Lightning, Ark, and Spark: How Bitcoin’s Payment Layers Compare

Bitcoin’s base layer settles a block roughly every ten minutes and was never built to clear millions of five-dollar payments. That gap between what Bitcoin’s ledger can process and what people actually want to do with it, pay each other quickly and cheaply, has spawned an entire category of layer 2 networks. The Lightning Network was the first serious attempt to close it, and by mid-2026 it is carrying more than a billion dollars in monthly payment volume. But it is no longer the only option.

Two newer designs, Ark and Spark, spent 2025 and into 2026 arguing that Lightning’s payment channels are the wrong building block entirely. Both promise Bitcoin-backed, trust-minimized transfers without the channel management that has made Lightning notoriously fiddly to run. Neither has anywhere near Lightning’s scale yet. This is a look at how the three actually differ under the hood, who is backing each one, and what a developer or a payments company should weigh before picking a rail in 2026.

Why Bitcoin needed a second layer in the first place

Bitcoin’s block size effectively caps the network at somewhere around seven to ten transactions per second, and that ceiling has not moved since the network’s original design. Every four years the block subsidy that pays miners for securing the chain cuts in half, which means transaction fees have to eventually shoulder more of that security budget. That dynamic, explored in depth in HOGE Wire’s coverage of Bitcoin’s halving cycle math, is part of why fee pressure on the base layer has never really gone away: it spiked during the 2017 bull run, again during the 2021 mania that spilled onto Bitcoin through wrapped assets, and once more in 2023 and 2024 when Ordinals inscriptions and BRC-20 tokens competed with ordinary payments for block space.

Layer 2 networks exist to route most of that everyday activity off the base chain while still settling disputes and final custody there. The Lightning Network, first described in a 2015 whitepaper by Joseph Poon and Thaddeus Dryja and live on mainnet since early 2018, was the first design to reach meaningful scale. It works by opening a funded, two-party payment channel on-chain, then letting the two parties exchange balances privately and instantly off-chain, with the underlying bitcoin only touching the blockchain again when the channel closes. For years, Lightning had the layer 2 conversation to itself. That changed as two very different architectures, Ark and Spark, moved from whitepaper to mainnet in 2025, each betting that channels are not the right building block for a payments network and that a shared, batched, or federated model can deliver similar speed with less operational overhead.

Lightning Network: payment channels, HTLCs, and onion routing

A Lightning channel begins with an on-chain transaction that locks bitcoin into a two-of-two multisignature address controlled by both participants. From that point on, the two parties simply sign updated commitment transactions between themselves, each one reflecting a new balance split, without broadcasting anything to the Bitcoin network. Either side can close the channel at any time by publishing the latest signed state; a penalty mechanism, often called a justice transaction, lets an honest party claim the entire channel balance if the other side tries to cheat by broadcasting an old, more favorable state.

Payments that need to travel beyond a single channel move through Hash Time-Locked Contracts, or HTLCs, which chain several channels together into a route. Each hop is cryptographically bound to the same payment hash, so the payment either completes across every hop or fails everywhere; nobody in the middle can walk away with funds without forwarding them onward. Onion routing, borrowed conceptually from Tor, wraps the route in layers of encryption so each node only knows the hop immediately before and after it, not the full path or the final recipient.

That design is what makes Lightning trust-minimized rather than trust-free: at no point does a routing node take custody of somebody else’s funds for longer than a payment is in flight, and the underlying channel balance is always recoverable on-chain. The specification lives in the BOLT (Basis of Lightning Technology) documents, implemented independently by four major pieces of software: LND from Lightning Labs, Core Lightning (CLN) from Blockstream, LDK from Spiral, and Eclair from ACINQ, which powers the Phoenix wallet. The catch is operational. Someone has to open channels, fund them, and keep enough inbound liquidity on hand to actually receive payments, not just send them, which is exactly the gap that channel-free designs like Ark and Spark are trying to close.

The state of Lightning in mid-2026

Public Lightning capacity has been volatile over the past year, and bitcoin itself has been trading in the mid-60,000-dollar range through mid-July 2026. Capacity bottomed out around 4,200 BTC in August 2025, spiked to a fresh all-time high in December 2025 as large exchanges moved bitcoin into channels, and has since given back a chunk of that gain, according to tracking from Spark’s Lightning Network research and reporting from Bitcoin Magazine, whose figures are pulled from Amboss and Bitcoin Visuals.

