The headline number was 1,915.5 BTC. The number that matters is 1,217.
On Sept. 8, Canaan published a set of figures that most of the tape filed under "treasury update." I filed it under collateral disclosure. Of the Bitcoin miner and Avalon ASIC manufacturer's 1,915.5 BTC held at June 30, 1,117 BTC were pledged against secured term loans and 100 BTC had been transferred into a fixed-term product. That is 1,217 BTC — roughly 63.5% of the headline balance — that could not be sold at will, spent on operations, or used to defend anything. It was collateral, and collateral has a counterparty.
The revenue line framing the release was bad enough. Second-quarter 2026 revenue came in at $31.9 million, well below the $35 million to $45 million range management guided in May. Third-quarter guidance landed at $11 million to $15 million. That is a 53% to 65% sequential collapse stacked on top of an already broken quarter.
Silence in the ledger speaks louder than hype.
Why this quarter is different from every previous Canaan drawdown
Canaan has always been a cyclical business wearing a growth story. Avalon was one of the first commercially viable SHA-256 ASIC lines, and the company spent most of the last decade selling machines into whatever part of the halving cycle happened to be paying for hashrate at the time. The 2019 Nasdaq listing was never really about the miners. It was about giving public-market capital a way to express a view on Bitcoin's compute layer without holding Bitcoin directly.
What changed is that Canaan stopped being purely a vendor. Over the last two cycles it became a self-miner, then a treasury holder, then — quietly — a leveraged treasury holder. That progression is the entire story of the Sept. 8 disclosure. In the first quarter of 2026, the company's combined BTC and ETH treasury approached $148 million. That figure was the marketing. The second quarter is the accounting.
The distinction matters because the market is still pricing Canaan on the treasury narrative while the treasury itself has been partially converted into loan collateral and a buyback funding source. In a bull market, that gap gets ignored. In a bull market, the reflex is to take the Bitcoin balance, multiply it by spot, add the mining segment to the multiple, and call the equity cheap. That reflex is wrong here, and the arithmetic in this release shows exactly why.
There is also a regulatory layer. Under FASB's fair value rules for crypto assets — effective for fiscal years beginning after Dec. 15, 2024, and therefore fully in force across Canaan's 2026 reporting — digital assets are remeasured to fair value each reporting period, with changes running through net income. The stated intent is comparability. One asset, one date, one price. That is the standard's promise, and it is precisely the promise this balance sheet does not keep.
The same asset, carried at two prices, on the same day
Run the division.
The pledged and fixed-term Bitcoin was recorded as cryptocurrency receivables worth $70.9 million at June 30. That bucket is 1,217 BTC. Divide one by the other and the implied carrying price is roughly $58,258 per coin.
Separately, the 698.5 BTC sitting in the cryptocurrency assets category was valued at $47 million. That implies roughly $67,287 per coin.
Two buckets. One asset. One measurement date. A 15.5% spread between them.
The disclosure does not reconcile that gap. It offers no note explaining whether the fixed-term product is discounted for counterparty terms, whether the pledged coins are carried at something other than spot, whether the receivables line contains a non-Bitcoin component, or whether the loan agreement itself imposes a valuation convention. Any of those explanations would be legitimate. None of them is stated. What is stated is two prices for the same thing on the same page.
Then there is the third price — the one the company actually got.
In late August, Canaan sold 3,952 ETH and 54 BTC for approximately $13.9 million. Work backward from those totals and the implied execution is roughly $2,520 per ETH and about $72,800 per BTC, and the two lines sum to $13.8 million against a disclosed $13.9 million. The reconciliation is almost exact, which is what makes it useful. The quarter's mining revenue supports the same number: 243 BTC produced against $17.7 million of mining revenue implies an average realized price near $72,840.
So the realizable price of Bitcoin in Canaan's own hands was roughly $72,800. The unencumbered bucket was marked at $67,287. The encumbered bucket was marked at $58,258 — a 20% discount to a price the company demonstrably transacted at two months later.
Data does not negotiate; it only confirms.
I spent 72 hours in 2017 reverse-engineering Solidity on a token launch that most of the market had already priced at a premium, and the lesson that stuck was not about reentrancy. It was that the most expensive line item in a disclosure is the one the preparer chose not to explain. I have applied that filter to every balance sheet since, and it has never once been wrong about where the risk sits.
What "positive cash contribution" actually measures
Management was careful with language in this release, and the care is instructive.
Mining produced 243 BTC and generated $17.7 million in the quarter. Management said those operations made a positive cash contribution before depreciation. That is a true statement and a narrow one. It stops above the depreciation line, above the impairment line, above corporate overhead, above interest on the secured term loans, and above every noncash charge that turns a segment-level contribution into a company-level loss.
Run the segment split. Of $31.9 million in total revenue, $13.6 million was product revenue — down from $42.9 million in the first quarter, a 68% sequential decline that management attributed to less computing power sold and lower selling prices amid weaker demand. Mining at $17.7 million was therefore roughly 55% of the top line. Read that again: the mining segment, not the miner business, was the majority of Canaan's revenue in the second quarter. The company that lists itself as an equipment manufacturer is now primarily a self-miner with a hardware side project.
That inversion has a consequence for the third-quarter guide. If total revenue lands at $11 million to $15 million, and product revenue is still falling, then the mining line has to carry a disproportionate share of what remains. Which brings us to Ethiopia.
Canaan disclosed that paused mining in Ethiopia accounted for nearly 35% of its July operating hashrate total. Thirty-five percent of operating hashrate is not a rounding error. It is the difference between a segment that clears its cash cost and one that does not. The second quarter's $17.7 million mining revenue was earned with that capacity in some state of availability. The third-quarter guide was issued against a July hashrate base that had already lost more than a third of its operating total.
