Two contract addresses. One behind a proxy on Ethereum mainnet, one an older non-upgradeable ERC-20. Between them, more than $250 billion in dollar claims settle every day across exchanges, lending desks, and AMM pools that price their entire universe against these two symbols.
I pulled the transparency pages again last week. Then I pulled the mint functions. They do not look like each other, and neither of them looks like the "stablecoin" the market thinks it is trading.
The number everyone repeats this cycle is 85%. Tether and Circle control roughly 85% of circulating stablecoin supply, a level that sits near a multi-year high. It is the sort of statistic that gets read into a regulatory hearing and forgotten by the time the next block finalizes.
But 85% is the surface. The number that actually matters lives one layer down, inside an access-control role mapping that most holders will never open.
State root mismatch. Trust updated.
Context
A stablecoin, at the contract level, is not complicated. It is an ERC-20 with three privileged surfaces bolted on top: a mint authority, a burn authority, and a freeze or blacklist mapping. Everything else โ the name, the decimals, the transfer event โ is decoration.
For USDC, those surfaces are gated by a role-based access control system. A masterMinter role can add and remove minters up to a defined ceiling. A pauser role can halt all transfers. A blacklister role can freeze any address. And the contract itself sits behind an upgradeable proxy, which means the logic behind every one of those roles can be replaced by whoever holds the proxy admin key.
For USDT on Ethereum, the design is older and blunter. There is an owner address, an addBlackList function, and a destroyBlackFunds function that strips tokens from a frozen wallet and reduces total supply. There is no proxy on the original contract, but there is a deprecate function and a long history of migration proposals.
None of this is hidden. It is all in verified source. The point is that "85% market share" is not a story about two products winning a competition. It is a story about two key sets holding unilateral authority over the supply and the mobility of the dollar layer of crypto. The token is a receipt. The authority behind it is a corporate ledger.
The consolidation is not new. USDT launched in 2014 as Realcoin. USDC arrived in 2018 as a consortium product backed by a Circle and Coinbase joint venture. What changed is the regulatory perimeter. MiCA's e-money token regime took effect in mid-2024. The US GENIUS Act moved through 2025. Both frameworks raised the capital and compliance bar to a height that a distributed issuer simply cannot clear. The 85% is partly market-driven and partly engineered by rulemaking.
Core
Start with the arithmetic, because the arithmetic is where the narrative tends to break.
Market concentration is normally measured with a Herfindahl-Hirschman Index โ the sum of squared market shares. If two issuers each hold 42.5%, the HHI lands around 3,600 on a 0โ10,000 scale. Anything above 2,500 is classified as a highly concentrated market in standard antitrust practice. Stablecoins have been sitting in that band, and often above it, for years. The current print is near the historical ceiling.
Then add the layer the index cannot capture. HHI treats each firm as independent. In stablecoins, the two dominant issuers are not independent at the compliance layer. Both USDT and USDC blacklist addresses flagged by the same handful of chain-analytics vendors. Both respond to the same sanctions lists. Both have frozen funds in the same high-profile enforcement actions, sometimes within days of each other.
So the effective concentration at the policy layer is not 85%. It is closer to 100%. Holding both USDT and USDC is not diversification. It is the same instruction delivered to two extension phones in the same building.
This is the finding I keep returning to. When I traced the event-emission logic across the L2 standard bridge contracts in early 2024, the lesson was structural: the security surface is never where the user interface points. The interface shows a bridge; the risk sits in the wrapper's race condition. The interface shows a stablecoin; the risk sits in the role mapping.
Where the reserves actually sit
The reserve question is the one everybody asks and nobody answers cleanly.
Tether publishes quarterly attestations from BDO. Circle publishes monthly attestations from Deloitte. Attestation is not audit. An attestation is a point-in-time snapshot of specified accounts, reviewed under agreed-upon procedures. It does not test the completeness of liabilities. It does not opine on internal controls. It does not give the auditor unrestricted access to the full counterparty book.
Tether has never published a full audit under PCAOB standards. This has been true since 2017 and it remains true today. The industry has collectively decided to price around it, because USDT is the most liquid dollar in crypto and refusing it means refusing the market.
What an attestation can tell you: that at 11:59 PM on a given date, the reported assets exceeded the reported liabilities by some margin. What it cannot tell you: whether the liabilities are complete, whether the assets are encumbered, whether the custodian relationships are as described, or whether the next thirty days of redemption flow can be met without dumping illiquid paper into a falling market.
