Investing in Stablecoins and On-Chain Yield

On the night of March 10, 2023, Circle disclosed that $3.3 billion of the cash backing USDC, about 8% of the reserve, was sitting in Silicon Valley Bank, which regulators had closed that afternoon. USDC is the second-largest dollar stablecoin and the one that had spent four years marketing itself as the transparent, regulated alternative to Tether. By two in the morning it traded at 87 cents. More than a billion dollars was redeemed on the Friday. DAI, which held USDC as most of its collateral, fell with it, and for a weekend the cleanest coin in crypto was worth thirteen cents less than the dollar it was supposed to be. On Sunday evening the FDIC guaranteed all SVB deposits, the $3.3 billion was whole, and USDC was back at a dollar by Monday.

Nobody who held USDC through that weekend lost a cent unless they sold. That is the whole subject in one scene. A stablecoin has no upside; you are lending a dollar to a company, or to a smart contract, for the convenience of a dollar that moves in seconds and settles at three in the morning. The only question that matters is what happens to that dollar when something breaks, and the answer depends on what the issuer holds, who can redeem, how fast, and whether you were paid enough for the wait.

By 2026 the market that survived SVB was a $300 billion one, its largest issuer was, by its own account, among the twenty biggest holders of US government debt, Congress had written a statute around the business, and the same coins were being offered to savers with 4% to 15% yields attached. This guide is about the difference between the coin and the yield. The coin, held correctly, is the one product in crypto that has worked. The yield is where most of the damage has been done.

Every yield, depeg and statute in this guide follows from one mechanism: how a dollar gets onto the chain and back off it. An issuer takes your dollar, puts it in a bank account or a Treasury bill, and hands you a token on a public blockchain that it promises to buy back for a dollar. The token then moves like any other crypto asset: between wallets, across exchanges, into lending contracts, at any hour, with settlement in seconds and a fee that is usually cents.

The dollar in the bill earns interest; under the model that has dominated since 2019, the issuer keeps the interest and the holder gets a dollar that does nothing. Tether’s 2025 profit of more than $10 billion, roughly $4 billion of it bill interest on a coin that ended the year near $187 billion (the rest was mostly gains on gold and bitcoin), is what a 4% bill yield looks like when the depositors are paid zero.

Three families exist, and the distinctions decide how a coin fails:

• Fiat-backed coins hold dollars, bills and repo: USDT (Tether), USDC (Circle), PYUSD (PayPal, issued by Paxos), RLUSD (Ripple), USD1 (World Liberty Financial), FDUSD (First Digital). They fail when the reserve is impaired (SVB) or when redemption is slow or restricted.
• Crypto-collateralised coins are minted against over-collateralised crypto loans, with a savings rate paid out of the protocol’s lending income: DAI and its successor USDS, both from Sky (formerly MakerDAO). They fail when the collateral falls faster than liquidations can sell it, or when the collateral is itself another stablecoin, as it was in March 2023.
• Synthetic dollars hedge a crypto position with a short futures position and pay the funding that leveraged longs pay to stay long: Ethena’s USDe. They fail when funding goes negative for long enough, when an exchange holding the hedge fails, or when an exchange’s own price feed misreads them, as Binance’s did on October 10, 2025.

A fourth family, the algorithmic coin backed by nothing but a sister token whose supply expands to defend the peg, is extinct. Terra’s UST was the largest ever built and it took five days to go from $18 billion to zero in May 2022; its founder, Do Kwon, was sentenced to 15 years on December 11, 2025.

Who is on the other side of your trade? On the primary market, only approved counterparties: Tether redeems for verified institutional clients with a $100,000 minimum and a 0.1% fee; Circle Mint is open to verified businesses and, since March 15, 2026, charges 2 basis points on daily redemptions between $40 million and $100 million and 5 above that. Everyone else trades on the secondary market, against other holders and against arbitrageurs who buy at 99.9 cents and redeem at a dollar. That arbitrage holds the peg on a normal day, and its absence breaks it on an abnormal one: on March 11, 2023 Circle’s redemptions ran through banks that were shut for the weekend, so nobody could complete the loop.

Before any yield is layered on, the return on a stablecoin held in a wallet is zero, minus the depegs, minus inflation, and the thing to compare it with is the Treasury bill it imitates. That is the design, and what it means for you is that a year in a coin is a year of Treasury interest forfeited to the issuer.

From July 2023 to September 2024 the Federal Reserve’s target range was 5.25% to 5.50%, so a holder of $100,000 in USDT or USDC was giving up something like $5,000 a year for instant settlement. As of the July 28–29, 2026 FOMC the range was 3.50% to 3.75%, held on a 9–3 vote with the three dissenters wanting a hike. On September 9, 2026 the three-month bill yielded 3.89% (Trading Economics), with futures putting roughly 63% odds on a quarter-point increase at the September 16 meeting. The forfeit is smaller than it was in 2024 and the arithmetic is the same.

