One Custodian, 84% of the ETFs: Crypto Custody in 2026

The phrase that built a generation of crypto, not your keys, not your coins, was meant as a warning against trusting institutions with your money. In 2026 it reads like an irony. The spot Bitcoin and Ethereum exchange-traded funds that pulled Wall Street into the asset class, the single biggest demand story of this cycle, overwhelmingly park their coins with one company. Coinbase custodies roughly 84% of the assets in the US spot Bitcoin ETF complex, worth tens of billions of dollars, and its chief executive calls that dominance a strength.

That tension sits at the centre of any honest custody comparison in 2026. The question is no longer whether regulated custody is safe; it largely is. Charters, audits, insurance and cold storage have matured to the point where losing coins at a qualified custodian is genuinely rare. The real question is what happens when almost everyone uses the same one. This piece compares the custody models, then the firms, and explains why 2026 became the year issuers quietly began handing a second set of keys to someone else.

The number that should worry you

By mid-October 2026, with Bitcoin trading around $80,900 and Ether near $2,363 after a hotter-than-expected inflation print knocked risk assets lower (Yahoo Finance), the spot ETFs had become Bitcoin’s dominant on-ramp. They also share a dominant custodian. Data cited by Forbes earlier in the year put Coinbase’s Bitcoin ETF holdings at about $77.1 billion, roughly 84% of the entire $91.7 billion US spot Bitcoin ETF complex (Forbes). Research firm Bitwise made the same point bluntly: Coinbase is the primary custodian for the large majority of US spot ETFs (Crypto Briefing).

Translate that share into failure modes. If Coinbase suffered a serious operational outage, a crippling breach or an abrupt regulatory halt, it would not hit one fund. It would hit the custody layer under most of the market at the same moment. Forbes called it a choke point, and the label stuck.

Coinbase’s own chief executive does not dispute the share, only the framing. “We do have pretty dominant market share in terms of custody for the ETFs. I see that as a strength,” Brian Armstrong said on a February company call, adding that Coinbase holds “80% plus market share” of US Bitcoin ETF custody and that the firm is “the trusted counterparty on the institutional side” (Bitcoin Magazine). He is not wrong that Coinbase is good at the job. The worry is structural rather than about competence: a market whose founding slogan warns against single points of trust has rebuilt one at industrial scale.

Custody is not one thing

Before comparing firms, it helps to be precise about what custody even means, because the word covers at least seven very different arrangements. They differ along two axes that people constantly conflate: technical control (who can actually move the coins) and legal recourse (who is liable, and who gets paid, when something goes wrong).

At one end sits pure self-custody: you hold the private keys on a hardware or software wallet, nobody can freeze or seize your coins, and nobody will help you if you lose the seed phrase. Our field test of MetaMask, Phantom and Rabby covers the software end of that spectrum. Collaborative multisig splits signing authority, usually two-of-three, between you and a key agent such as Casa or Unchained, so no single party can move funds alone. MPC (multi-party computation) wallets, the institutional default, never assemble a full private key at all: shards sign jointly, off-chain. Exchange custody means your balance is an IOU on a company’s internal ledger. A qualified custodian is a regulated entity, usually a trust company or national trust bank, that holds client assets under a legal duty to keep them segregated. Bank custody is the newest branch. And the ETP or spot ETF wrapper outsources all of it: you own a security, and a custodian holds the coins behind it.

The table below maps that spectrum. Read the right-hand columns together: as you move down toward institutions, your technical control falls while your legal recourse, at least on paper, rises.

What a qualified custodian actually is

Qualified custodian is not marketing language; in the US it is a legal status with teeth. Under the SEC’s custody rule for investment advisers (Advisers Act Rule 206(4)-2), a registered adviser that holds client assets must place them with a qualified custodian: a bank, a registered broker-dealer, or a trust company that meets specific standards. That rule is why an ETF issuer cannot simply keep coins in a hot wallet on a laptop. It must use an entity a regulator will recognise, with segregation, audits and a fiduciary duty attached.

In crypto, three charter types do the heavy lifting. The first is a New York limited-purpose trust company supervised by the New York Department of Financial Services (NYDFS); Coinbase Custody Trust Company, chartered in 2018, is the archetype and a fiduciary under New York banking law. The second is a national trust bank chartered by the Office of the Comptroller of the Currency (OCC); Anchorage Digital Bank won the first such approval for a crypto firm in January 2021 (CoinDesk). The third is a state trust company, the model BitGo used before its own federal upgrade.

