Canaan Q2 2026: 63.5% of the Bitcoin Was Already Spoken For Before the Buyback Started

Canaan had 1,915.5 Bitcoin at June 30. Only 698.5 of them were sitting in the bucket the company calls cryptocurrency assets.

The rest was already working for somebody else.

On Sept. 8, the Bitcoin mining-equipment maker reported second-quarter revenue of $31.9 million, below the $35 million-to-$45 million range it published in May. Product revenue collapsed to $13.6 million from $42.9 million in the first quarter — a 68% sequential drawdown. Third-quarter guidance came in at $11 million to $15 million, and that number deserves more scrutiny than the miss itself.

And 1,117 BTC were pledged against secured term loans. Another 100 BTC had been moved into a fixed-term product. Those 1,217 coins — 63.5% of the headline balance — were booked as cryptocurrency receivables worth $70.9 million, on a separate line from the $47 million of cryptocurrency assets.

The number most people quote is not a pool of money. It is a schedule of claims, and 63.5% of it is already claimed.

The Subsidy Ended. Nobody Sent a Memo.

ASIC manufacturers are the highest-beta instrument in the crypto capital stack. They are levered to hashprice, to miner capex, and to the willingness of strangers to finance machines against future block rewards. When any one of those three legs moves, the revenue line does not drift. It detonates. A 68% quarter-over-quarter collapse in product revenue is not a demand wobble. It is the sound of a replacement cycle ending.

Here is the part the notes keep skipping. The 2023-to-2025 ASIC replacement cycle was not funded by block rewards. It was funded by inscriptions. Ordinals gave Bitcoin block space a second buyer and handed miners a fee stream that had nothing to do with settlement demand. That fee stream flowed into miner margins, and miner margins flowed into new machine orders. Structurally, it was a liquidity mining program running on top of Bitcoin's fee market — a subsidy that produced TVL-shaped hashrate numbers and evaporated the moment the incentive stopped paying. Without that inscription wave, Bitcoin's fee market would already be in a harder conversation about what secures the chain after the next halving, and Canaan's order book would have thinned two years earlier.

We didn't get here because ASIC demand disappeared. We got here because it was rented.

I have watched this exact shape three times. In 2021 I built a Resonance Index for Bored Ape holders that ignored floor price and measured social capital instead — celebrity ownership as a proxy for network effect — and it peaked weeks before the floor did. In 2022 I spent three months inside the Terra mechanism and published a long autopsy arguing the failure was never a flaw in the peg, it was a flaw in the assumption that growth is infinite. Both times the tell was identical: a metric that looked like demand but was actually a subsidy.

Canaan's income statement now has three revenue lines and one independent variable. Hardware sales, self-mining, and treasury — all three are the same trade expressed three ways. Self-mining produced 243 BTC and $17.7 million in the quarter. That works out to roughly $72,800 of realized revenue per coin, which is a perfectly respectable number and a completely correlated one: it is paid in the same asset that collateralizes the company's borrowings and marks its receivables.

Zero diversification. Maximum reflexivity. When one asset is your product, your yield, and your collateral, you have not built a business. You have built a mirror.

The treasury was supposed to be the answer. In the first quarter, Canaan was still selling the reserve story — a record BTC and ETH position approaching $148 million, presented as a strategic asset base. The pitch worked, and it worked on people I know. I spent 2025 advising three Swiss banks on how to frame this exact structure, and I told them then what I will repeat now: institutional adoption requires narrative dilution, and narrative dilution means the hardware wrapper gets stripped off the moment the treasury needs a credit rating rather than a chart.

Two quarters later the reserve is a receivables line and the strategic part is the loan. The company did not change its strategy. The market changed the price of the strategy, which is the same lesson with a new timestamp.