I have spent enough hours inside state-transition systems to know that a snapshot is not an invariant. The reserve is a balance-sheet claim evaluated off-chain, cleared off-chain, and reported on a schedule the issuer controls. The token contract knows none of this. The token contract knows only that a mint was called.
The off-chain settlement rail
Here is the part the on-chain framing hides. The blockchain is the thinnest slice of a stablecoin's stack.
When you redeem USDC through Circle Mint, you are not interacting with the ERC-20 contract in any meaningful sense. You are sending tokens to an issuer address, the issuer burns them, and a wire lands in your bank account after a compliance review. When Tether processes a large redemption, the mechanics are a treasury desk and a banking relationship, not a smart contract.
The settlement of the dollar claim happens at the corporate ledger. The token is a receipt on top of that ledger. The blockchain provides transferability and composability; it does not provide the finality that institutions actually care about.
This distinction matters because it reframes every "on-chain" risk metric. The 85% concentration is not a property of Ethereum. It is a property of the banking relationships and the licensing regimes that sit underneath the tokens. You cannot fix a banking-concentration problem with a better contract.
The DeFi dependency graph
Stablecoins get described as infrastructure, which undersells them. They are the substrate.
Walk the graph. Every major AMM routes through a stable pool, historically Curve's 3pool and now often a USDC/USDT pairing with a third asset. Every lending market denominated in dollars prices collateral against a stablecoin unit. Every perp exchange settles margin in USDT. Every OTC desk quotes against a stablecoin pair.
Remove one edge and the graph does not degrade gracefully. It reconfigures violently. The March 2023 USDC depeg is the cleanest natural experiment we have. Circle disclosed $3.3 billion of its reserve sitting at the failed Silicon Valley Bank. USDC traded as low as $0.877 on some venues. Curve's pools skewed hard. Arbitrageurs burned and redeemed at par by the end of the weekend, but only because the Fed backstopped SVB depositors and Circle's remaining reserves were intact.
The reflexivity is the point. A reserve shock does not stay in the reserve. It transmits into every pool that quotes the token, which means it transmits into every position that uses those pools as a price oracle, which means it transmits into every protocol that treats the oracle as truth. One disclosure, one weekend, and a trillion-dollar graph re-priced.
Opcode leaked. Liquidity drained.
Blacklist forensics
Watch the freeze counters. They tell a cleaner story than any transparency report.
The August 2022 Tornado Cash sanctions were the first large-scale demonstration. Circle complied and froze USDC held in sanctioned wallets; Tether initially resisted, then complied under pressure. The two issuers, nominally competitors, converged within a short window because they answered to the same enforcement authority.
Since then the pattern has repeated. Sanctioned entities, bridge hackers, and darknet-linked addresses have all seen balances frozen on both major tokens. Each freeze is a small demonstration that the supply of the dollar layer is discretionary. The transfer you sign is final on-chain, but it is reversible at the issuer level, and at scale that reversibility is indistinguishable from a permissioned system.
This is also where the correlation becomes measurable. If you plot blacklist events from both issuers on one timeline, you do not see two independent policies. You see one policy with a short lag. That lag is the only competitive difference, and it is shrinking.
The mint-burn asymmetry
Supply control has a second dimension that gets almost no attention: the direction of the flow.
Minting increases circulating supply. Burning decreases it. In a fractional banking system, this is monetary policy. Two private companies control the aggregate supply of the dollar unit in crypto, subject only to demand and to their own compliance thresholds.
When demand surges, the issuers expand supply and collect the float income on the reserves. When demand collapses, the issuers contract supply and the system has to find new dollars elsewhere. That is a procyclical mechanism running inside a market that marketing describes as a hedge against inflation. The hedge is administered by a treasury desk, and the desk answers to a board.
Stablecoins as the real settlement L2
One more framing, because it matters for how the infrastructure competition gets scored.
The L2 wars are fought over throughput and fees, but the layer that actually moves dollars between counterparties is the stablecoin rail. Every rollup eventually settles its users' value claims through an exchange or a bank, and both route through USDT and USDC. In that sense the most-used settlement layer in crypto is not an OP Stack chain or a ZK rollup โ it is a role mapping with a banking relationship behind it. The real difference between L2 stacks was never cryptographic. It is who can convince more projects to route through them first. The stablecoin layer won that contest by a mile, and it won it with a centralized issuer.
What the market is actually pricing
Here is where I part ways with most of the commentary.
