## References

- [Ark protocol topic page](https://bitcoinops.org/en/topics/ark/), Bitcoin Optech
- [Press Start: Arkade Goes Live](https://blog.arklabs.xyz/press-start-arkade-goes-live/), Ark Labs
- [State of the Lightning Network in 2026](https://www.spark.money/research/lightning-network-2026-state), Spark Money
- [Ark: Offchain Transaction Batching in Bitcoin](https://arxiv.org/abs/2605.20952), Keer, Alexopoulos, Maffei, Argentieri, Camilleri, Avarikioti