The record of the peg itself is better than the reputation. Coin Metrics’ daily reference price for USDT bottomed at $0.950 on November 14, 2018, a month after the rumour-driven run of October 15 had printed $0.955, and before Tether published regular reserve reports; earlier prints of $0.82 in June 2015 and $0.85 in May 2017 came when the coin was under $100 million and barely traded. Through the Terra collapse it printed $0.9954 on May 11, 2022, and through the SVB weekend it never closed below $0.9999. USDC’s low is the $0.9706 reference of March 11, 2023 (the 87 cents was an intraday exchange print), and DAI’s is $0.9670 the same day, because DAI was mostly USDC. Between January 2024 and May 2026 neither USDT nor USDC closed a day below $0.9977. The synthetic and crypto-collateralised coins have a different record, covered in the depeg section.

What has moved is the supply, the cleanest measure of what the product is for. The ten coins Coin Metrics tracks summed to $26.0 billion at the end of 2020, $130.9 billion at the end of 2021, fell to $116.5 billion through the 2022 bear market as leverage left the system, then grew every year: $125.9 billion at the end of 2023, $195.1 billion at the end of 2024, and $296.8 billion on May 22, 2026. Public trackers that count every coin put the total near $322 billion at a May 2026 peak and about $302 billion on September 3, 2026, the first sustained decline of this cycle; treat those as secondary figures. The pattern that matters is that supply grew through a year in which, on our tape, bitcoin fell 53% from its October 7, 2025 peak to its July 1, 2026 low. Dollars stayed on the chain while the assets priced in them halved.

Two things about that chart. The 2022 decline of 11% is the only year of shrinkage, and it coincided with Terra, Celsius and FTX taking leverage out of the system; stablecoin supply is, at bottom, a measure of how much money wants to be in crypto without being in a crypto asset. And the composition moved: USDC was 32% of the total at the end of 2021 and 20% at the end of 2023 after SVB, then 26% by May 2026, while USDT went from 60% to 73% to 64%. Trust moved in response to specific events, which is the argument for reading reserve reports rather than brand names.

Two issuers, Tether and Circle, are 90% of the market, and they are different businesses wearing the same product. The rest share the remaining tenth.

Tether is a private company registered in El Salvador that issues USDT and reported more than $10 billion of profit for 2025 after about $13 billion for 2024; on analysts’ breakdowns of the attestation, roughly $4 billion of the 2025 figure was interest on the bills behind the coin and most of the rest was unrealised gains on gold and bitcoin. Its BDO attestation for the fourth quarter of 2025 showed $122 billion of direct Treasury holdings ($141 billion with overnight repo), $6.3 billion of excess reserves and, more controversially, about $17 billion of gold and $8 billion of bitcoin held as reserves. The second-quarter 2026 attestation (June 30, 2026) showed a $1.5 billion quarterly profit, $115 billion of Treasuries, $18.8 billion of gold, 98,932 bitcoin, $184.6 billion of USDT in issue, and an excess-reserve buffer that had halved to $4.11 billion from $8.23 billion a quarter earlier.

Tether has never published a full audit. It settled with the New York Attorney General for $18.5 million in February 2021 and with the CFTC for $41 million in October 2021 over misrepresented reserves; the CFTC found USDT fully backed only about a quarter of the time between 2016 and 2018, and that from June to September 2017 there was never more than $61.5 million behind roughly 442 million coins. USDT is not listed on MiCA-compliant European exchanges. On January 27, 2026 Tether launched USAT, a separate US coin issued through the OCC-chartered Anchorage Digital Bank, to live under GENIUS while USDT stays offshore.

Circle issues USDC and is the opposite in structure: a US public company (NYSE: CRCL since June 5, 2025, priced at $31, first-day close $83.23, a $299 peak on June 23, 2025) that publishes monthly reserve reports with a Big Four attestation over a reserve held mostly in a government money-market fund managed by BlackRock plus bank cash. Its second quarter of 2026, reported August 5, showed $701 million of revenue, of which $668 million was reserve income, USDC circulation of $73.3 billion at quarter end, and adjusted EBITDA of $143 million.

Coinbase, under a distribution agreement, booked $292 million of stablecoin revenue in the same quarter, about 44% of Circle’s reserve income, and Circle’s total distribution, transaction and other costs were $412 million, which is why its margin after them was 41.2%. Circle is a leveraged bet on USDC supply and short-term rates: the shares closed at $95.88 on September 8, 2026, about two-thirds below the June 2025 peak, on a business that grew.

Sky (formerly MakerDAO) issues DAI and USDS against over-collateralised crypto loans and its own Treasury-bill holdings, and pays a governance-set Sky Savings Rate to holders who lock USDS into sUSDS. USDS supply was $10.6 billion on May 22, 2026 on Coin Metrics’ data and about $10.0 billion at the end of June per Sky’s own quarterly release; gross protocol revenue was $107 million in the second quarter of 2026, and the savings rate was cut from 3.75% to 3.60% during that quarter to rebuild reserves. DAI, the older coin, has been flat at about $4.3 billion since 2024.

Ethena issues USDe, the synthetic dollar, which went from $5.6 billion on July 18, 2025 to a peak of $14.83 billion on October 4, 2025 and back to $4.44 billion on May 23, 2026; the basis section explains why.