Node and channel counts differ slightly between trackers depending on whether they count private and unannounced channels, but the broader trend is consistent: the public node count has fallen by roughly 16 percent since its 2022 peak of around 20,700, even as capacity swung higher. That split, more bitcoin locked in fewer, larger channels, points to a story told throughout 2025 and 2026: growth is increasingly institutional, driven by exchanges and payment processors rather than individual hobbyist nodes.

Volume tells a similar story. Lightning processed roughly $1.17 billion in payments in November 2025, up sharply from about $286.5 million in November 2024, according to data cited by DL News. The average payment size has grown to around 74,000 sats, worth roughly $50 at recent prices, which BlockSpaceForce managing partner Spencer Yang has pointed to as evidence the network’s center of gravity is shifting. “The average transaction value on the Lightning Network has actually increased and it goes to show that it’s not so much peer-to-peer payments for coffee or whatever; it’s more like a settlement between exchanges or between businesses,” Yang told DL News. Voltage, a Lightning infrastructure and liquidity provider, has separately claimed that close to 29 percent of bitcoin transfers now move over Lightning rather than settling directly on-chain, though it has not published the methodology behind that figure.

BOLT12, splicing, and why LND is still catching up

Two protocol upgrades matter most for Lightning’s usability in 2026. The first is BOLT12, a spec that replaces single-use invoices with reusable “offers”, static codes a merchant or a person can post once and reuse for repeat payments. BOLT12 merged into the official specification in September 2024 and is supported natively by Core Lightning, LDK, and Eclair. LND, which still runs the largest share of public nodes, has kept the feature behind an experimental flag since v0.18.4 and relies on a separate sidecar project called LNDK to bridge the gap. LND v0.21, released on June 11, 2026, added the ability to forward onion messages on behalf of peers, which Lightning Labs describes as a first concrete step toward native offer support, though full BOLT12 in LND itself had still not shipped as of this writing.

The second upgrade is splicing, which lets a node resize an open channel, adding or removing funds, without closing it and paying to reopen a new one. Previously, adjusting liquidity meant an on-chain close and reopen, plus the fees and confirmation delay that come with it. Core Lightning moved splicing out of experimental status in its 26.04 release in April 2026, alongside a new routing engine called askrene and roughly 20 percent smaller binaries, the product of 421 commits from 23 contributors over about 110 days.

LND v0.21 also carried a less flashy but arguably more consequential change: migrating its payment store to a native SQL backend, which Lightning Labs says cut wall-clock time on large payment history queries by more than 97 percent. Simple taproot channels, which reduce a channel’s on-chain footprint and improve privacy, also graduated to production-ready status with support for RBF-enabled cooperative closes. None of this closes the gap with Ark or Spark’s zero-channel-management pitch, but it shows Lightning’s own maintainers feel the competitive pressure and are sanding down the rough edges rather than standing still.

Ark: shared UTXOs, VTXOs, and the ASP model

Ark takes a fundamentally different starting point than Lightning. Instead of two parties locking funds into a dedicated channel, every user transacts through a coordinator called an ASP, or Ark Service Provider, which batches many users’ transactions into a single shared on-chain UTXO at the start of each round. Inside that shared structure, users hold what are called Virtual Transaction Outputs, or VTXOs: off-chain claims on a slice of the shared UTXO, represented by a tree of pre-signed Bitcoin transactions that only the user can ultimately unlock.

The security promise is that the ASP never actually takes custody of user funds. Every VTXO is backed by a pre-signed exit path, so a user can unilaterally broadcast their branch of the transaction tree and reclaim their bitcoin on-chain at any time, with or without the ASP’s cooperation, a mechanic explained in detail in Bitcoin Optech’s technical overview of the protocol. That is a meaningfully different trust model from Lightning’s penalty-transaction approach: Ark has no equivalent of a Lightning justice transaction, so if an ASP tries to publish a stale state, users fall back on timelocks rather than a punitive claim on the operator’s funds. An ASP cannot steal a user’s VTXO outright, but it can censor a user by refusing to include them in a future batching round, forcing a slower, costlier on-chain exit.