Two policy reversals in 2025 turned the US from hostile to hospitable for bank custody, and they explain why 2026 looks the way it does. In January 2025 the SEC rescinded Staff Accounting Bulletin 121 and replaced it with SAB 122, ending the rule that had forced banks to carry safeguarded client crypto as a balance-sheet liability, an accounting treatment that made custody uneconomic for most lenders (XBRL). Weeks later the OCC issued Interpretive Letter 1183, confirming that national banks may custody crypto, and later buy and sell it on a customer’s behalf, without seeking prior supervisory sign-off (OCC). Remove the accounting penalty and the permission hurdle, and the banks arrive. That is the backdrop to the comparison that follows.

The big five custodians, compared

Five names dominate institutional crypto custody in 2026: Coinbase, BitGo, Anchorage, Fireblocks and Komainu. They are not interchangeable. Two are trust-company custodians in the classic sense (Coinbase and BitGo), one is a national trust bank that was early to the charter (Anchorage), one is a technology layer that many banks and exchanges run but that is not itself a qualified custodian (Fireblocks), and one is a multi-jurisdiction specialist built by a global bank and a wallet maker (Komainu). The table compares them on charter, model, insurance and role. A warning on the insurance column: these are company-reported crime or specie policies that cover theft and key loss, not the custodian’s own insolvency, and none of them is FDIC or SIPC coverage.

Coinbase: the default, and the choke point

Coinbase became the default for one unglamorous reason: it was ready. Coinbase Custody Trust Company has operated as an NYDFS-chartered fiduciary since 2018, runs a cold-storage framework the firm traces to 2012, and carries SOC 1 and SOC 2 Type II attestations; its institutional arm markets exactly the segregation and reporting an ETF sponsor’s lawyers demand (Coinbase). When BlackRock, ARK and most other issuers filed to launch spot Bitcoin funds, Coinbase was the counterparty that already cleared the bar. The result is the 80-plus-percent share Armstrong defends.

On security, the pitch is credible. Armstrong has pointed to cold-storage systems that are regularly penetration-tested and audited, to custody-related patents, and to in-house cryptographers hardening the stack (Bitcoin Magazine). There is no public evidence that Coinbase has lost ETF coins, and the firm has never suffered a cold-storage breach of client assets. The problem is not Coinbase’s security; it is the arithmetic of everyone depending on it.

By one widely cited estimate, a serious operational failure at Coinbase could affect nine of the eleven US spot Bitcoin ETFs at the same time. Concentration converts a company-specific risk into a market-wide one: a single breach, outage or court order would propagate across most of the product category at once. The lone structural exception among the big issuers is Fidelity, which self-custodies the Bitcoin behind its FBTC fund through its affiliate Fidelity Digital Assets rather than outsourcing to Coinbase, keeping the vast majority of coins in air-gapped cold storage (The Block). It is the one top-tier issuer that holds its own keys, and, as we will see, even Fidelity added a backup name in 2026.

BitGo: the public, federally chartered challenger

If there is a credible number two, it is BitGo, and 2026 was its breakout year. BitGo pioneered enterprise 2-of-3 multisig custody in the 2010s and had long held state trust charters. In December 2025 the OCC granted it a national trust bank charter, and on 22 January 2026 BitGo listed on the New York Stock Exchange under the ticker BTGO, becoming, in its own words, the first public, federally chartered digital-asset infrastructure company (BitGo). The listing paired a public balance sheet with a federal charter, a combination none of its rivals could claim.

BitGo then bought growth. In August 2026 it acquired NYDIG’s institutional trading business for up to $57.5 million in cash and stock, bringing roughly 30 staff and a book of institutional trading relationships, and bolting derivatives and financing onto its regulated custody core (Markets Media). The strategic logic is that custody alone is a low-margin utility; the money is in the trading, lending and settlement that sit on top of safe storage.

BitGo matters here for a second reason: it is the custodian issuers increasingly reach for when they want a name that is not Coinbase. It is the primary custodian for Grayscale’s Bittensor product, a settlement rail in the FTX creditor distributions, and, per early-2026 filings, a named backup custodian for Fidelity’s FBTC. A public, federally chartered, multisig-rooted custodian is precisely the counterweight a concentrated market needs. Whether issuers actually shift meaningful assets to it, rather than merely naming it, is the open question of the dual-custody era.