PayPal’s PYUSD, issued by Paxos, peaked at $4.21 billion on March 6, 2026 and was about $2.7 billion to $2.9 billion by mid-June, down roughly 31%, after the incentives that built it were tapered; PayPal advertised a variable 4% reward in 2026. Ripple’s RLUSD passed $2 billion in 2026, about 57% of it on Ethereum on Ripple’s own figures. USD1, the Trump-affiliated World Liberty coin, was $4.7 billion in May 2026. Each is a rounding error against the two leaders and each exists to sell something else: payments, a settlement rail, a political brand.

The lesson of the issuer table is that “stablecoin” describes the promise, not the business. Tether is a bill fund with a gold and bitcoin side pocket and no audit; Circle is a bill fund with an audit and a distribution deal; Sky is a lender; Ethena is a hedge fund. You would not put those four in one asset class anywhere else, and you should not here.

On a normal day the peg is held by arbitrage, and that machinery, not the marketing, is what you are relying on at two in the morning. If USDC trades at $0.998, a market maker buys it, sends it to Circle, receives a dollar and keeps the difference; at $1.002 the same firm wires a dollar, mints a coin and sells it. The loop needs an issuer that honours redemption at par, a banking rail that is open, and a counterparty with the standing to use the primary market. You have none of the three directly. What you have is the secondary market, and it is only as deep as the arbitrageurs’ confidence that the loop will close.

The loop breaks in four ways, and each has a date. The reserve is impaired, or believed to be: Tether in October 2018, USDC in March 2023. The rail is shut: SVB failed on a Friday, and Fedwire does not run on weekends, so a coin whose redemptions settle in dollars could not be redeemed until Monday. The collateral behind a non-fiat coin falls faster than it can be sold: DAI on March 12, 2020, when an ether crash of more than 40% overwhelmed its auctions and some collateral was liquidated for nothing. And a venue misprices the coin: Binance on October 10, 2025 valued USDe from its own thin order book, marked collateral to 65 cents, and liquidated positions against a price that existed nowhere else; the depeg section has the full account.

Two more features matter to a holder. Every fiat-backed coin can be frozen at the address level by its issuer: Circle froze USDC in the addresses the US Treasury sanctioned with Tornado Cash on August 8, 2022 (the sanctions were lifted in March 2025; the freeze function was not), and Tether freezes USDT at law-enforcement request. A frozen balance is still “backed”; it is just not yours. And the coins live on many chains at once (Ethereum, Tron, Solana, Base and others), so the same USDT is a different token with a different fee and risk on each; a transfer to the wrong network is one of the commoner ways a beginner loses a balance outright. The chains themselves, as assets, are the subject of the sister guide Investing in Solana and Altcoins; here they are rails.

Reserve reports are short, the differences between them are large, and the market has punished holders who did not read them. Five questions extract what matters, once you have read the auditor’s word for what they did. An attestation is a point-in-time check that the numbers on a date reconcile to bank and custodian statements; an audit tests controls over a period. Circle’s monthly reports are attestations by a Big Four firm over a reserve that is almost entirely a government money-market fund and bank deposits; Tether’s quarterly reports are attestations by BDO over a reserve that includes gold, bitcoin, secured loans and “other investments”. Neither is an audit, and as of September 9, 2026 no large issuer has published one. GENIUS requires audited annual statements from issuers above $50 billion once in force.

Then ask five things. What is the reserve made of? The GENIUS standard is cash, bills of 93 days or less, overnight repo and government money funds. Tether’s June 30, 2026 sheet had $115 billion of Treasuries against $184.6 billion of coin, with the balance in gold ($18.8 billion), bitcoin (98,932 coins), repo and loans; that is a reserve with price risk in it. How big is the buffer? Excess reserves are the issuer’s own capital between a mark-to-market loss and the holders; Tether’s fell from $8.23 billion to $4.11 billion in one quarter, about 2.2% of supply, as it bought gold and bitcoin.

Where is the cash? Named banks and custodians and how much at each; SVB was one line in Circle’s report that nobody read until it mattered. Who can redeem, at what minimum, how fast, at what fee? If you cannot find the terms, the coin is not redeemable in any sense that helps you. Does the liability side match? Coins on the report should match supply on the chain.

The statute that turned stablecoins from a tolerated product into a licensed one was signed on July 18, 2025 and, as of September 9, 2026, was not yet in force. What it requires is settled; the argument it left open is yield.

The Guiding and Establishing National Innovation for U.S. Stablecoins Act passed the Senate 68–30 on June 17, 2025, the House 308–122 on July 17, and was signed on July 18, 2025 as Public Law 119-27.

Its terms are the reserve checklist written into law. A “permitted payment stablecoin issuer” must hold one-for-one reserves in cash, insured deposits, Treasury bills of 93 days or less, overnight repo and government money-market funds; publish the composition monthly with an accounting-firm examination; obtain an annual audit above $50 billion of issuance; give holders priority over every other creditor in an insolvency; and, the clause the banks fought hardest for, pay no interest or yield to holders. Issuers are limited to subsidiaries of insured banks, OCC-approved non-banks, and state-licensed issuers under $10 billion. The statute took the offshore model, Tether’s, and made the onshore model, Circle’s, the only legal one for a coin sold in the United States.