Understand what the secured term loans actually are. Somewhere between a margin loan and project finance, a lender takes Bitcoin as collateral, applies a loan-to-value ratio, and wires dollars. The borrower keeps the upside, the lender keeps the right to sell at a threshold, and both parties agree not to describe it as a forced-sale option until the day it becomes one. Canaan disclosed $70.9 million of cryptocurrency receivables against 1,217 encumbered coins, which approximates the value of the pledged and time-locked assets rather than the principal of the debt. The debt itself is not itemized in the release. That omission is not accidental. It is how these structures are always marketed.

The wrapper is gone. What remains is a credit story with a factory attached, and the credit story starts with where the collateral actually sits.

A Balance Sheet That Answers a Different Question

Most coverage will tell you Canaan holds 1,915.5 BTC. Technically true. Practically useless.

Strip it apart. At June 30: 1,117 BTC pledged for secured term loans. 100 BTC transferred into a fixed-term product. 698.5 BTC in cryptocurrency assets. The three buckets sum cleanly to the headline total, and that is exactly what makes the disclosure so easy to misread — the arithmetic reconciles, so nobody asks whether the categories are equivalent. They are not. One bucket is encumbered, one bucket is time-locked, and one bucket is spendable.

This is the same forensic habit I picked up auditing Golem's pre-sale distribution contracts in 2017. I spent a full day building the token distribution logic in pseudocode, found three flaws that could have inflated supply, and filed a GitHub issue that forced a protocol pause. The lesson was never about Golem. The lesson was that a number and its name are two different things, and the name is where the deception lives.

So write the function the way an auditor would:

function spendable_treasury(report):
    gross = report.crypto_assets + report.crypto_receivables   # 1915.5 BTC
    encumbered = report.btc_pledged_as_collateral              # 1117 BTC
    restricted = report.btc_in_fixed_term_product              # 100 BTC
    free = gross - encumbered - restricted                     # 698.5 BTC
    return free, encumbered + restricted

The function returns 698.5 free and 1,217 restricted. 36.5% of the treasury is liquid. The other 63.5% is collateral, and collateral has a counterparty.

Now follow the marks, because the marks disagree with each other. The $70.9 million of cryptocurrency receivables covers 1,217 encumbered coins. That implies roughly $58,300 per Bitcoin. The $47 million of cryptocurrency assets covers 698.5 coins. That implies roughly $67,300 per Bitcoin. A gap of approximately 13 percent between two buckets of the same asset, on the same balance sheet, dated the same day.

If the cryptocurrency assets bucket contains any ETH — and it does, because Canaan sold 3,952 ETH in late August, which means it was carrying a real position — then the per-Bitcoin mark on the free bucket is higher still and the gap is wider.

There are only a handful of ways to produce a 13% spread on identical assets. Option one: the encumbered coins carry a collateral haircut, which is what a lender demands when it wants a cushion. Option two: the free bucket is fair-valued to spot while the receivable bucket sits closer to cost. Option three: the receivables line is net of something the company has chosen not to itemize.

I do not know which one it is. Neither does anyone reading the press release. But every one of those explanations says the same thing from the equity holder's seat: the encumbered Bitcoin is worth less per coin to shareholders than the free Bitcoin, and the financial statements have already started admitting it. The bug wasn't the collateral. It was the classification.

The Cash Pile Grew While the Business Burned

Here is where the quarter gets genuinely interesting.

Canaan reported a $97.6 million net loss. It also reported cash of $66 million at June 30, up from $43.5 million at March 31 — a $22.5 million increase in the middle of a quarter in which it lost nearly a hundred million dollars.

Read those two lines together and you learn more than either one alone. The loss included real noncash charges: $25.3 million in inventory and prepayment write-downs plus purchase-commitment provisions, and $9.2 million in property and equipment impairment. That is $34.5 million of the loss that never touched a bank account. The remainder — call it $63 million — is the ceiling on what operations could have consumed, and the true figure is lower once you net out the mining cash contribution management flagged as positive before depreciation.