That combination, no ability to seize funds but a real ability to degrade service, puts Ark somewhere between a federated sidechain and Lightning on the trust spectrum. What it buys in exchange is a much simpler experience for builders: rather than opening and managing individual channels, an application just talks to an ASP, and rather than worrying about inbound liquidity, VTXOs can represent arbitrary amounts. Ark’s design was proposed in 2023 and stayed largely theoretical until Ark Labs shipped the first production implementation, called Arkade, in the second half of 2025.

Inside Arkade: launch partners, funding, and Arkade Assets

Ark Labs launched Arkade in public beta on Bitcoin’s mainnet on October 21, 2025, after roughly two years of development, calling it Bitcoin’s first major new layer 2 since Lightning itself. The launch shipped with support from wallet and infrastructure partners including Breez, BlueWallet, BTCPayServer, BullBitcoin, and LayerZ Wallet, and the company counts Draper Associates, Axiom, and Fulgur Ventures among its backers.

“The Bitcoin L2 landscape has been full of promises but light on shipping. Today’s release marks the beginning of Bitcoin’s evolution as programmable money,” Ark Labs CEO Marco Argentieri said at launch, a pointed comment aimed at the string of Bitcoin scaling proposals that never made it past a whitepaper or a testnet.

Ark Labs has since pushed toward multi-asset support with Arkade Assets, a framework meant to bring stablecoins and other tokens onto the same VTXO-based execution layer, with USDT support planned as the flagship use case. Tether’s involvement is not incidental: the stablecoin issuer backed a $5.2 million seed round in Ark Labs in March 2026, part of a broader pattern of Tether investment across nearly every Bitcoin-native payment rail in 2025 and 2026, alongside its Taproot Assets rollout on Lightning and its parallel bet on the RGB protocol. Tether is not picking a single winner among Bitcoin’s competing layer 2 designs; it is hedging across Lightning, Ark, and RGB at once, betting that stablecoin demand on Bitcoin will be large enough, and the winning architecture uncertain enough, that spreading the bet is cheaper than guessing wrong.

Spark: statechains reinvented with FROST and leaves

Spark starts from a different lineage again: statechains, a concept Bitcoin researcher Ruben Somsen proposed in 2018, in which ownership of an entire on-chain UTXO transfers between parties by handing over a rotating private key rather than by broadcasting a new transaction. The original design had an obvious limitation: because it transferred a whole UTXO at a time, it was clumsy for everyday payments of arbitrary, uneven amounts.

Lightspark’s Spark protocol addresses that with what it calls a leaf architecture: balances live as leaves in an off-chain tree that can be split and recombined, so a user can send an arbitrary amount without moving an entire UTXO. Custody of the underlying funds works through a two-of-two arrangement between the user and a set of independent operators, collectively called the Spark Entity, which hold their half of the signing key using FROST threshold signatures. Because FROST distributes key shares across the operator set, no single operator ever holds the complete key, and the resulting on-chain transaction looks like an ordinary 110-vbyte taproot spend regardless of how many operators are involved, a privacy property neither Lightning nor Ark shares: an outside observer cannot tell whether a payment was signed by one operator or a hundred. Spark describes this as a one-of-n trust model: user funds stay safe as long as at least one operator in the set behaves honestly, and a user can always fall back to a unilateral on-chain exit.

Two operators, Lightspark and Flashnet, currently run the Spark Entity, with more expected to join over time. It is worth being precise about naming here: this Spark, the Bitcoin statechain network built by Lightspark, has no relationship to Spark Protocol, the Ethereum-based lending market with ticker SPK that trades under the Sky ecosystem, formerly MakerDAO, and carries billions of dollars in an entirely separate part of DeFi. The two happen to share a name and nothing else. Lightspark’s Spark network carries a fraction of Lightning’s public capacity in total value locked, a modest figure that is notable mainly because the protocol only became commercially relevant through 2025 and 2026.