The dates are the part to memorise. The Act takes effect on the earlier of January 18, 2027 or 120 days after the primary regulators publish final rules. Each had until July 18, 2026 to finalise; none did. The OCC issued its proposal on February 25, 2026 (Federal Register, March 2; comments closed May 1); the FDIC’s proposal for bank-subsidiary issuers appeared April 10, 2026; Treasury, FinCEN and OFAC followed in April on money-laundering and sanctions rules. On August 19, 2026 Comptroller Jonathan Gould said the OCC intended to finalise by November and to process applications in the new year. Until a final rule appears, January 18, 2027 binds, and the coins you can buy in September 2026 are the same coins, under the same voluntary disclosures, as a year earlier.

The unresolved argument is yield. GENIUS bars the issuer from paying interest. It says nothing about the issuer’s distribution partner, which is how Coinbase pays about 4.10% on USDC balances (4.5% for Coinbase One members, as of August 2026) out of its share of Circle’s reserve income, and how PayPal pays a reward on PYUSD.

More than forty banking associations, led by the American Bankers Association, asked Congress and the OCC to extend the ban to affiliates and exchanges, arguing that a 4% instant-settlement dollar drains retail deposits; Coinbase and the exchanges argued the text bars only issuer-paid yield. The dispute stalled the broader market-structure bill through 2026 and is the most important open question for anyone earning a “reward” on a regulated coin: if the affiliate route is closed, the 4% ends the day the rule takes effect, and the coin reverts to what the statute intends, a dollar that pays nothing.

The banks are also building. JPMorgan rolled out JPMD, a deposit token (a tokenised claim on a bank deposit, which may pay interest because it is a deposit), to institutional clients on Base on November 12, 2025; Western Union launched USDPT on Solana in 2026 through Anchorage Digital Bank for its remittance network. The likely shape of 2027 is three tiers: licensed issuers under GENIUS, bank deposit tokens for institutions, and the offshore coin, USDT, serving the trading and emerging-market demand the statute does not reach. The flagship Investing in Crypto guide covers the wider regulatory picture, including the CLARITY market-structure bill; this guide stays on the coin.

Four failures define the risk, told here in order of what actually broke, and the pattern in them is the test to apply to every yield in the rest of the guide.

Terra, May 7–12, 2022

UST was an algorithmic coin: no reserve, only a promise that a dollar of UST could always be swapped for a dollar’s worth of newly minted LUNA. Anchor, its lending protocol, paid about 20% on UST deposits, and no borrower earned that 20%; it was subsidised from a reserve Terraform topped up with investors’ money. When large holders withdrew from Anchor and sold UST on May 7, each dollar redeemed minted more LUNA, LUNA’s price collapsed, and the collateral behind the promise went to zero in five days. UST had about $18 billion in issue in early May 2022 and about $40 billion of value disappeared across the two tokens, on the figures cited at Kwon’s sentencing.

Celsius, Three Arrows and Voyager failed within two months, and the supply chart above shows the only down year in its history. Kwon’s 15-year sentence on December 11, 2025 was for fraud: he had secretly propped the peg during an earlier wobble while claiming the algorithm worked.

USDC, March 10–13, 2023

Told in the cold open. What broke was not the reserve, which was whole by Monday, but the redemption rail. Note that a fully reserved coin lost 13 cents on the exchanges and nobody who waited lost anything; the loss went to the holders who sold into the panic and to DAI holders whose collateral was USDC.

USDe, October 10–11, 2025

In the largest liquidation cascade on record, about $19.4 billion of leveraged positions closed in a day after a tariff headline. Binance’s internal oracle priced USDe from its own $8 million order book instead of the $400 million pools on Curve and marked it to $0.65 while it traded at $0.99 elsewhere; $500 million to $1 billion of positions collateralised with USDe were liquidated against a price that existed only on that venue, and Binance later paid out a reported $283 million to the users its pricing had liquidated. The protocol was over-collateralised and its hedges held; the failure was the venue.

But the confidence loss was real: USDe supply fell from $14.83 billion on October 4 to $9.55 billion by October 31 and $4.44 billion by May 23, 2026, a 70% decline, because the funding yield that had pulled money in went with the leverage that paid it. Our tape shows why: bitcoin closed at $121,719 on October 10, 2025 and $113,044 the next day, and did not stop falling until $58,566 on July 1, 2026.

Stream Finance, November 3–4, 2025

Stream’s xUSD advertised double-digit yields from an “external fund manager” who lost about $93 million of the assets; xUSD fell 73% on the day of disclosure and later traded at seven to fourteen cents.

The contagion was the lesson. Lending vaults on Morpho and Euler had accepted xUSD as collateral at a hard-coded $1, so liquidations never fired while borrowers drained real USDC against worthless collateral; Elixir’s deUSD, which had lent $68 million to Stream (65% of its backing), fell to 1.5 cents; total exposure was put at about $285 million and Euler faced roughly $137 million of bad debt. No contract was hacked. A yield without a named payer was rehypothecated through three layers of “risk-managed” vaults until the loss landed on people who thought they were lending USDC.