Cash still went up. The quarter was financed, not earned. The financing showed up as secured term loans against pledged Bitcoin, which is why 1,117 coins moved into the collateral bucket in the first place.

This is the mechanism I spent two weeks modeling on Uniswap V2 in 2020, and it is worth stating plainly because people keep getting it backwards. Permissionless liquidity is not a balance. It is a permission — a right to exit that exists only as long as the counterparty's incentives hold. Code is law, but liquidity is truth. A collateralized loan is the same instrument in a different costume: the Bitcoin is yours until the margin call, and the margin call is the real owner.

Scale check: at the end of 2025, Canaan held $80.8 million in cash. Six months and roughly a year of losses later, it holds $66 million. The company did not fund the loss out of cash flow. It funded the loss out of its own collateral and reported the result as a stronger liquidity position.

And the inventory line is sending a message the cash line cannot. A $25.3 million write-down plus purchase-commitment provisions against $13.6 million of product revenue means the provisions are 1.9 times the revenue they are meant to support. You book a purchase-commitment provision when you already know you will take delivery of silicon you cannot sell at a profit. That is not a lagging indicator of a bad quarter. It is a scheduled entry for the next two.

Now put a floor under the mining line. Canaan produced 243 BTC in the quarter at an implied realized rate near $72,800 per coin — before power, before hosting, before the Ethiopia site went quiet. With roughly 35% of July operating hashrate tied to a paused Ethiopian facility, self-mining is not a stable base. It is a spread that widened in 2024 and has been narrowing since. Hashprice does not negotiate. It is block subsidy divided by network hashrate, and when the subsidy halves and hashrate keeps climbing, the only variable left standing is how much cheaper your electricity is than the next operator's. That is a competition Canaan can win in Ethiopia and lose the day the power contract is renegotiated.

Take the segment split seriously. $13.6 million of product plus $17.7 million of mining lands almost exactly on the $31.9 million total. So when management guides third-quarter revenue to $11 million to $15 million, the midpoint of $13 million is a figure mining alone could plausibly occupy — against a business that just wrote down $25.3 million of inventory and commitments in a single quarter.

Read literally, the Q3 guide implies equipment revenue compressing toward the noise floor. Nobody is pricing that.

The Buyback Is the Tell

After quarter-end, Canaan sold 3,952 ETH and 54 BTC in late August for approximately $13.9 million, and used part of the proceeds for share repurchases.

By Sept. 8, the program had retired about 16.4 million American depositary shares for $7.4 million, including $5.4 million spent in late August. Average cost per ADS: roughly $0.45.

Four ways to read that number, and only one of them is bullish.

First, the timing. $5.4 million of crypto was liquidated in late August and $5.4 million of equity was repurchased in the same window. That is 39% of the crypto sale proceeds routed into share retirement rather than operations, at a moment when the company was guiding revenue down by more than half sequentially. Cash is a war chest in a bear market. Canaan spent nearly two-fifths of a liquidation event on its own stock.

Second, the scale. The entire repurchase program — $7.4 million — equals 7.6% of the $97.6 million the company lost in one quarter. This is not capital return. This is a press release with a wire transfer attached.

Third, the direction of the trade. MicroStrategy sells equity to buy Bitcoin. Canaan sells Bitcoin to buy equity. Those are mirror images, and they are not morally equivalent. One expresses a view that hard money is cheap relative to your paper. The other expresses a view that your paper is cheap relative to hard money — or, more likely inside a collateralized structure, that your paper is the only asset you can still defend. A treasury company that becomes a net seller of its treasury asset has stopped being a treasury company and started being a deleveraging story.

There is also a mechanical reason buybacks in this sector deserve suspicion. Retiring ADS at roughly $0.45 shrinks the denominator, which flatters per-share metrics, which supports the narrative, which supports the equity — and the equity is precisely the instrument the company may eventually need to issue to raise real money. A buyback funded by liquidating the treasury is not returning capital. It is re-levering the story. The hype here is that a repurchase is a signal. The liquidity here is that the repurchase was paid for with the last genuinely mobile asset on the balance sheet.