Anchorage, Fireblocks and Komainu: the rest of the field

The other three names fill distinct niches. Anchorage Digital Bank is the purist’s institutional custodian: the first crypto firm to win an OCC national bank charter, back in January 2021, it secures keys in hardware security modules released only by a biometric quorum of officers, and it layers staking and on-chain governance on top of storage. Its 2026 significance is that BlackRock chose it as the second custodian for both its Bitcoin and Ether funds, the clearest institutional vote yet that one custodian is not enough.

Fireblocks is the odd one out, and the nuance matters. Fireblocks is not a qualified custodian; it is a software platform built on MPC, in which, as the company puts it, the full private key is never created, never stored and never assembled. Banks and exchanges license Fireblocks and remain the custodians of record themselves. That design removes single-key risk, but it introduces a different concentration: one technology vendor sits underneath more than eighty banks and exchanges, so a flaw in that shared layer would ripple across many institutions that look independent from the outside. Diversifying custodians does little if they all run the same software.

Komainu is the reminder that custody is global, not an American story. A venture of the Japanese bank Nomura, the asset manager CoinShares and the hardware-wallet maker Ledger, Komainu has been regulated by the Jersey Financial Services Commission since 2019 and later added a Dubai VARA licence, a UK registration and an Italian footprint, holding client assets in segregated Ledger Vault cold storage with staking inside custody (Ledger Insights). For a European or Gulf institution that wants regulated custody outside the US charter system, Komainu is the obvious call.

The dual-custody turn: 2026’s quiet correction

Through 2026 a pattern repeated in the filings: an issuer adds a second custodian and describes it, almost word for word, as risk management as the trust grows. The language is bland; the signal is not.

The most recent example is Grayscale’s Bittensor Trust, the vehicle that holds the TAO token behind the Bittensor network. On 29 September 2026 the trust became a party to the Coinbase Prime broker agreement, adding Coinbase as a second custodian alongside its primary, BitGo, a change disclosed to the SEC in an 8-K in early October; the sponsor noted it had not yet determined how much TAO it would move (SEC). BlackRock had done the mirror image earlier in the year, naming Anchorage Digital as an additional custodian for its IBIT and ETHA funds in an April filing, again citing ongoing risk management as the trusts grew (etf.com). Even Fidelity, the self-custody holdout, named BitGo Bank & Trust as an additional custodian for FBTC.

There is a catch that keeps the correction honest: most of these are additions on paper, not migrations of coins. BlackRock’s own quarterly filing said it had no current plans to move Bitcoin to Anchorage, and Grayscale left the amount blank. Naming a backup custodian is not the same as splitting the keys. Jean-Marie Mognetti, chief executive and co-founder of CoinShares, was blunt about why that matters: “Right now they are all using one custodian, which is Coinbase, creating a massive concentration risk,” he told a CoinDesk panel, adding that “from a protection and diversification point of view, it’s a zero. If you were in any hedge fund, you would want to get a number of prime brokers to diversify your risk” (CoinDesk). The dual-custody turn is real, but in 2026 it is mostly an intention.

Why concentration is a systemic risk, not a Coinbase risk

It is worth separating two claims critics often blur. The first is that Coinbase is unsafe. There is little evidence for that. The second is that the market’s dependence on Coinbase is unsafe regardless of how good Coinbase is. That one is true by construction. A system with a single dominant node fails in a correlated way: the same event, a court order freezing assets, a zero-day in a shared system, a liquidity crunch at the parent company, hits everything downstream at once. Farmers call it monoculture, and it fails the same way in finance as in agriculture.

The reach runs further than most ETF buyers realise. When a European investor buys a physically backed Bitcoin ETP through a Nordic or German broker, the coins behind that wrapper are often held, at the end of a chain of issuers, by the very same US custodians. BlackRock’s funds route to Coinbase; a retail holder of an iShares product therefore inherits the identical concentration, several legal layers removed and entirely invisible on the brokerage screen. Concentration at the custody layer does not stay in the institutional tier. It flows down to the smallest retail position.

Some commentators have reached for a Lehman analogy, asking what happens to the whole market if the dominant custodian stumbles. The comparison is imperfect: a qualified custodian holding segregated client assets is legally very different from an investment bank levered against its own book. But the instinct behind it is sound. Resilience comes from redundancy, and in 2026 the custody layer still has very little of it. That is why the slow, paper-first dual-custody turn matters more than its modest size suggests: it is the market starting to buy itself insurance it does not yet think it needs.

The bankruptcy test: who actually gets the coins back?

Strip away the branding and one question decides everything in a crisis: when a custodian fails, are your coins legally yours, or are you just another creditor? The answer turns less on technology than on the fine print, and recent history is blunt about it.

