Riot Platforms retired a $200 million bitcoin-backed credit facility from Coinbase Credit on September 21. The collateral — 5,821 BTC — was released. The security interest was terminated. No prepayment penalty was paid. Riot's secured debt went to zero.
That is the clean version.
Here is the arithmetic that is not clean. The filing marks the pledged collateral at roughly $340.7 million. Divide that by 5,821 coins and the implied mark is $58,530 per bitcoin.
Riot drew the first $100 million of this facility in April 2025 and upsized to the full $200 million the following month. The repayment lands roughly five months after origination. Bitcoin did not print $58,530 anywhere inside that window.
Three explanations survive contact with the numbers. The valuation date on the collateral schedule is older than the filing implies. The coins were marked with a haircut that was never disclosed at that scale. Or the dollar figure is stale residue carried forward while the coin count is current.
Each possibility rewrites the leverage math. Only one of them is harmless.
Data over drama. Start with the integers, not the headline.
What the facility actually was
Strip the language. This is not a DeFi position. It is a bilateral, off-chain, centrally intermediated credit line — Coinbase Credit lending against bitcoin it custodies, with a legal security interest over the coins.
No smart contract. No oracle. No liquidation bot. No reentrancy surface. The risk lives somewhere else entirely: in custody, in counterparty solvency, and in the discretion of a lender to demand collateral when a threshold is crossed.
The structure is worth spelling out because the market persistently confuses the two species of bitcoin lending.
A MakerDAO-style CDP vault is a rules engine. The collateral ratio is public. The liquidation price is deterministic. The oracle feed is the failure point, and everyone can see it. You are trusting code and a price feed.
Coinbase Credit is a credit desk. The terms are private until disclosed. The margin-call trigger is contractual, not algorithmic. The collateral sits with the lender. You are trusting a balance sheet and a legal agreement enforced in a jurisdiction.
The trade-offs invert cleanly. On-chain lending fails through oracle manipulation and liquidation cascades. Off-chain lending fails through counterparty insolvency and custody. Neither is safe. They are unsafe in different directions, and you size them differently.
Riot borrowed at a fixed 6.15% annual rate. That single number tells you more about the deal than anything else in the disclosure.
Fixed-rate secured paper to a high-beta borrower at 6.15% is not a predatory rate. It is a good rate — the kind of rate you get when the lender has already decided the relationship is worth more than the spread.
That distinction matters, and I'll return to it.
The timeline is the tell
Walk the sequence.
Credit facility established. First $100 million drawn in April 2025. Facility upsized to $200 million in May. Full repayment executed September 21. Security interest released on settlement. Total exposure window: approximately five months.
Five months is not a financing. Five months is a bridge.
Long-duration corporate debt against a treasury asset gets structured over three to five years, with covenants tied to collateral ratios and periodic re-marking. A facility drawn and retired inside two quarters behaves like working capital — a liquidity buffer against a specific, time-bound need, then unwound.
What was the need? The disclosure doesn't say. But the shape of the instrument points at something transactional rather than structural. A capex tranche. A vendor settlement. A bridge ahead of equity issuance or an asset sale.
And the exit was cheap. No termination fee. No prepayment penalty. That clause is not standard generosity. It is a negotiated term, and it tells you which side had leverage at the drafting table.
A borrower with weak bargaining position does not get penalty-free early repayment on a fixed-rate facility five months into the term. The lender is giving up $12.3 million of contractual annual interest. Coinbase agreed to that. Read into it what you will, but read something.
The negative carry nobody priced
Now the economics, which are simple and brutal.
$200 million at 6.15% fixed costs $12.3 million per year. That is roughly $1.025 million per month. About $33,700 per day.
What does the borrowed capital earn? Nothing. Bitcoin does not pay a coupon. It does not distribute a dividend. It has no cash flow. Holding BTC is a zero-yield position, and financing it at 6.15% creates a guaranteed negative carry of 6.15% per annum before custody costs and before price movement.
This is not an opinion about bitcoin. It is arithmetic about the financing structure.
I ran this exact math on my own book. During the DeFi Summer of 2020 I deployed $200,000 into liquidity positions because the headline APY printed triple digits. By August, impermanent loss had consumed 40% of principal even while token prices appreciated. The lesson was not "avoid yield." The lesson was that I had been comparing a headline yield to a financing cost without netting out correlation and slippage. The number on the banner is never the number in your P&L.
Same error here, scaled up. A leveraged bitcoin treasury is a long position with a funding cost attached. The levered holder's break-even is not "bitcoin goes up." It is "bitcoin goes up by more than 6.15% annualized, plus custody, plus the opportunity cost of the equity slice."
In a strong uptrend that hurdle is trivial. In a flat or declining market it is a slow bleed with no floor. And in a bear tape, a fixed-rate financing cost on a zero-yield asset is the definition of a position that has to be closed or defended.
The 2.42x that Riot just gave back
Here is what the facility actually did to Riot's exposure.