Read the chart from the top. The two fully reserved coins lost a few cents on their reference prices and recovered in days; the synthetic coin lost 35 cents on one venue for eight hours and 70% of its supply over seven months; the coins whose yield had no payer went to zero. The pattern is exact enough to be a rule, and the next section turns it into one.

A dollar coin earns nothing by itself. A yield on it has to be paid by someone, there are only five candidates, and which one it is separates the yields that are income from the ones that are your own principal coming back to you slowly.

The Treasury, through the issuer. Tokenised money-market funds pass bill interest straight through: BlackRock’s BUIDL held about $2.5 billion to $2.9 billion depending on the tracker, Circle’s USYC about $2.6 billion to $3 billion, Franklin Templeton’s BENJI about $2.3 billion and Ondo’s USDY about $740 million in May 2026; the five largest funds together held about $6.8 billion on rwa.xyz’s dashboard, in a category that secondary surveys put near $15 billion that month once every product is counted. These are securities, sold to institutions and accredited investors (BUIDL’s minimum subscription is $5 million and it is limited to qualified purchasers; USDY is sold only to non-US persons), and they yield a bill fund’s return less 15 to 50 basis points. Payer: the US government. Risk: the fund’s custody, and the law that says you may not hold it.

The issuer’s distribution partner. Coinbase’s 4.10% on USDC (August 2026), PayPal’s 4% on PYUSD and Kraken’s default 1.75% are paid out of the reserve interest the issuer shares with the platform, as a marketing cost to keep balances there. Payer: the issuer’s margin, ultimately the Treasury again. Risk: the platform as counterparty, and a rate that is set unilaterally, tracks fed funds, and may be outlawed by the GENIUS rulemaking.

Borrowers. Aave, Morpho, Compound and Sky lend your coin to someone who has posted more collateral than they borrow and pass you most of the interest. Aave’s USDC rate on Base was about 3.6% in August 2026, its trailing thirty-day range across networks 3.8% to 5.2%; curated Morpho USDC vaults paid 4.5% to 6.5%; Sky’s savings rate was 3.60%. Payer: a leveraged trader, mostly. Risk: the contract, the oracle, the collateral, and the curator who chose them.

Leveraged longs. Ethena holds ether or bitcoin, sells the same amount of perpetual futures, and collects the funding that long traders pay to hold leverage, plus staking yield on the ether. sUSDe’s seven-day yield was about 9.4% in late April 2026 with a ninety-day average near 11.8%; early in 2026, after the October cascade, Messari had it at 3.72%. Payer: whoever is long on margin this week. Risk: funding goes negative, an exchange fails, a venue misprices you.

The next depositor. Anchor’s 20%, Celsius’s up-to-18%, Stream’s double digits, and every exchange “promotional” rate of 7% to 15% on a coin whose real lending rate is 4%. Payer: nobody, or you. Risk: total.

The chart says what the marketing does not: yields paid by the Treasury and by the issuer’s marketing budget sit within a quarter-point of the bill rate (Kraken’s default 1.75% is the exception, well below it), and everything above them is compensation for a risk with a name. On-chain yield is neither free money nor a scam; it is a menu of risk premia, and the only mistake is to take one without knowing which line you are on.

A lending protocol is a pool, and the pool sets the rate, which is why on-chain lending is the one yield above the bill rate that a non-specialist can defend, and why its losses have specific names. You deposit USDC and receive a receipt token that grows as interest accrues; a borrower deposits ether or bitcoin worth more than the loan, takes USDC out, and pays a rate the contract sets from utilisation, the share of the pool that is lent.

The curve is steep above a target utilisation, usually around 90%, so when demand for leverage spikes the rate spikes with it, which is why stablecoin lending paid 4% to 10% at the top of 2024–2025 and closer to 4% after the October 2025 unwind. If collateral falls toward the loan, anyone can repay it and seize the collateral at a discount; that liquidation protects lenders, and it depends on a price feed (an oracle) and on a buyer for the collateral in a crash.

Aave is the largest, $18.3 billion of deposits on September 9, 2026 on DeFiLlama’s count, in operation since 2017 with no depositor loss on its main markets that its own reserves did not cover. Morpho is the challenger at $9.8 billion, with outstanding loans at a record $5 billion on September 1, 2026 on Messari’s count, 95% of them in stablecoins, much of it on Base through Coinbase’s crypto-backed loan product, in which a Coinbase customer’s bitcoin collateral and USDC loan are actually a Morpho position. Sky’s Spark, at $6.8 billion, is third.

The difference between them is where risk is decided. Aave’s markets are governed collectively: one list of accepted collateral, one oracle standard, one safety module. Morpho’s markets are isolated and permissionless, and the choice of collateral is delegated to a curator, a named firm that earns a fee for it. That is where Stream landed: curators had accepted xUSD at a hard-coded dollar, the vault names said “USDC”, and the loss fell on depositors who had never heard of Stream. A curated vault paying 6.5% when Aave pays 3.6% is telling you its collateral is riskier; that is the whole reason for the extra three points.