The Celsius bankruptcy set the precedent. In January 2023 Judge Martin Glenn of the Southern District of New York ruled that assets deposited into Celsius’s Earn program were property of the bankruptcy estate, following the plain terms of the user agreement, which made Earn depositors general unsecured creditors at the back of the queue, while customers who had used Celsius’s separate custody product were treated differently (CoinDesk). The lesson is that the label on the account, not a user’s intuition about whose coins they are, governs who gets paid.

FTX drove home a second lesson: even a successful recovery can be a loss. The FTX Recovery Trust has returned roughly $10 billion to creditors across 2026, with major customer classes reaching 100% or more of their allowed claims (Crypto Briefing). But claims were dollarized at the November 2022 petition date, so a creditor who had held Bitcoin on FTX is repaid the 2022 dollar value, not the coins, and misses the rally that followed. A 100% dollar recovery can still be a heavy loss denominated in Bitcoin.

This is where a genuine qualified custodian differs from an exchange balance. Client assets held in segregation by a trust company or trust bank are meant to be bankruptcy-remote, outside the custodian’s estate and beyond its creditors; an omnibus balance on a trading platform is not. Crucially, no government backstop fills the gap: SIPC and FDIC do not cover crypto, which is neither a bank deposit nor a registered security (SIPC). The European Union went further here. Under MiCA’s Article 75, a crypto-asset service provider must legally segregate client holdings from its own estate and is liable for their loss, a statutory protection the US has not matched with a single federal custody law (EUR-Lex). The same getting-it-back question even shadows staked assets, as our look at Lido, Rocket Pool and Frax shows: wherever someone else can delay or deny the return of your coins, you are making a custody decision, whether or not it is labelled one.

Proof of reserves is not proof of solvency

After Mt. Gox and FTX, exchanges rushed to publish proof of reserves, and the phrase has been doing misleading work ever since. As Nic Carter, who helped popularise the Merkle-tree method, has long cautioned, a proof of reserves demonstrates assets, that the custodian controlled certain coins at a snapshot in time, but it says nothing about liabilities (Nic Carter). A firm can prove it holds 100,000 BTC and still be deeply insolvent if it owes customers 200,000, or if the coins were borrowed the night before the snapshot and returned the morning after.

Vitalik Buterin framed the missing half precisely: a real proof of solvency combines proof of assets with proof of liabilities, showing both that customer deposits equal some amount X and that the firm controls X in coins (vitalik.eth.limo). Without the liabilities side, a cryptographic reserve attestation is theatre. Newer zero-knowledge schemes aim to prove both without exposing customer data, but adoption is uneven.

There is a live tension between verification and security, and Coinbase sits on one side of it. Chief financial officer Alesia Haas has said the firm will not publish the wallet addresses holding ETF coins, citing security and client confidentiality, while maintaining that custody clients can independently verify their holdings on-chain (Bitcoin Magazine). Both things are true at once: disclosing addresses would help the public audit reserves, and disclosing addresses would hand attackers a map. The result is that outsiders mostly take reserve claims on trust, backstopped by SOC audits and the custodian’s reputation rather than by open verification.

This is why verifiability deserves to be treated as a third axis of custody, alongside technical control and legal recourse. It is also why the audit layer matters: independent attestations, and the smart-contract reviews that firms such as OpenZeppelin perform on the multisig and MPC systems underneath, are often the only external check a client ever gets. Trust, in regulated custody, has not been eliminated. It has been relocated to the auditors.

MiCA, the SEC and the regulatory gap

The deepest reason US custody concentrated is that, unlike the European Union, the US never wrote a single custody law for crypto. What it has instead is a patchwork: the SEC’s adviser custody rule, NYDFS trust charters in New York, OCC national trust banks at the federal level, and a scatter of state trust companies. An ETF issuer picking its way through that maze rationally chooses the custodian everyone else already uses and regulators already know. The patchwork does not merely tolerate concentration; it quietly rewards it.

The SEC’s own posture flipped in 2025. Repealing SAB 121 removed the accounting rule that had kept banks out, and the agency has signalled a more accommodating line on digital-asset custody than the enforcement-heavy stance of prior years. Congress added the GENIUS Act, a federal framework for payment stablecoins, in 2025, while a broader market-structure bill that would clarify how tokens are custodied and classified stalled in the Senate. The upshot is that in 2026 the rules are friendlier but still incomplete, and completeness is exactly what a diversified custody market needs.

