Hook
Two issuers. Roughly 85% of the dollar-pegged float. I have been computing the Herfindahl-Hirschman Index for stablecoin issuance every quarter since DeFi Summer 2020, and the number has never left the 6,800โ7,200 band. The US commercial banking sector โ the industry that gave us antitrust law โ runs between 1,000 and 1,400.
Read that again. The settlement asset for the most permissionless financial system ever built is roughly five times more concentrated than the banking system it was designed to circumvent.
That is not a headline. It is a structural constant. And structural constants are where analysts stop looking, because they feel like weather rather than news. What follows is an attempt to price the weather.
Context: How I Built the Dataset
Methodology first, because methodology is the only part of stablecoin research that cannot be astroturfed.
I pull mint and burn events directly from issuer contracts on Ethereum, Tron, Solana, Base, and Arbitrum โ the five chains carrying more than 90% of the float. I join those events against monthly and quarterly attestation PDFs, then cross-check reserve composition against US Treasury auction data. Where attestations are quarterly, I interpolate. Where they are monthly, I do not. The asymmetry in attestation cadence is itself a variable in the model, not a footnote.
Two issuers matter here. Tether runs the offshore model: dollars into a set of reserve managers, tokens out, quarterly attestations, and in most jurisdictions no enforceable legal claim on the issuer. Circle runs the regulated model: monthly attestations, a French EMI license under MiCA, and a corporate treasury that behaves like a small money-market fund. Different legal wrappers. Identical product.
And the product is a claim on a bank deposit or a Treasury bill, wrapped in a token, sold as infrastructure. The token is not the asset. The bank account is the asset. Everything downstream of that sentence is an engineering detail.
Why does 85% matter? Because stablecoins are the denomination layer. Between 60% and 70% of on-chain DEX volume pairs against a stablecoin, the majority of perpetual futures collateral is a stablecoin, and every DeFi lending market prices risk in one. Concentrate the denomination layer and you have concentrated the entire credit surface of the ecosystem into two corporate bank accounts.
Core: The Evidence Chain
I want to be precise about what the data does and does not show, because precision is the only defensible position in a market this reflexive.
Finding one: supply velocity is trending down, not up. Mint events over the last four quarters are roughly flat in count but larger in average size. Fewer, bigger mints. That is the signature of institutional treasury operations โ a handful of desks moving nine figures โ rather than retail demand. Retail usage has migrated into yield-bearing wrappers that do not show up in the issuer's headline float. The 85% figure measures issuance, not usage. That distinction is where most public analysis breaks.
Finding two: the concentration is not an accident of competition. It is the equilibrium of it. Liquidity is a network effect with negative returns to fragmentation. A trader does not want a decentralized dollar; a trader wants the dollar with the deepest exit and the tightest spread. Given two functionally identical claims, demand flows entirely to the deeper one, because depth is the only differentiator that matters when you are unwinding a nine-figure position. This is not a market failure. It is the market working. Which makes it far harder to fix with better code.
Finding three: MiCA made this worse, and it did so by design. This is the part I would flag for anyone building a stablecoin business in Europe. MiCA's reserve rules require 1:1 backing with a cap on the share held at commercial banks, plus custody and segregation requirements. On their own, those are reasonable. But the compliance cost curve is largely fixed โ legal, audit, custody, capital, reporting, the EMI licensing minimums โ which means per-unit compliance cost falls as float grows. A stablecoin with $500M outstanding carries essentially the same regulatory overhead as one with $50B. MiCA is not a neutrality layer. It is a scale filter. Small issuers are not killed by the reserve rules. They are killed by the fixed cost of proving they follow them.
On-Chain Truth: Three Stress Tests, One Lesson
In May 2022, UST broke because its collateral was endogenous. In March 2023, USDC traded to $0.87 because $3.3B of its reserves sat at Silicon Valley Bank. In the same window, USDT briefly traded at a discount on Curve. Three different mechanisms, three different failure narratives โ and an identical on-chain signature: the assets that recovered were the ones with a redemption path to a large, liquid, exogenous collateral pool. The decentralized option was not the safe option. The liquid option was.
Early Warning Indicators
The checklist I actually run, weekly:
- Issuer float change above 3% in seven days without a matching rise in aggregate DEX volume โ signals redemption pressure, not demand.
- Stablecoin borrow rates on Aave or Morpho diverging from issuer redemption throughput โ signals a queue forming.
- Attestation date slippage beyond the stated cadence. Tether's cadence has drifted before; each slip is a repricing event.
- Stablecoin share of DEX pair volume falling below 55% โ signals rotation into a non-USD denominator, the only scenario that genuinely threatens the 85%.
- Any single reserve custodian exceeding 20% of an issuer's attested holdings.
- MiCA-passported EMI additions in the EU falling to zero quarter-over-quarter โ signals the scale filter has fully closed.
None of these are price signals. All of them are plumbing signals. Correlation is a whisper; causation is a scream โ and screams come from the plumbing.
Contrarian Angle: The Risk Is Not in the Contract
Here is where I part ways with most of the commentary.
The reflexive take is: 85% concentration equals systemic risk, equals regulatory crackdown, equals the decentralized alternatives finally get their moment. Three steps, each reasonable, and the chain is wrong in the middle.
The transmission channel for stablecoin risk is not a smart contract. There is no reentrancy bug in this story, no oracle manipulation, no MEV extraction. There is a redemption queue at a custodian bank. A stablecoin run is not an on-chain event; it is a bank run with a blockchain veneer โ and unlike a bank run, it has no deposit insurance, no lender of last resort, and no resolution framework. It has a Telegram channel and a legal opinion. Somebody's Solidity audit is irrelevant to whether $20B can exit in 48 hours.
The second mistake is treating concentration as evidence of failure. It is evidence of product-market fit. The market discovered the cheapest, most liquid dollar and bought it. Anyone who claims a decentralized alternative would capture share on the merits of its architecture has not watched what happens to the spread on a Curve pool when depth halves. Users do not pay for ideology. They pay for slippage.
The deepest blind spot is that we have been measuring the wrong thing. Decentralization in stablecoins was always a claim about governance and redemption rights, not about the count of issuers. A market with twelve issuers, all holding the same T-bills at the same custodians and all redeemable at the same banks, is not decentralized. It is diversified in name and concentrated in substance. Opacity is the original sin of valuation โ and an issuer that publishes an attestation quarterly is asking you to price eleven weeks of nothing.
Takeaway
Watch the threshold, not the narrative. If the combined share holds at 85% through the next two quarters, the debate is settled: the dollar layer belongs to two companies, and regulation will rationalize that outcome rather than reverse it. If it breaks below 70% on the back of a genuine non-custodial denomination rail โ not a tokenized T-bill in a different wrapper โ then the architecture argument finally has evidence behind it.
Mathematics respects no community, only consensus. The consensus right now is that stablecoins are infrastructure. The math says they are two balance sheets wearing a token.
The question I keep failing to answer: if 85% of the settlement layer is two corporate treasuries, what exactly was decentralized โ the issuance, or just the marketing?