The market prices reserve risk. That is the visible risk โ the one with a transparency page and a social media account. The market does not price the upgrade path.
Both the USDC contract and USDT's successor designs are reachable by an administrative key. In USDC's case, the proxy admin can point the implementation at new logic. There is no timelock on the upgrade. There is no on-chain governance vote. There is a key, held by a corporation, that can rewrite the rules of the contract that the entire ecosystem uses as its unit of account.
This is not a hypothetical exploit. It is a deliberate design choice, and it is what makes the token legally an e-money product rather than a bearer instrument. Regulators require the issuer to be able to freeze, seize, and reverse. That ability is exactly the ability to rewrite the contract.
The blind spot is that almost nobody models it. Risk frameworks model depeg. They model reserve shortfall. They do not model "the implementation address was updated at block N and the semantics of transfer changed."
I made a version of this mistake myself. In 2025 I modeled DA-layer slashing conditions and assumed validator independence. The simulation showed that under consolidation, the economic-security assumption collapsed faster than the throughput assumption. The stablecoin case has the same shape. The security model assumes two independent issuers. The actors are correlated by regulation, by vendors, and by the same enforcement calendar.
โ ๏ธ Deep article forbidden
The regulatory transmission channel
The regulatory layer is not external to any of this. It is a load-bearing beam.
MiCA's e-money token regime requires issuers to hold reserves in segregated accounts with EU credit institutions, to redeem at par, and to submit to supervision. The GENIUS Act in the US imposes reserve-composition rules and a federal or state licensing path. Both frameworks read, on paper, as consumer protection. In operation, they are a moat.
I have written before that exchange fines function as regulatory licenses โ Binance emerged more entrenched after its $4.3 billion settlement, because the compliance overhead it now carries is a wall no newcomer can climb. The same logic applies one layer down. The compliance cost of issuing a regulated stablecoin is now high enough that the addressable set of issuers is small and shrinking. Tether and Circle are not the incumbents despite regulation. They are the incumbents because the regulation selected for them.
This produces a strange incentive loop. The more the market worries about concentration, the more it asks for regulation. The more regulation arrives, the more concentrated the market becomes. The drain is circular, and it is running in the direction of fewer issuers, not more.
Contrarian
The consensus critique of this cycle is that stablecoin concentration creates systemic risk, and that decentralized stablecoins are the answer.
Both halves are suspect.
The systemic-risk claim is true but mis-specified. The danger is not that one issuer fails. The danger is that both issuers behave identically at the moment of stress โ and because they are correlated at the compliance layer, they will. A decentralized stablecoin outside the two-issuer compliance perimeter solves a different problem than the one the market believes it is solving. It solves censorship resistance. It does not solve the concentration of the dollar-denominated settlement rail, because that concentration is a property of the underlying base money and the banking system, not of the token.
The second half is worse. "Decentralized stablecoin" is a narrative category that has, so far, produced over-collateralized capital-inefficient designs and fractional-algorithmic designs that imploded. It has not produced a unit that a pension fund will settle in. Treating its revival as the fix misreads what institutional users actually want, which is a predictable freeze-and-seize regime, not its absence.
The real blind spot is simpler. The industry has spent a decade arguing about reserves and almost no time arguing about the upgrade key. The reserve question is eventually auditable. The upgrade question is answerable today, in verified source, and almost nobody asks it.
Takeaway
Watch three signals, not the price.
First, the concentration. If the two-issuer share falls below 70%, the HHI drops out of the high-concentration band and the policy-layer correlation loosens. That is the only number that changes the structural picture.
Second, the upgrade path. If either issuer adds a timelock to its proxy admin, that is a genuine reduction in unilateral authority. If neither does, the reserve attestations are the smaller of two problems.
Third, the freeze cadence. Count blacklist events per quarter. If both issuers freeze the same addresses inside the same window, the effective concentration is confirmed, and the "diversification" of holding both is an illusion.
As AI agents begin settling autonomously in stablecoins โ a shift already visible in 2026 oracle and payment prototypes โ the upgrade key becomes a machine-to-machine authorization question, not just a legal one. A smart contract that trusts a stablecoin is trusting a role mapping it cannot see and a corporate board it cannot vote on.
The question is not whether stablecoins are centralized. That was settled years ago. The question is whether the next decade of dollar settlement on public chains will be governed by two boards, or by a role mapping the market can read, verify, and replace.
State root mismatch. Trust updated.