Contrast that with MiCA. Article 75 hands EU clients a statutory segregation-and-liability regime by default, so the protection a US investor must infer from a trust company’s charter and terms of service is, in Europe, written into the law that licenses the custodian. Neither system is automatically safer in practice, since a well-run NYDFS trust can be sounder than a thinly capitalised EU licensee, but the EU default is clearer. Custody rules are also where the stablecoin fight is being settled, since reserves have to be held somewhere by someone, which carries the same custody questions straight into the dollar-token world. For now, the American investor’s safeguard is mostly contractual, and the burden of reading the contract sits with the investor.

How to choose, whether you hold $500 or $5 billion

Custody is not a single best answer; it is a matching problem. The right model depends on who you are afraid of and what you need to be able to prove. Run any option past the three axes this piece has used: how much technical control do you keep, what legal recourse do you have if the holder fails, and can you verify the assets are really there?

For an individual, the old maxim still holds. If you can secure a seed phrase, self-custody removes counterparty risk entirely, which is why keys, not computing breakthroughs, remain the real attack surface, a point our piece on quantum computing and Bitcoin makes at length. The failure mode simply flips from someone else’s bankruptcy to your own mistakes: lost seeds, no inheritance plan, a careless phishing signature. Collaborative multisig is the sensible middle for larger personal holdings, removing the single point of failure of one seed while keeping you in control.

For an institution, the lesson of 2026 is plural. One qualified custodian, however good, is a concentration; the professional answer is at least two, with assets actually split rather than merely a second name added to a filing. That is the discipline Mognetti was demanding and that the BlackRock and Grayscale filings gesture toward without yet delivering. And whoever you use, the contract is the product: the Celsius ruling proved that the terms of service, not the marketing, decide whether you own your coins or merely hold a claim.

A short checklist, whatever your size:

• Separate what you hold for sovereignty (self-custody) from what you hold for trading (an exchange), and never confuse the two.
• Read the account terms for the words custody and segregation; an Earn-style yield product is a loan to the platform, not safekeeping.
• If you rely on a custodian, ask who the backup is, and whether assets are genuinely split or just dual-named in a filing.
• Remember that SIPC and FDIC do not cover crypto; the only backstop is the custodian’s private insurance, which covers theft, not failure.
• Treat verifiability as a feature: prefer custodians that offer meaningful attestation, and never mistake a proof of reserves for a clean bill of health.

The maturing of crypto custody is real and worth celebrating; the infrastructure that lost billions in 2022 would not lose them the same way today. But maturity brought its own fragility. The market solved the problem of whether institutions could hold crypto safely and replaced it with the problem of too many of them holding it in the same place. The fix is unglamorous and already underway: more custodians, assets genuinely split, contracts actually read. In 2026 the keys are safer than ever; the open question is simply whose hands they are concentrated in.

Frequently Asked Questions

Who holds the Bitcoin behind spot Bitcoin ETFs?

Most US spot Bitcoin ETFs use Coinbase as their custodian, which holds an estimated 84% of the assets in the US spot Bitcoin ETF complex. BlackRock added Anchorage Digital as a second custodian in 2026, and Fidelity is the main exception, self-custodying its FBTC fund through Fidelity Digital Assets.

Is my crypto safe with a qualified custodian?

A qualified custodian such as Coinbase, BitGo or Anchorage holds client assets in segregation under a trust or national-bank charter, which is far safer than an ordinary exchange balance. However, SIPC and FDIC do not cover crypto, so the only backstop is the custodian’s private theft insurance, which does not protect you against the custodian’s own insolvency.

What is the difference between self-custody and exchange custody?

With self-custody you hold the private keys yourself, so no third party can freeze or seize your coins, but you alone are responsible if you lose the seed phrase. With exchange custody the exchange controls the keys and your balance is an IOU on its ledger, which can become an unsecured creditor claim if the exchange goes bankrupt.

Why are ETF issuers adding a second custodian in 2026?

Issuers are adding a second custodian to reduce concentration risk, the danger that one custodian holding most of the market becomes a single point of failure. Grayscale added Coinbase Prime alongside BitGo for its Bittensor trust, and BlackRock named Anchorage Digital for its Bitcoin and Ether funds, though in most cases the assets have not yet been moved.

Does proof of reserves mean an exchange is solvent?

No. Proof of reserves shows only that a firm controlled certain assets at a snapshot in time; it does not show the firm’s liabilities. A real proof of solvency combines proof of assets with proof of liabilities, so a reserve attestation alone cannot confirm that a custodian can cover everything it owes customers.

Marcus Okafor is a markets correspondent at HOGE Wire, covering custody, market structure and the plumbing of crypto’s institutional era.