Grid: Lightspark’s bigger bet on statechains

Spark was never the end goal for Lightspark so much as the plumbing underneath a bigger product. At the Bitcoin 2026 conference in Las Vegas in April, chief executive David Marcus unveiled Grid Global Accounts, a dollar-denominated account layer built on top of Spark that gives businesses a single wallet address capable of holding and routing both bitcoin and dollar balances. Lightspark simultaneously became a principal member of the Visa network, which the company says gives Grid account holders spending access to Visa’s roughly 175 million merchants, live payouts to more than 65 countries and 14,000 banks, and Visa debit cards, according to reporting from Forbes. Marcus has said he is targeting 75 countries and 100 Visa markets by the end of 2026.

Grid also shipped with what Lightspark calls agent delegation, letting an AI agent execute payments or account actions inside a user’s wallet under a defined, revocable set of permissions, a feature aimed at the same agentic-commerce trend that has pulled in competitors building directly on Lightning.

Marcus’s rationale for building on statechains rather than expanding non-custodial Lightning support goes back further than the April announcement. In late 2023, well before Spark existed as a shipped product, he argued that Lightning’s non-custodial model forces an uncomfortable trade: “To be blunt, if you want full support for non-custodial Lightning with offline receive and want to make it economically viable, you have to accept some form of compromise on the trustlessness level of the solution.” Spark, and now Grid, are Lightspark’s answer to that problem: instead of pretending the compromise does not exist, the company built a protocol that names it explicitly, a one-of-n operator trust assumption, and optimized everything else around it.

Trust models side by side

Strip away the marketing and the three networks sit at genuinely different points on a trust spectrum, not just different points on a speed or cost spectrum. Lightning asks the least trust of any third party, a channel partner can only cheat within a narrow, penalty-enforced window, but it demands the most operational effort from users and node runners. Ark removes channel management and lets a single ASP batch many users together, at the cost of letting that ASP censor participation, even though it can never seize funds outright. Spark removes channel management entirely and adds arbitrary-amount transfers and offline receive, but it asks users to trust that at least one operator in a small, currently two-member set is honest.

None of these is simply more secure or less secure than the others in isolation; they are different answers to the question of how much inconvenience a user should tolerate in exchange for removing trust in a third party.

The centralization question nobody has solved

Every one of these designs eventually runs into the same tension: Bitcoin’s base layer is decentralized by design, but the layer built to make it usable at scale keeps concentrating around a small number of operators. Lightning is the furthest along and the clearest cautionary tale. An often-cited 2024 analysis found that close to half of all identifiable Lightning nodes were hosted on just two cloud providers, split roughly 29 percent to 19 percent between Amazon Web Services and Google Cloud, with the rest scattered across smaller providers and self-hosted setups. A node being cloud-hosted does not compromise the underlying channel security, since funds are still only spendable by whoever holds the private keys, but it does mean a large share of the network’s routing capacity could vanish in a single cloud outage or, in a more adversarial scenario, could be pressured by a subpoena or a change in a provider’s terms of service.

That is a strikingly similar shape to the concentration risk HOGE Wire has covered in Bitcoin’s mining pool landscape, where a handful of pools control the majority of hashrate despite thousands of individual miners contributing hardware. In both cases, the base protocol is permissionless, but the practical infrastructure running on top of it tends to consolidate for cost and convenience reasons that have nothing to do with the protocol’s design.

Ark and Spark did not solve this problem; they moved it earlier in the stack and made it explicit. Arkade currently depends on Ark Labs’ own ASP for the large majority of its activity, since third-party ASPs are only beginning to appear. Spark’s operator set consists of exactly two companies. Both projects say their operator sets will grow and decentralize over time, and both point out, correctly, that unilateral exit means a user’s funds cannot be stolen even in a worst-case scenario. But censorship resistance and liveness are different guarantees from theft resistance, and on those, a two-operator federation or a single ASP looks a lot more like a well-run business than a decentralized network, at least for now.

Rivals or plumbing? How Lightning, Ark, and Spark interconnect

Despite the framing of a three-way race, the networks are not entirely separate islands. Boltz, a cross-chain swap service, already lets users move value between on-chain bitcoin, Lightning, and increasingly Ark, without any party taking custody during the swap, the same atomic-swap plumbing that underpins USDT’s move onto Lightning through Tether’s Taproot Assets protocol, which Lightning Labs has continued to update with new SDK tooling through 2026. Multi-asset support is turning out to be a shared obsession across all three ecosystems: Lightning has both Taproot Assets and, separately, Tether’s newer RGB-based rollout; Ark has Arkade Assets; Spark’s leaf architecture was built with tokenized assets in mind from the start, part of why Grid can move both bitcoin and dollar balances through a single wallet address.