The hack record is the other cost. Chainalysis counted $3.4 billion stolen across crypto in 2025, $2.02 billion of it by North Korean groups, most in one exchange breach (Bybit, $1.46 billion, February 21, 2025) rather than in lending contracts.

The first half of 2026 ran about $1 billion in DeFi exploits alone, led by a $292 million bridge exploit at KelpDAO on April 18, a $285 million key compromise at Drift on April 1, and a March 21–22, 2026 breach at Resolv in which an attacker with the signing key minted about 80 million unbacked USR against a deposit under $200,000 and cashed out roughly $25 million in ether in seventeen minutes. None touched Aave or Morpho’s core contracts, and that record is why a reader who wants a lending yield should be in the oldest, largest, most boring pool and nowhere else.

The one stablecoin yield that has paid double digits without being a fraud rests on a trade older than crypto: hold an asset, sell a futures contract on it, collect the difference. The number on Ethena’s front page is a trailing average of a rate that can go negative, and the mechanism explains both the size of the yield and its sign.

In crypto the futures are perpetual swaps that never expire and stay tethered to spot through a funding payment that flows, every eight hours, from the crowded side to the other. In a bull market the crowd is long on leverage and pays the shorts; Ethena stands on the short side with ether and bitcoin as collateral, collects the funding, adds the staking yield on the ether, and passes the total to holders who lock USDe into sUSDe (about 55% of supply in early 2026). The collateral sits with off-exchange custodians and the shorts sit on exchanges, which is the counterparty exposure: if an exchange fails with the hedge open, the dollar is no longer hedged.

The record on our tape is the record of leverage. USDe grew from $5.6 billion to $14.83 billion between July 18 and October 4, 2025 as bitcoin ran to its $124,740 peak on October 7; funding was rich because everyone was long. It then lost 70% of its supply in the months that bitcoin lost 53% and ether 66.7% on our tape, because the longs who paid the funding were the ones being liquidated. Funding flipped negative during the October 2025 crash and for stretches afterwards, meaning the strategy paid to hold its hedge; Ethena’s reserve fund covers such weeks (it stood at $61 million in March 2026 against $5.6 billion of supply, about 1.1%, per Eco), and the question is for how long. The 9.4% seven-day yield of April 2026 came back with leverage in the spring; the 3.72% of early 2026 is what the strategy pays when the crowd has gone home.

Our tape is one observation a day from CoinGecko, 374 observations from August 29, 2025 to September 8, 2026, and the IA Crypto sub-index is our own basket rebased to 100 on September 2, 2025 with a 9.7% weight in the composite; none of it is a market-wide index. What it shows is that the fully reserved coins were the only line that did not fall, and the synthetic one fell furthest. USDe never broke its peg on-chain through any of it; its holders lost the yield, which was the reason they held it.

The yield most beginners actually take is the one the exchange offers on the balance already sitting there, and the history of that product is short and expensive. Celsius paid up to 18% on some coins, high single digits on stablecoins, called it “Earn”, and filed for bankruptcy on July 13, 2022 owing customers about $4.7 billion; its founder, Alex Mashinsky, was sentenced to 12 years on May 8, 2025. Voyager filed on July 5, 2022 and BlockFi on November 28, 2022 with similar products; FTX followed on November 11, 2022. In every case the customer’s deposit became the company’s asset, was lent or invested at its discretion, and in bankruptcy the customer was an unsecured creditor waiting years for a partial recovery. The yield had been paid, in part, from newer deposits, and the “earn” label was the tell.

The 2026 versions are smaller and better disclosed. Coinbase’s USDC reward is funded by its share of Circle’s reserve income, is not a loan of your coin, and is paid on balances Coinbase custodies (its separate USDC lending product, advertised near 7% in 2026, is a loan, routed on-chain through Morpho); Kraken’s rewards are automatic at 1.75% and up to 3.75% for Kraken+ subscribers; Binance’s “Simple Earn” headline rates of up to 15% on USDC (December 2024 and March 2025 promotions) and up to 7% (2026) are promotional caps for short windows over a standing rate near the market lending rate.

The distinction is between a reward paid for a balance the platform custodies and a yield paid for a balance the platform lends out. The first is a marketing expense with custody risk; the second is Celsius with a better website. Read the terms for “lend”, “loan” or “rehypothecate”, and read them again after every update.

Either way the balance is an exchange balance: inside the company’s bankruptcy estate if it fails, subject to withdrawal halts, and beyond any deposit insurance. That is the cost of the 4.10%, and against Treasury bills at 3.89% on September 9, 2026, exempt from state tax, the worked example below finds it is not worth paying at all: the bills come out $58 ahead on $50,000 over three years.

Losing the principal outright, in stablecoins, happens not through the peg but through the wallet, the approval, the exchange and, increasingly, the front door. There are two places to keep a coin. On an exchange, where the platform holds the key and you hold a claim, the risks are the platform’s failure, its breach and its freezes; the Bybit theft of $1.46 billion on February 21, 2025, by a North Korean group that compromised a wallet provider’s developer machine, was the largest crypto theft ever and came out of an exchange’s cold storage. In your own wallet the risks are yours: a seed phrase photographed to the cloud, a signature granted to a malicious contract, a transfer to the wrong network, and the wrench. Investing in Bitcoin covers exchange tiers, multisig and inheritance in depth; what follows is the part specific to a dollar balance.