If the collateral was marked at $340.7 million against $200 million of debt, the borrower's equity slice in that collateral stack was $140.7 million. Control of $340.7 million of bitcoin for $140.7 million of own capital is 2.42x effective leverage on the treasury position.
At 30% BTC appreciation, the equity slice returns roughly 72% before carry. At 30% depreciation, the same slice returns roughly minus 72% before carry. Symmetric, violent, and entirely dependent on a price you do not control.
Now add the LTV constraint. Loan-to-value on this facility sits near 58.7% — $200 million of debt against $340.7 million of collateral.
Industry practice for miner secured lending generally sits between 40% and 60%. At 58.7%, Riot was at the aggressive end of the band, not the conservative end. That is not reckless, but it is not a cushion either.
Run the stress test.
| BTC drawdown from mark | Collateral value | Resulting LTV | Likely lender action | |---|---|---|---| | 0% | $340.7M | 58.7% | none | | 10% | $306.6M | 65.2% | watch | | 16% | $286.2M | 69.9% | approaching typical trigger | | 20% | $272.6M | 73.4% | margin call territory | | 27% | $248.7M | 80.4% | collateral top-up or forced reduction |
Read the middle row twice. A 16% bitcoin drawdown puts a 58.7% LTV loan against a conventional 70% trigger threshold.
Bitcoin prints 16% moves routinely. Not in tail scenarios. In ordinary quarters. A mining treasury carrying half its coins as pledged collateral was one ordinary correction away from a collateral top-up demand, in a market where the miner's own equity had already correlated downward with the same asset.
That is the real reason the loan got retired. Not a directional call on bitcoin. A reduction of the probability-weighted cost of a margin event.
What happened to the balance sheet
The structural change is the part worth respecting.
Before repayment, roughly 5,821 BTC — about 51% of Riot's implied treasury of ~11,414 coins — sat under Coinbase's security interest. Riot held a contractual claim, not control. It could not sell those coins, pledge them elsewhere, or deploy them without lender consent.
After repayment, the full stack moved to unencumbered status. Free coins went from roughly 5,593 to roughly 11,414. Debt went from $200 million to zero.
That is a meaningful reclassification. Half a treasury moving from pledged to free is not a headline event; it is a change in the option set.
Unencumbered bitcoin can be sold into strength. It can be re-pledged at better terms in a tighter market. It can collateralize a completely different financing — an equipment facility, a data-center build, an acquisition. It can simply sit as a reserve against an operating loss.
I learned the value of that optionality the hard way. In 2022, the Terra collapse and the FTX bankruptcy erased $1.2 million of my portfolio. The mechanism was not a bad trade. The mechanism was counterparty failure — capital I believed was mine sitting on balance sheets that were not solvent. I liquidated every leveraged position in March of that year, preserved 60% of what remained, and moved the rest to self-custody.
The specific lesson: custody is not a legal status, it is an operational fact. You either control the keys or you control a claim on someone who does. Riot's 5,821 coins were, until September, a claim.
Liquidity vanishes. Lessons remain.
The DeFi comparison, and why it is not equivalent
There is a temptation to frame this as "centralized lending versus DeFi lending" and score it. That framing is lazy.
The on-chain analogue to Riot's facility would be a collateralized debt position on Aave or MakerDAO. Same asset, same directional exposure, same leverage mechanic.
What differs is where the uncertainty sits.
On-chain, the liquidation threshold is public and static. Every participant knows the exact price at which positions get seized. The failure mode is mechanical: a fast candle, a congested block space, a liquidation cascade, and a holder is wiped in a single block with no negotiation.
Off-chain, the trigger is contractual and private. The borrower and lender can renegotiate. Nobody gets liquidated by a bot at 3 a.m. But the borrower also cannot verify the lender's solvency, and the lender can change the terms of the relationship in ways that never appear in a transaction log.
I watched the on-chain borrow rate for stablecoins against BTC collateral swing through a multiple of its own level inside a single quarter while the underlying credit risk did not change at all. The rate was moving because utilization moved, not because anyone's probability of default moved. That is a utilization curve wearing the costume of a market price.
This is exactly why the fixed 6.15% is the more informative number here. Someone looked at Riot's balance sheet, Riot's collateral, and Riot's cash flow, and wrote down a rate. That is a judgment. An algorithm cannot make it.
A rate can be wrong. A judgment can be wrong. But a judgment can also be renegotiated, and a curve cannot.
The equity side: where the volume tells you the truth
On the equity, this is a modest positive against a neutral tape. Debt eliminated, collateral freed, no penalty paid. Expectations were probably 60% to 80% aligned with this outcome already — deleveraging at listed miners has been telegraphed for quarters.
So the announcement itself is not the trade. The reaction volume is.
I have a rule that has survived a lot of damage: when price diverges from volume, exit. It came out of the NFT cycle, where I flipped roughly fifty blue-chip assets for a 300% aggregate return and then got trapped holding illiquid positions after the turn because I refused to diversify out of an ecosystem I believed in. Community conviction is a leading indicator. Volume is the sustainment mechanism, and when they separate, volume wins.