The more interesting long-term question is whether these become genuinely complementary rather than competing. Lightning’s biggest structural advantage is simply time in the market: thousands of nodes, tens of thousands of channels, and years of liquidity relationships that a brand-new ASP or operator set cannot replicate overnight. That makes it a plausible settlement backbone between larger players, exchanges, market makers, payment processors, even as Ark and Spark compete for the retail and application layer where channel management was always the biggest turnoff. A wallet could, in principle, hold Spark leaves for everyday spending, tap Boltz to move into a Lightning channel when it needs to reach a Lightning-only counterparty, and fall back to an Ark VTXO or an on-chain UTXO when it needs programmability channels do not offer. None of the three protocols is trying to prevent that kind of interoperability, and the wallets and SDKs shipping in 2026 increasingly assume multi-protocol support is the default rather than the exception.

The regulatory clock ticking in the background

None of this is developing in a regulatory vacuum, particularly for the stablecoin ambitions that Arkade Assets, Taproot Assets, and Grid all share. The GENIUS Act, signed into law in the United States on July 18, 2025, gave the law’s primary regulators, the OCC, the FDIC, the NCUA, the Treasury Department, FinCEN, and OFAC, exactly one year to finalize the rules that will govern who can legally issue a payment stablecoin to US persons and under what capital, reserve, and redemption conditions. That statutory deadline, set out in the bill text itself, falls on July 18, 2026, the day after this article was published, and as of this writing every agency’s rule remains a proposal rather than a final regulation. Comment periods only closed earlier this summer, and reporting on the rulemaking sprint has noted the law leaves no explicit mechanism for extending that deadline, even though agencies have missed statutory deadlines elsewhere before.

For a network like Arkade, whose flagship multi-asset use case is a Tether-backed stablecoin, or Grid, which is explicitly building dollar-denominated accounts on Bitcoin rails, the shape of the final GENIUS Act rules matters as much as which layer 2 wins on technical merits. Questions about whether a given token counts as a payment stablecoin, a security, or something else entirely sit at the center of a related fight in Washington over the CLARITY Act’s attempt to split digital asset oversight between the SEC and the CFTC, and over how regulators apply the Howey test to assets like VTXOs or Spark leaves that do not map cleanly onto any category that existed when securities law was written. None of that is likely to slow Ark or Spark’s shipping schedules, but it will shape which of their stablecoin ambitions can legally operate in the largest market in the world.

Choosing a rail: a developer’s decision framework

For a team actually deciding where to build in 2026, the choice mostly comes down to what kind of trust and operational burden they are willing to carry, not which protocol is objectively best.

Lightning still makes the most sense for high-volume settlement between businesses that can afford to run or rent proper node infrastructure: exchanges routing customer withdrawals, market makers, remittance companies, anyone who benefits from the network’s existing liquidity and its bilateral, penalty-enforced trust model. Running that infrastructure well is its own discipline, not unlike the choice HOGE Wire has covered between running a solo validator and delegating to a staking service in Ethereum solo staking: full control and minimal trust, in exchange for real operational responsibility. Ark fits teams that want Bitcoin-backed programmability and a straightforward unilateral exit without wanting to own channel liquidity, an appealing profile for wallets and application developers who would rather integrate with a single ASP than manage a routing graph. Spark, and by extension Grid, fits products where consumer-grade simplicity matters more than minimizing trust: offline payment receipt, arbitrary transfer amounts, and a stablecoin-friendly account model, at the cost of trusting a small operator set.

A few practical questions worth asking before committing to any of the three:

• Does the application need to reach existing Lightning-connected wallets and exchanges today, or can it onboard users onto a newer wallet entirely.
• How much operational capacity exists in-house to manage channel liquidity, versus outsourcing that to an ASP or operator set.
• Does the product need offline payment receipt, a Spark strength and a genuine Lightning weakness.
• Does the roadmap depend on stablecoins or other tokenized assets, and if so, which of Taproot Assets, RGB, or Arkade Assets already has the needed wallet and liquidity support.
• How much regulatory certainty is required before launch, given that the GENIUS Act’s final rules are still pending as of mid-2026.