Chainalysis counted 46 violent attacks on holders through June 2026 taking more than $30 million, a pace that would beat 2025’s full-year record of $58 million, with France the hotspot at 30 incidents after a tax official allegedly sold dossiers on wealthy holders; home invasions rose from 14% of incidents in 2025 to 37%. A stablecoin balance is the most attractive target on a phone because it does not need to be sold.

The self-custody stack is cheap: a hardware wallet bought direct from the manufacturer ($79 to $169 in 2026 for the mainstream Ledger and Trezor models), a seed phrase on metal stored where you would store a passport, a separate hot wallet holding a week’s spending, and a habit of revoking contract approvals after every DeFi interaction. What it does not fix is the approval problem: depositing into Aave means signing a permission for the contract to move your USDC, and a fake site that looks like Aave asks for the same signature. Wallet-drainer phishing took about $84 million from some 106,000 wallets in 2025 on Scam Sniffer’s count, down 83% from 2024, and Permit-style signature approvals were the commonest tool in the largest cases.

Freezing is the risk nobody prices. USDT and USDC can be immobilised by their issuers at an address, with no published timeline for release; USDS carries a freeze function in its code that DAI never had, and USDe has none at the token level though its staked form, sUSDe, has a blacklist role, a feature or a bug depending on which side of the freeze you are on. For an American holder with a clean history the risk is small, not zero, and one more reason to keep the operating balance in a coin and the savings in a bank.

The costs are small enough to ignore and the tax is not. Buying is close to free. Coinbase converts dollars to USDC one-for-one with no fee; buying USDT on Coinbase Advanced costs 1.20% taker or 0.60% maker at the entry tier and 0.40% / 0.25% between $10,000 and $50,000 of monthly volume, Kraken Pro 0.40% / 0.25%, as checked September 9, 2026. Moving a coin costs cents on Base, Solana or Tron and a dollar or a few on Ethereum in 2026. Redeeming at the issuer is for institutions: Tether charges 0.1% on a $100,000 minimum and Circle Mint nothing on the first $40 million a day; retail sells on an exchange at a spread of hundredths of a cent. The all-in friction on a $50,000 round trip through a lending pool is under $100.

The tax is where the money is, and it has three parts. First, a stablecoin is property under IRS Notice 2014-21, so every sale, swap or payment is a disposal with a gain or loss, usually of a few cents; the bookkeeping, not the liability, is the burden. Since tax year 2025 brokers report gross proceeds on Form 1099-DA, and from 2026 cost basis for covered assets bought in the same account; a broker may aggregate “qualifying stablecoin” sales and omit them if they total $10,000 or less for the year, but you must report regardless.

Second, every reward, savings-rate payment, lending interest and funding distribution is ordinary income at its dollar value when you control it, at 10% to 37% federal plus state, plus the 3.8% net investment income tax above $200,000 of income ($250,000 married) where it applies; there is no long-term capital gains rate on stablecoin yield, because there is no capital gain.

Third, the accrual forms are unsettled: a receipt token that grows in value (Aave’s aUSDC, sUSDS, sUSDe) may be argued to defer income until redemption, and the wrap itself may or may not be a taxable exchange; the IRS has not ruled, and you should fix your position with a preparer before the first deposit. The wash-sale rule of §1091 did not apply to digital assets as of September 9, 2026, though bills to extend it were pending. Treasury interest, by contrast, is exempt from state income tax, a point or more of after-tax advantage in a high-tax state that no stablecoin yield can match.

Four ways to earn on $50,000 of dollars, with fees and tax, produce gaps that are smaller than the risks separating them: three of the routes land within $350 of one another over three years, and the fourth swings by thousands on a single event.

Assumptions, stated once: rates held constant at their August and September 2026 levels, monthly compounding, a 24% federal and 5% state bracket, no NIIT, all yield taxed as ordinary income in the year received, and the coin sold at exactly a dollar at the end. Every rate is variable and tracks fed funds, so the gaps below shrink if rates fall.

Route A: Treasury bills at 3.89%

The rate is the three-month bill on September 9, 2026; a government money-market fund would pay within a few basis points of it. Three years compounds $50,000 to $56,179, a gain of $6,179. Federal tax at 24% is $1,483; state tax on Treasury interest is zero. Net $4,696. Costs: none at TreasuryDirect; a fund would charge 0.09% to 0.15%.

Route B: Coinbase USDC reward at 4.10%

$50,000 becomes $56,532, a gain of $6,532, taxed at 29% federal plus state: $1,894. Net $4,638. Costs: none. What you take on: Coinbase as custodian of the whole balance for three years, a rate it can change on a Tuesday, and a rulemaking that may abolish it.