The positive cash contribution claim is therefore a claim about a configuration of assets that no longer exists in the same form.
A $97.6 million loss that did not cost $97.6 million in cash
The reported net loss for the quarter was $97.6 million. That number will get quoted as a cash burn. It is not one.
The loss included $25.3 million in inventory and prepayment write-downs and purchase-commitment provisions, plus $9.2 million in property and equipment impairment. Those are noncash charges — inventory that was carried above its net realizable value, prepayments to suppliers against machines that are no longer expected to sell at the price that justified the deposit, and hardware whose carrying value exceeds its recoverable amount. Together they represent roughly 35% of the reported loss, and they represent a judgment about the future of the ASIC market rather than a check written in the second quarter.
That distinction matters because it tells you what management privately thinks about the equipment pipeline. A purchase-commitment provision is an admission that contracted supply is now worth less than the contracted price. In a bull market where mining economics look fine on paper, that is an unusual thing to concede. It is also consistent with everything else in the release: the compute market Canaan sells into is not clearing at the prices the industry is advertising.
Cash, meanwhile, rose to $66 million at June 30 from $43.5 million at March 31 — though still below the $80.8 million held at the end of 2025. That sequential increase did not come from operations. It came from the balance sheet, and the September disclosure confirms exactly how: asset sales after quarter-end.
The buyback arithmetic that does not reconcile itself
The August liquidation raised roughly $13.9 million. Canaan said part of the proceeds went to share repurchases.
By Sept. 8, repurchases under the existing program totaled about 16.4 million American depositary shares for $7.4 million, including $5.4 million spent in late August. So of the $13.9 million raised from selling ETH and BTC, $5.4 million went into the buyback in the same window. The remaining $8.5 million is not itemized in this disclosure. Working capital is the obvious destination. It is also the destination the disclosure declines to name.
The audit trail never lies, only the auditor can. And here is the arithmetic that deserves an auditor's attention: 16.4 million ADS for $7.4 million implies an average cost of roughly $0.45 per ADS. Readers should reconcile that against Canaan's ADS ratio and outstanding share count, because the disclosure as issued does not make the implied average self-evident against the company's historical trading range. Either a very large fraction of the program was executed during an extreme dislocation, or the ADS-ratio arithmetic needs a footnote that was not provided. Either way, the number as printed is striking, and striking numbers are where the reconciliation work starts.
Now the structural point, which is more important than the per-share price. Canaan sold a volatile, partially encumbered asset and used the proceeds to retire equity. In isolation, buying back stock below intrinsic value is rational. But the equity here is not being retired with surplus cash from a profitable business. It is being retired with the only genuinely liquid portion of a treasury that is more than half pledged against loans. The buyback shrinks the share count and simultaneously shrinks the unencumbered collateral buffer supporting the secured term loans the rest of the treasury is pledged against.
Also note what the 100 BTC fixed-term allocation represents, and apply the fourth law of crypto accounting: yield is not income; it is risk repackaged. A fixed-term product converts a self-custodied, immediately spendable asset into a counterparty obligation with a maturity and a credit exposure. It shows up on the balance sheet as a receivable. Receivables get discounted because receivables can fail.
The contrarian read: this is a leveraged Bitcoin proxy, not a Bitcoin proxy
The consensus framing of Canaan in a bull market is simple: it is a high-beta Bitcoin play with an equipment option attached. Buy the miner, get the machine business for free.
That framing is now operationally wrong, and not because the treasury shrank. It is wrong because the treasury was converted into a structure the equity holder does not control. Sixty-three and a half percent of the Bitcoin is pledged against secured term loans. The determining variable is no longer Bitcoin's price alone — it is Bitcoin's price against a loan-to-value threshold that the company has not disclosed, on a loan facility whose terms the company has not disclosed, secured by collateral whose carrying value the company reports at a 20% discount to the price it later realized in the open market.
In 2022, I activated my emergency protocol within four hours of the UST de-peg and published withdrawal thresholds and liquidation prices for lending-market exposure. The whole point of that exercise was that in a contraction, collateral does not sit still — it is either sold by the holder or seized by the counterparty, and the difference between those outcomes is decided by a covenant most holders have never read. Canaan in 2026 is the same structure at a different scale. The equity is a residual claim behind a pledge.
Speed without structure is just noise. And the noise here is bullish: 1,915 BTC, a mining segment, an ASIC franchise, a buyback. The structure is bearish: a collateralized treasury, a $11 million to $15 million revenue guide, an Ethiopia hashrate base that lost 35% of its operating total, a 68% sequential collapse in the actual product business, and a $97.6 million quarterly loss of which a third is management marking its own pipeline down.
The blind spot is not that Canaan is in trouble. The blind spot is that the market is modeling Canaan's downside using a treasury balance that is no longer freely available to absorb it.
What to watch from here
The third-quarter print is the obvious checkpoint, but it is not the most informative one. The most informative item will be any disclosure that reconciles the two carrying prices on the June 30 crypto balance, because that reconciliation will explain the terms of the secured term loans without the company having to state them directly. After that, watch whether a second crypto-to-cash conversion follows the August sale — a third liquidation in a rising market would tell you the funding pressure is operational, not opportunistic. Then watch Ethiopia. Hashrate that goes quiet for a quarter and returns is maintenance; hashrate that stays quiet is a stranded asset, and a stranded asset is the collateral problem arriving one layer earlier than anyone modeled.
The Bitcoin is real. The question is who can move it.