Open Wins. It Always Has.
--
On self-hosted wallets, permissioned chains, and why the industry keeps learning the same lesson.
There is a recurring pattern in crypto that does not get enough attention. Every few years, a group of institutions announce a new permissioned blockchain. The press release is professional. The consortium is impressive. The stated goals are cost savings, efficiency gains, and interoperability across the network. And then, a few years later, the project quietly stalls, pivots, or shuts down.
This has happened before. It is happening again now. And the self-hosted wallet data is showing us why.
The Exchange Custody Problem
Early stablecoin and crypto adoption moved through centralized exchanges. That made sense. Exchanges are easy to use, they handle key management, and they give users a familiar interface. They function as onboarding infrastructure. I have no issue with that framing.
The problem is that exchange custody is not where adoption matures. It is where it starts.
When I started tracking stablecoin flows, my hypothesis was simple: users would onboard through exchanges but transition to self-hosted wallets over time. The risk case for exchange custody is severe. Using a seven-channel risk framework across regulatory, reputational, reserve, redemption, routing, and rehypothecation dimensions, exchange custody scores high to very high on most of them. Self-custody primarily exposes users to regulatory and routing risk, at lower levels.
The real-world data has borne this out in a brutal way. Mt. Gox lost 850,000 BTC in 2014. Twelve years later, creditors are still waiting on repayment, with a deadline now extended to October 2026. FTX collapsed in November 2022 with an $8 billion shortfall affecting over a million users. A federal judge ruled in 2023 that the $4.2 billion in Celsius Earn deposits belonged to the company, not the depositors. In five months across 2022, roughly 4.3 million crypto investors lost approximately $46 billion across FTX, Celsius, Voyager, BlockFi, and Genesis. In February 2025, Bybit lost $1.5 billion in a single attack, the largest crypto heist in history.
The lesson from every one of these events is the same. When you hold crypto on someone else’s platform, you hold a claim on that platform, not the asset itself.
What the Data Actually Shows
Bitcoin held on centralized exchanges peaked at roughly 3.2 million BTC in early 2020 and has declined steadily since, falling to around 2.3 million BTC by late 2023. That drawdown represents a structural shift, not just a price cycle.
The self-hosted wallet side of the ledger has grown in parallel. MetaMask hit 30 million monthly active users in early 2024. Phantom surpassed 17 million MAU with over $25 billion in self-custodied assets. Ledger has sold 7.5 million hardware wallets and grew revenue 93% between 2023 and 2024. Hardware wallet unit sales grew 31% year over year in 2025. Self-hosted wallets are not a niche. They are the direction of travel.
The stablecoin data tells the same story, just with more precision. Stablecoin transaction volume reached $33 trillion in 2025, up 72% year over year. But the more important figure is behavioral. A 2024 Artemis study found that 47% of Ethereum stablecoin transfers are wallet-to-wallet payments rather than exchange or DeFi flows. In Argentina, where inflation ran above 140% in 2023, stablecoins represent over 61% of all crypto volume. In Brazil, 59%. In Nigeria, MetaMask’s user base has a 12.7% concentration. These are not speculators. These are people using self-hosted stablecoins as a financial tool.
In February 2025, when U.S.-Israeli strikes hit Iran, Nobitex saw a 700% outflow spike within minutes. Roughly $10 million fled Iranian exchanges to personal wallets. About 60% of those wallets still held the assets weeks later. That is not trading behavior. That is custody behavior.
The conclusion I drew, and still hold, is that exchanges are onboarding infrastructure. Self-custody reflects sustained adoption. Volume can mask where real activity is occurring. Behavior is the better signal.
The Permissioned Chain Trap
I want to be fair here. I do not think permissioned chains fail entirely. My view is more specific: they have historically struggled to expand beyond their initial networks, and that is the core assumption in the current narrative I would want to test.
The history is not encouraging.