Route C: Aave USDC, own wallet, at 4.5%

The rate is the middle of Aave’s August 2026 range. $50,000 becomes $57,212, a gain of $7,212; tax $2,092; a hardware wallet at $129 and about $50 of network fees. Net $4,942. What you take on: contract, oracle and collateral risk, your own key, and a tax return with an accrual question in it.

Route D: sUSDe at 8.0%

The rate is chosen between the 3.72% of early 2026 and the 11.8% ninety-day average of April 2026, which is to say a guess. $50,000 becomes $63,512, a gain of $13,512; tax $3,918; the same $179 of wallet and network costs as route C; net $9,415 if nothing goes wrong. Now assume one event at the start of year two: a 15% loss, less than half of what Binance’s oracle did to USDe holders on October 10, 2025 and a fraction of what Stream did to its depositors. $54,150 after year one falls to $46,027, compounds back to $53,985 by the end of year three; the $12,108 of yield along the way is taxed as income ($3,511), and the $8,122 capital loss is deductible against ordinary income at $3,000 a year. Net after the income tax and the $179 of costs, and before the carryforward: $295, about 6% of what the bills paid, for three years of hedge-fund risk.

The chart is the guide in one picture. Routes A, B and C are within $350 of one another over three years on $50,000, and the order is not the one the marketing implies: after state tax the bills beat the Coinbase reward by $58, so the counterparty is being taken on for nothing, and the $246 Aave adds over the bills is the price of the contract and the key. Route D is the only one that changes the outcome materially, in either direction, and the direction depends on an event you cannot forecast. A reader who cannot say why they would take route D over route C should take route A.

A stablecoin belongs in operating cash, not in savings, and the steps below are in the order that keeps the principal safe.

1. Decide what the coin is for. Trading, moving money between venues, a dollar balance abroad, or a yield. If the honest answer is “a yield”, compare it with a Treasury money-market fund first; in September 2026 the gap was about a fifth of a point (4.10% on Coinbase against a 3.89% bill), and negative once state tax is counted.
2. Pick the coin by its reserve report, not its brand. USDC for a US holder, by the rule in the reserve section; USDT only as a trading balance where USDC is not quoted; nothing synthetic or crypto-collateralised in the operating balance.
3. Open the on-ramp. A Coinbase or Kraken account, bank-linked, with hardware two-factor authentication. Convert dollars to USDC at par; do not market-buy USDT at the 1.20% entry tier.
4. Buy the hardware wallet from the manufacturer and move everything beyond a week’s spending to it. Practise with $20 first, on the network you will actually use.
5. If, and only if, you want a lending yield, use the largest, oldest pool (Aave’s main USDC market), deposit from the hardware wallet, and revoke the approval after each interaction. Size it at no more than 10% of a crypto allocation that is itself 1% to 5% of investable assets; Investing in Bitcoin and Investing in Ethereum cover that allocation.
6. Set up the tax file the same day. A crypto tax package connected to the exchange and the wallet, and a written position on how accruing receipt tokens will be reported.
7. Write the exit rule. The next section’s thresholds are ours; adopt them or write your own before the next depeg, not during it.

The case for each coin is a set of published numbers, and these are the readings, with levels and dates, that would change our view.

The GENIUS final rules. An OCC final rule, promised for November 2026, starts a 120-day clock; absent one, January 18, 2027. Read the treatment of affiliate rewards: if Coinbase’s 4.10% survives, the regulated coin becomes a money-market fund with instant settlement; if not, exchange USDC balances fall and the yield case for the coin ends.

Tether’s buffer and composition. Excess reserves were $4.11 billion on June 30, 2026, 2.2% of supply, and gold plus bitcoin were about 13% of total reserve assets ($18.8 billion of gold and about $5.8 billion of bitcoin against $187.75 billion). Below $2 billion of buffer, or above 20% in gold and bitcoin, USDT is a sell for us. A Big Four audit would be the upgrade.

Circle’s circulation. USDC was $73.3 billion on June 30, 2026; two consecutive quarters below $70 billion would mean the regulated coin is losing to the offshore one even after the statute. CRCL at $95.88 (September 8, 2026) is the market’s verdict on reserve income at a 3.50% to 3.75% fed funds rate.

The Fed. Every yield here tracks the target range, 3.50% to 3.75% as of July 29, 2026 with a hike the majority expectation for September 16. At a 2% fed funds rate Tether’s profit halves, Coinbase’s reward falls toward 2%, and the only yields left above the bill rate are the risky ones; that is when the next Anchor gets launched.

Total supply on the public trackers. About $302 billion on September 3, 2026 against a May 2026 peak near $322 billion. Below $280 billion would be the first real contraction since 2022.

sUSDe’s seven-day yield and USDe supply. Yield below the bill rate: exit. Supply below $3 billion (the Coin Metrics low was $3.75 billion on April 28, 2026): the strategy is in run-off.

Aave utilisation and Morpho bad debt. Aave’s USDC market above 90% utilisation for a week is the top of a leverage cycle; any curated Morpho vault reporting bad debt is the next Stream, and the response is to withdraw first and read second.

The peg itself. USDC or USDT below $0.995 on Coin Metrics’ daily reference for two days, absent a known bank event, would be new; the 2024–2026 floor was $0.9977.