Applied here: if RIOT rallies on the disclosure with above-average volume, the market is repricing the balance sheet structurally — lower leverage risk, better access to institutional capital, a cleaner story for risk-averse allocators who will not touch a miner with pledged treasury. That is a demand-driven move.
If RIOT ticks up on thin volume and fades, it is supply finding a bid, nothing more. The deleveraging was already in the price and the announcement is a liquidity event for sellers.
Watch the volume. The price will lie to you for a week. Volume will not.
The part the repayment does not fix
Here is where the celebration should stop.
Repaying a loan does not change miner economics. Zero.
The post-halving subsidy is 3.125 BTC per block. Hashprice compresses as total network hashrate climbs. Revenue per unit of energy is a function of BTC price, difficulty, and fleet efficiency measured in joules per terahash. All-in cost per mined coin is a function of electricity tariffs, curtailment credits, and machine depreciation.
None of those variables moved because Riot paid off Coinbase.
A miner's structural exposure is a levered long position on two inputs it does not control — bitcoin price and network difficulty — financed by an operating cost line denominated in dollars and megawatt-hours. Deleveraging the treasury makes the balance sheet more resilient through a drawdown. It does not make the mining operation more profitable at any given bitcoin price.
The live strategic question is the AI/HPC pivot, and it is a real one. Revenue per megawatt for high-performance computing hosting has been quoted at multiples of mining revenue per megawatt. The catch is capital intensity: converting mining sites into Tier III-grade compute facilities requires power infrastructure upgrades, cooling redesign, and customer contracts that take quarters to sign.
Freed-up bitcoin is precisely the ammunition for that capex. But ammunition is not a contract. Until a signed colocation agreement or a named counterparty appears in a filing, the pivot is a hypothesis, not a thesis, and it should be weighted accordingly.
The contrarian read: three things the consensus is getting wrong
The consensus interpretation is already forming. Riot deleveraged, therefore management is bearish, therefore caution. That reading is structurally confused.
Closing a leveraged position is not the same as abandoning the asset. A treasury desk can be maximally constructive on bitcoin over three years and still refuse to pay 6.15% to express it today. When the marginal cost of financing exceeds the risk-adjusted expected return on the financed asset, the mathematically correct action is to close the position — regardless of your price target. Being long and being levered are separate decisions, and conflating them is the most expensive habit in this market.
The "5,821 BTC of sell pressure" narrative is lazy. Collateral release changes legal control. It does not change intent. Every person repeating that line is inferring a seller from a contract termination. If those coins move, they move on-chain and you will see it in block explorer data before you see it in a headline. Monitor addresses. Ignore press releases.
The rate is a signal about the lender, not the borrower. A fixed 6.15% secured facility to a high-beta mining company is not a pure credit trade. That is a relationship trade — a customer-acquisition instrument underwritten at a spread that may not reflect the true risk-adjusted cost of the capital. Which means the facility Riot just retired might not be available at the same price in a tighter market, and every miner that built treasury strategy around assumed access to cheap bitcoin-backed credit should re-underwrite that assumption today.
Retail sees a repaid loan. The desk sees a subsidized rate that may not come back.
The data inconsistency you should not inherit
One more pass on the numbers, because this is where most readers will get misled.
The implied mark of $58,530 per bitcoin is inconsistent with the origination window and inconsistent with a treasury of roughly 11,414 coins being half-encumbered at those collateral values. If the mark is stale, the real LTV at repayment was materially lower than 58.7% because bitcoin appreciated after the valuation date — meaning Riot had more headroom than the ratio suggests, and the margin-call scenario was less acute than my stress table implies.
If the mark is current, the ratio is right and the date is wrong.
Either way, you cannot inherit a ratio from a secondary summary. Pull the Form 8-K and the corresponding 10-Q. Reconcile the collateral schedule against the debt schedule. Rebuild the LTV from raw integers. Numbers do not lie. They just get ignored.
I spent five years building models on top of unverified secondary figures before I understood that the input error dominates the output error every single time. If your LTV is wrong by ten points, your margin-call price is wrong by a double-digit percentage, and your position sizing is fiction.
What to actually watch
Three observable signals, ranked.
RIOT equity volume. If the disclosure produces above-average volume with a sustained bid, the market is repricing balance-sheet quality. If volume is thin, the move is noise. Either way, volume resolution is the trigger, not the headline.
On-chain movement from Riot-linked addresses. Any large transfer to a known exchange deposit cluster is the only real evidence of sell intent. Everything else is speculation about someone else's intentions.
The next miner credit print. The most informative data point in this entire story is the coupon on the next comparable bitcoin-backed facility written to a listed miner. If the rate comes in above 6.15%, Riot's deal was a relationship price and the market has repriced crypto credit. If it comes in flat or lower, the facility was market and Riot simply chose to retire it.
That single number will tell you more about the state of institutional crypto lending than any quarterly earnings call.
The bear market does not care about your narrative. It cares about whether your assets are actually yours, whether your carry is positive, and whether your counterparty is solvent when you need them to be.
Riot answered two of those three questions on September 21.