R3 Corda launched in September 2015 with nine founding banks. By mid-2016 it had 42 members. By October 2016, Goldman Sachs and Santander had let their memberships lapse. JPMorgan left to build Quorum in-house. R3 cut its funding target from $200 million to $150 million and only secured $56.5 million of that. It has since gone through multiple restructuring rounds and staff cuts. Marco Polo, a trade finance network built on Corda with over 30 bank participants including BNP Paribas and HSBC, filed insolvency in February 2023 with roughly $5.7 million in debts.
Hyperledger Fabric, backed by IBM and the Linux Foundation, powered TradeLens. That project had 15 major ocean carriers covering over 60% of global containerized trade, 10 trade-finance banks, and 270+ port terminals. Maersk and IBM shut it down in November 2022. The official reason was that the project had not achieved the full global industry collaboration it needed. Per IBM’s own Hyperledger TOC reports, 81 to 82% of Fabric code activity still comes from IBM employees. Chainstack dropped Fabric node support in July 2024 citing low demand.
Libra is the sharpest example. Facebook launched it in June 2019 with 28 founding partners. In October 2019, Visa, Mastercard, eBay, Stripe, and Mercado Pago all withdrew in a single day under regulatory pressure. Facebook rebranded to Diem. Meta eventually sold the tech to Silvergate for $182 million in January 2022. Silvergate took a $196 million impairment charge on those assets and collapsed in March 2023. The engineers who built Libra went on to found Aptos, which launched as an open public L1.
The Current Cohort
This pattern is playing out again, with a new set of names.
JPMorgan’s Kinexys (rebranded from Onyx in late 2024) has processed over $3 trillion cumulatively since 2019 and now runs at roughly $5 billion in average daily volume. That sounds significant until you compare it to JPMorgan’s conventional USD payment engine, which processes approximately $10 trillion per day. After six years, Kinexys represents about 0.05% of JPMorgan’s own payment flows. In December 2025, JPMorgan quietly launched its JPMD deposit token on Coinbase’s Base, a public Ethereum L2. That is not a small footnote. It is an acknowledgment that institutional demand exists on public infrastructure.
Canton Network launched in May 2023 with approximately 30 institutional founding members including Goldman Sachs, Deutsche Börse, and Microsoft. It processes roughly 600,000 transactions per day across about 575 validators. Solana runs around 75 million non-vote transactions per day. Canton is more than 100 times smaller by that measure, after two years of operation.
Tempo, the Stripe and Paradigm-backed payments blockchain, raised $500 million at a $5 billion valuation and launched its mainnet in March 2026 with four team-run validators. Circle’s Arc blockchain unveiled in August 2025 is still in testnet as of April 2026. Both projects are, to their credit, signaling moves toward permissionless validation over time. That is exactly my point. The structural pull toward open systems is so strong that even the permissioned projects eventually feel it.
The irony I keep coming back to is this: these systems rely on open-source infrastructure but limit the open access that drives adoption. Ethereum added 16,000 new developers in the first nine months of 2025 alone. Solana added 11,500. These numbers do not have a permissioned chain equivalent because the permissioned chains do not have the token incentives to bootstrap a developer ecosystem.
What I Think Is Actually Happening
My view is not that permissioned systems have no value. They may find a durable niche in specific institutional settlement use cases where regulatory requirements make open participation impractical. Kinexys might be one of those.
But I do not believe these systems will expand beyond their initial networks at anything close to their stated ambitions. The open-source alternatives are improving too fast, the developer ecosystems are too large, and the network effects are on the wrong side of the comparison. The fact that JPMorgan went to Base, that Libra’s engineers went to Aptos, and that every 2025 permissioned chain is quietly exploring tokenization and open validation, tells you where the structural pressure is pointing.
Exchanges taught us that onboarding infrastructure is not the same as sustained adoption. Permissioned chains may be teaching us the same thing about institutional infrastructure.
Open wins. It has not stopped winning yet.
