Stablecoin Reserves Are Safe. Their Transport Networks Aren't.

Can a dollar token be perfectly safe and still break? The obvious answer is no. After the Federal Reserve staff paper modeling stablecoin runs on congested blockchains, that answer needs a qualifier: safe on the balance sheet, maybe. Safe in practice, not always.

The Data Behind the Risk

Start with the number that jumps out. From 2021 through 2025, for below-median USDC transfers on Ethereum, the fee-to-value ratio at the 75th percentile frequently passed 100%. Paying more in gas than the value of the transfer itself. For above-median transfers, that ratio almost never exceeded 5%.

That's not a claim that users routinely paid $5 to send $3. But it shows how congestion rations access by transfer size. Small payments become uneconomic. A holder can wait, batch activity, move through a custodian or just leave the chain.

The Fed economists built a model where the stablecoin is fully and safely backed. No bad reserves. No liquidity mismatch. The fragility comes from somewhere else entirely: the interaction between transaction fees and network effects. People value a payment asset partly because other people accept it. When fees climb, use drops. When use drops, the network gets less attractive. When the network gets less attractive, more holders have reason to exit.

The empirical work backs the mechanism. A one-standard-deviation, $10.83 increase in gas was associated with a roughly 0.9 percentage-point rise in weekly redemptions, but only when network effects were low. That state covered just 7% to 7.5% of observations. And here's the kicker: gas by itself was statistically insignificant. It's the interaction that matters.

Another test matched 1,230 pairs of identical USDT transfers on Ethereum and Tron within 60-minute windows. The average matched transfer was about $176 million. Each $1 increase in lagged, demeaned gas correlated with 3% to 4% more net value moving from Ethereum to Tron. That's real migration, not just theory.

A Gap in the GENIUS Act

Here's where the policy question gets uncomfortable. The GENIUS Act does a lot of things right. It requires one-to-one reserves in specified liquid assets. It mandates public redemption procedures. Issuers have to disclose purchase and redemption fees, file monthly reports, submit to exams and meet capital, liquidity and operational standards.

All of that protects the dollar claim. None of it protects the route to that claim.

Treasury's proposed rule, published in the Federal Register on Aug. 18, focuses on section 3's restrictions on offering or selling payment stablecoins in the United States. Comments are due Oct. 19. The issuer licensing framework is expected to take effect Jan. 18, 2027. The digital asset service provider rules follow on July 18, 2028.

The distinction between direct self-custody transfers and compensated services like exchanges or custodians is useful. But the economics of a congested base layer persist across both categories. A reserve can stay liquid while a user still faces a transaction fee larger than the intended payment.

Nobody in Washington is setting a price or capacity standard for public blockchains. The Fed paper doesn't forecast a current run. But it sharpens a question that GENIUS implementation has to answer: if the dollar token is sound while access to it's broken, is the system actually safe?

Who Feels This First

Think about who gets hit by a fixed network charge. It's not the institution moving $50 million. It's the person trying to send $20 for a coffee or a utility bill.

That's the pattern the transfer data shows. Small users get rationed out first. They delay payments. They combine transfers. They move through exchanges. Or they stop using that chain entirely. The token remains redeemable at par. The user still can't transact economically.

When small users exit, the next moves come from bigger players. Exchanges, market makers, bridges and treasury desks can shift enough liquidity to change chain-level circulation. But that's a reallocation, not a solution. The destination chain inherits both the activity and the pressure. Its fee market then faces its own test.

Tron's fee structure is different from Ethereum's. Solana's is different again. A snapshot from Sept. 3 showed Ethereum fees around 0.127 gwei, implying roughly two cents for a 65,000-gas ERC-20 transfer. ETH traded near $2,404. Calm conditions. But calm conditions don't measure the system's behavior under stress.

And here's what worries me. The Fed paper finds that a sufficiently large congestion shock can turn individual exits into coordinated redemptions, even with perfect reserves. Is the next stress event going to look like a classic bank run on an issuer? Probably not. It's more likely to look like a silent migration event.

Enterprise blockchain is boring. That's why it works. The dollar claims are backed. The attestations are clean. But the rails stay congested and the fee markets stay unregulated.

What to Watch Now

The paper carries no policy prescriptions. It doesn't say regulators should cap gas fees or centralize block production. But it gives market participants a monitoring framework that didn't exist before.

Watch fee-to-value ratios by transaction size. That's the distributional metric that actually matters. Watch abrupt changes in chain-level stablecoin circulation, especially on Ethereum. Watch matched cross-chain flows when gas spikes. The ETH-Tron migration data already shows that channel works.

The test cases are coming. GENIUS's dealer and purchase agreement rules could be subject to both Regulation II and CFTC rules, which would be important for Treasury's upcoming determinations.

There's also the question of what counts as a redemption under the law. The Fed paper measures it as a drop in Ethereum circulation. That can mean a cash-out to fiat. It can also mean a migration to Solana or Tron. The data captures pressure on Ethereum-based circulation, not a clean count of customers cashing out.

Would a chain migration count as a bank run? In the model, yes. The network effects weaken. The token utility falls. The dollar claim is fine, but the digital dollar on that particular rail loses users.

GENIUS gives supervisors broad authority over issuer operational and technological risks. They can examine how an issuer manages rail exposure. They just can't control public blockspace. That's a gap no reserve attestation can close.

Trade finance ran on fax machines for decades because the document was the system. Stablecoins run on public rails because the ledger is the system. You can't regulate away the fee market without changing what makes the ledger public in the first place.

The container doesn't care about your consensus mechanism. It cares whether it moves. Same for a dollar token on a congested network.

So the next stress episode might teach us something awkward. The issuer was solvent. The reserves were real. The redemption promise was honored. And the users still couldn't get their money where it needed to go at a price that made sense. That's not a bank run. That's something worse.

It's a network that doesn't work when it matters most.