Dai: Black Thursday and the PSM dependency

Maker's design assumes that when a vault falls below its minimum collateralization ratio, anyone can liquidate it: the collateral is auctioned for Dai, the debt is repaid, and a penalty is applied. The assumption that quietly underpins all of this is that the auction will attract competing bids. On 2020-03-12 that assumption broke.

Two failures stacked on top of each other. First, the oracle: gas prices climbed so high that Maker's Medianizer could not afford to push fresh prices on-chain, so it stalled and then jumped down more than 20% in a single update — liquidating a large cohort of vaults at the same instant. Second, the auction: the keeper bots that liquidators relied on used a shared Maker reference script that had no logic for paying competitive gas, so they simply stopped bidding. Into that vacuum, a handful of participants who were willing to pay gas won auctions for zero Dai.

The numbers most cited from the post-mortems are roughly $8.32M of ETH walking out the door across about 1,461 zero-bid auctions, and a resulting system shortfall around 5.67M Dai (some accounts tally net losses closer to ~$4M–$6.65M depending on timing and method). Vault owners whose collateral was sold for nothing lost everything; MKR holders later voted, controversially, against compensating them.

Crucially, Maker did not need an emergency shutdown. The protocol already had a recapitalization path written into its mechanism: when collateral auctions fail to cover a vault's debt and the Dai buffer is empty, the system triggers a debt auction. Debt auctions mint new MKR and sell it for Dai, deliberately diluting MKR holders — the governance token absorbs losses that the stablecoin holders should not.

Starting 2020-03-19, those auctions opened at 250 MKR per 50,000 Dai and raised over 5M Dai, enough to close the shortfall, with Paradigm taking roughly 68% of the MKR sold. The MKR-as-backstop design did exactly what it was meant to do: it kept Dai whole at the cost of MKR dilution.

The structural fixes that followed were as important as the recapitalization. Maker overhauled its collateral-auction system (eventually replacing the old Flip English auctions with the instant-settling Clip Dutch-auction model), lengthened oracle update windows, added circuit breakers, and broadened the keeper ecosystem. The lesson was specific: a CDP can be perfectly overcollateralized on paper and still fail if its liquidation venue cannot function under congestion.

By 2023 Dai looked very different. To hold the peg tightly and scale supply, Maker had introduced the Peg Stability Module — a contract that swaps Dai for USDC (and similar stablecoins) one-to-one. The PSM made the peg far more robust in normal conditions, but it also meant that a large share of Dai's backing was simply USDC. By the time SVB failed, USDC made up roughly half of the collateral behind circulating Dai — over $2 billion sitting in the PSM.

When USDC depegged on 2023-03-11, the PSM inverted from a stabilizer into a transmission line. Arbitrageurs deposited cheap USDC to mint Dai and exit, draining PSM capacity; USDC deposits in the module nearly doubled while Dai supply ballooned. Dai followed USDC down to about $0.85. Maker governance approved emergency parameter changes — lowering the USDC PSM cap and raising its fee — within hours, but the changes sat behind a 48-hour timelock and only executed on 2023-03-13, after the federal backstop had already restored the peg.

The episode crystallized the dependency-map view of Dai: a nominally decentralized, crypto-collateralized stablecoin whose actual single largest risk was a centralized issuer's banking relationship. Maker initially voted to keep USDC as its primary reserve, then spent the following year diversifying across multiple stablecoins and real-world assets — the same direction the current collateral mix reflects.

Two failure modes, one mechanism. In 2020 the danger was endogenous — the liquidation system that defines a CDP could not execute under congestion. In 2023 the danger was inherited — the PSM collateral chosen to stabilize the peg carried someone else's reserve risk. A CDP stablecoin has to be evaluated on both axes at once: can it liquidate, and what is actually behind the peg-stability rails.

Survival is not the same as being unchanged. Dai recovered both times and never required an emergency shutdown, which is the strongest possible evidence the backstop design works. But each event rewrote the protocol — auction and oracle redesigns after 2020, reserve diversification and the eventual Sky restructuring after 2023. Resilience here came from a willingness to absorb losses (MKR dilution) and to restructure, not from the original design being flawless.