What comes next

A handful of specific developments will settle some of these open questions before the end of 2026. LND v0.22 had not shipped as of mid-July, and its release notes will show whether Lightning Labs pushes native BOLT12 support past the experimental stage or continues to lean on LNDK as a stopgap. Arkade Assets’ USDT rollout is still pending; when it ships, actual on-chain volume will be the real test of whether Tether’s multi-front bet on Bitcoin stablecoin rails pays off evenly across Lightning, Ark, and RGB, or concentrates on whichever proves easiest to integrate. Grid’s stated goal of reaching 75 countries and 100 Visa markets by the end of 2026 is a concrete, checkable target for how far Lightspark’s statechain-based banking push has actually traveled.

The GENIUS Act’s implementing rules are due the day after this piece was published, and whether all six agencies actually hit that deadline, or join the list of regulators who have missed statutory rulemaking deadlines before them, will shape how aggressively stablecoin-carrying layer 2s can market themselves to US users and businesses over the back half of 2026.

None of that resolves the core architectural question this piece has tried to lay out: Lightning, Ark, and Spark are not simply competing for the same job. They are increasingly specialized tools, one optimized for trust-minimized settlement at scale, one for programmable Bitcoin-backed applications with clean exits, and one for consumer-grade simplicity backed by a small, named set of operators. The winner, if there is one, may just be whichever combination a wallet quietly stitches together behind the scenes.

Frequently asked questions

What is the Lightning Network and how does it work?

The Lightning Network is a Bitcoin layer 2 protocol that lets two parties open a funded payment channel, then exchange instant, low-fee payments off-chain by signing updated balance splits between themselves. Payments crossing multiple channels travel through Hash Time-Locked Contracts routed via onion encryption, so intermediate nodes forward value without seeing the full payment path. The channel only touches the Bitcoin blockchain again when it opens or closes, which is what lets Lightning process far more payments per second than Bitcoin’s base layer alone.

What is the difference between the Lightning Network and Ark?

Lightning requires two parties to open and fund a dedicated payment channel and manage its liquidity directly. Ark replaces that with a coordinator called an ASP, which batches many users into a shared on-chain UTXO and gives each of them a Virtual Transaction Output, or VTXO, representing their claim on that shared pool. Ark users can unilaterally exit to the base chain at any time using pre-signed transactions, without needing an open channel, but they depend on the ASP not to censor them from future batching rounds, a trust assumption Lightning’s bilateral channels avoid.

Is Spark the same thing as the Lightning Network?

No. Spark, built by the company Lightspark, is a separate Bitcoin layer 2 based on statechains rather than payment channels; ownership of off-chain balances moves by rotating cryptographic key shares among a federation of operators using FROST threshold signatures, not by opening a dedicated channel. Spark and Lightning can interoperate through swap services like Boltz, but they are distinct networks with different trust assumptions. Spark also should not be confused with the unrelated Ethereum-based lending protocol that also uses the name Spark and its SPK token.

Are Ark and Spark non-custodial?

Both are best described as trust-minimized rather than fully non-custodial in the way a Bitcoin wallet with a private key you alone control is. Ark’s ASP and Spark’s operator federation cannot outright seize a user’s funds because every balance is backed by pre-signed transactions the user can broadcast unilaterally, but both models depend on at least one honest party, an available ASP or one honest Spark operator, to avoid a slower, costlier fallback to an on-chain exit.

Will Ark or Spark replace the Lightning Network?

Unlikely in the near term. Lightning still carries far more public capacity, a larger base of connected wallets and exchanges, and years of accumulated liquidity relationships that neither newer network has matched. Ark and Spark are better understood as targeting the parts of Lightning’s experience, channel management, inbound liquidity, and offline receiving, that have proven hardest for ordinary users and app developers, rather than as wholesale replacements. Most 2026 wallets and infrastructure providers are building toward supporting multiple rails rather than betting on a single winner.

Written by the HOGE Wire research desk.