Fourth, the revealed preference. Management bought its own ADS at roughly $0.45 while pledging 1,117 BTC to lenders at an implied mark near $58,300 a coin. Both are votes. Only one of them is a vote with a counterparty attached.

The Blind Spot Nobody Is Pricing

The consensus read on this quarter will be that a miner-turned-treasury-vehicle monetized crypto to return capital. That framing is comfortable because it rhymes with every treasury narrative since 2020, and comfort is what the market buys in the ninth month of a bear market.

The unglamorous read is that 1,117 BTC is now a scheduled seller with no discretion. If Bitcoin marks down through the loan covenants, the collateral does not get to wait for a better price. It gets sold by whoever holds the other side. In notional terms it is a rounding error against daily spot volume — a bit over $70 million at the marks in the disclosure. In market-structure terms it is exactly the kind of flow that arrives at the worst possible hour, from a seller who is not expressing an opinion.

I have run this exercise before. In 2022 the entire Terra complex looked like a collateral system right up until the moment the collateral became the exit. The reflexive part was never the algorithm. It was the assumption that the thing you hold can always be converted at the price you marked it.

There is a second blind spot, and it sits upstream of Canaan entirely. Blob space is a subsidy too. Post-Dencun data availability made rollup fees artificially cheap, and the standard objection is that cheap DA is permanent. It is not. Blob demand is on a trajectory toward saturation, and when it gets there, rollup gas economics revert to something resembling cost. Every subsidy in this industry shares the same expiry profile — inscription fees on Bitcoin, blob space on Ethereum, liquidity mining APY on every DEX that ever printed a three-digit number. Liquidity mining APY was never yield. It was the project paying you to be a row in a dashboard. The ASIC replacement cycle was the same trade, settled in silicon.

Which brings the analysis back to the metric Canaan chose to highlight: mining operations that made a positive cash contribution before depreciation. That is an EBITDA-flavored number, and it belongs to the same species as a farm APY. It measures the subsidy, not the business. Depreciation on an ASIC fleet is not a footnote in a bear market. It is the entire cost structure, arriving on a schedule, in the exact period when hashprice will not cover it.

The sector-wide implication is the part worth internalizing. Canaan is not a special case; it is the most legible one. Every ASIC maker that pivoted to self-mining, every miner that issued convertible notes against its fleet, every treasury vehicle that pledged coins to fund a repurchase is running a version of this structure. The difference between a healthy one and a broken one is not the size of the Bitcoin stack. It is the share of that stack answering to a lender, and how many quarters of burn stand between the borrower and the covenant. Most of these companies report the first number and not the second. That asymmetry is the entire trade.

What to Watch, and What It Means If You Don't

Forget the 1,915.5 figure. It is a marketing number now.

Watch three lines instead. The ADS count, because every incremental repurchase funded by crypto sales is a transfer from a hard asset to a soft one, and the pace tells you how much management still believes in its own wrapper. The pledged-BTC line, because a rising collateral balance means the company is financing itself against an asset it cannot sell, and a falling one means the lender already took it. And the purchase-commitment provision, because silicon commitments do not disappear. They get delivered.

The optimistic case is unromantic and worth stating: Canaan converted $13.9 million of crypto into cash and used part of it to retire 16.4 million ADS. The sale worked. The mechanism functioned. In a bear market, a company that can still monetize its treasury is a company with options, and options are the only thing that matters when the revenue guide drops from $31.9 million to $13 million in ninety days.

The pessimistic case is that we are watching a balance sheet get converted, one line at a time, into the only number that will still be true in eighteen months — the one that is not encumbered, not time-locked, and not marked at a price somebody else controls.

Sixty-three point five percent of the treasury is already spoken for. Which raises the only question that matters: when the loans come due, what exactly is left to sell?