Contrary to the framing that circulated on Sept. 8, Canaan did not report "1,915.5 BTC on the balance sheet." It reported 1,915.5 BTC in total. Those are not the same claim.
Of that total, 1,117 BTC were pledged against secured term loans at June 30. Another 100 BTC sat in a fixed-term product. That leaves 698.5 BTC — 36.5% of the headline figure — in the bucket the company labels cryptocurrency assets. More than half of Canaan's Bitcoin was collateral, not reserves. The ledger doesn't round up.
The same release reported $66 million in cash, up from $43.5 million at March 31. It also reported a $97.6 million net loss for the quarter. Cash rising while the income statement bleeds is not a paradox. It is a description of how the cash arrived. The ledger doesn't care which bucket the press release puts the number in.
Canaan Inc. designs and sells Bitcoin mining ASICs and lists on Nasdaq through American depositary shares. It has survived three hardware cycles, and each one has taught the same lesson: ASIC margins compress faster than order books expand. Machines are a depreciating industrial good sold into a market where the buyer's payback period is set by hashrate difficulty and the cost of electricity — two variables the seller does not control.
Through the 2025 and 2026 bull market, the treasury-company template became a financing instrument. Hold coins, mark them at market, and let the equity trade as a leveraged proxy. For hardware makers, the template is seductive because it converts an inventory problem into a balance-sheet story. The narrative shifts from "we sell machines" to "we hold coins," and the second sentence is easier to market in a bull market than the first.
On May 2026, management guided second-quarter revenue to $35 million to $45 million. On Sept. 8, 2026, Canaan reported $31.9 million. Product revenue alone fell to $13.6 million from $42.9 million in the first quarter — a 68% sequential collapse. Management attributed the decline to less computing power sold and lower selling prices. Then it guided third-quarter revenue to $11 million to $15 million. At the midpoint, that is a further 59% sequential decline.
I read disclosures in a fixed order: units first, then categories, then cash, then management language last. That order is not stylistic. In 2017 I spent six weeks reverse-engineering the Paragon Coin reward-distribution contract and found an integer overflow that would have drained 12 million tokens during peak volatility. The lesson never left me. Code and ledgers do not have a tone of voice. They have line items.
The revenue mix has inverted, and almost nobody flagged it. Product revenue was $13.6 million. Mining revenue was $17.7 million. Together, $31.3 million of the $31.9 million total. The remainder is rounding and minor lines. For the first time in the company's reporting history, an equipment manufacturer earned more from operating its own machines than from selling them. That is a category change, not a mix shift. An ASIC vendor is priced on unit volumes, ASPs, and gross margin per machine. A miner is priced on realized hashprice, power contracts, and uptime. Those are different businesses with different multiples, and the second one is far more exposed to the difficulty curve.
The 243 BTC figure contains an implied price. Canaan produced 243 BTC and generated $17.7 million from mining in the quarter. Divide: roughly $72,800 per coin. That is the implied monetization price if every coin produced was converted. If Canaan instead recognized production at fair value while retaining coins, the implied figure says more about the measurement date than about execution. Both readings are defensible. Neither is disclosed in the release. I would not build a model on either without the cash flow statement.
The three-bucket ledger is where the accounting gets interesting. Canaan recorded the pledged and fixed-term Bitcoin — 1,217 coins in total — as cryptocurrency receivables worth $70.9 million. The remaining 698.5 coins sat in cryptocurrency assets at $47 million. Run the division. The encumbered coins carry an implied value near $58,300 each. The free coins carry an implied value near $67,300 each. That is a spread of roughly 13% inside the same asset class on the same measurement date.
I am not asserting manipulation. I am asserting that a 13% intra-asset discrepancy needs a footnote. Possible benign explanations: different impairment measurement dates, a fixed-term product measured on a discounted basis, or valuation inputs tied to the lending agreement rather than to a spot index. Possible non-benign explanations exist too, and I do not need them to justify the question. Correlation is not causation, and a valuation gap is not a fraud. It is an unexplained variance. Unexplained variances are where auditors earn their fees and where analysts lose their credibility by guessing.
The cash increase was financed, not earned. Cash moved from $43.5 million to $66 million — a gain of $22.5 million — in a quarter that produced a $97.6 million net loss. That is arithmetically possible only through a combination of noncash charges, asset liquidation, or borrowing. Canaan sold its crypto in late August, after the quarter closed, so those proceeds are not in the June 30 figure. The pledged coins were already serving as collateral. The most defensible inference is that the quarter's liquidity came substantially from secured lending against the treasury. The $66 million cash cushion and the 1,117 pledged coins are two views of the same transaction.
Not all of the loss was noncash, but more than you would expect. Management broke out $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 identified noncash charges against a $97.6 million net loss. Purchase-commitment provisions deserve their own paragraph. They mean Canaan signed supply contracts for machines it no longer expects to sell profitably. Those commitments were likely entered during the capacity scramble of the prior bull phase. They now sit on the balance sheet as a liability with a fixed clock and a floating value. Q3 product revenue guidance of $11 million to $15 million for the whole company tells you what management thinks of that clock.
In late August, the coins left the building. Canaan sold 3,952 ETH and 54 BTC for approximately $13.9 million and used part of the proceeds for share repurchases. The release does not disclose execution prices, but the aggregate is enough to back out a range. If BTC traded near $60,000, the 54 coins account for about $3.24 million, leaving roughly $10.66 million of ETH proceeds — an implied average near $2,697 per ETH. If BTC traded near $70,000, the implied ETH average drops to about $2,561. The August sale prices ETH somewhere in the $2,500s to $2,700s, and that number matters for anyone marking Canaan's remaining digital assets.
The buyback arithmetic is equally legible. By Sept. 8, repurchases totaled about 16.4 million ADS for $7.4 million, including $5.4 million spent in late August. That averages approximately $0.45 per ADS across the program, with the caveat that any change in the ADS ratio mid-program makes per-unit comparisons across periods unreliable. The sequencing is the point: coins were liquidated, then equity was retired. The buyback was funded by selling the collateral asset, not by operating cash flow.
That raises the only question that actually decides whether this was a good decision or a bad one: what is the equity worth relative to the assets? Build the bridge. Cash of $66 million, plus cryptocurrency assets of $47 million, plus cryptocurrency receivables of $70.9 million, equals $183.9 million of tangible, crypto-linked value. Subtract the outstanding secured term loan — secured by 1,117 BTC — and you have net asset value. The receivables are contractual claims, not spendable cash, so the spendable portion is narrower than the headline suggests. But if the market capitalization sat below that net figure, retiring equity at a discount is one of the few genuinely rational cases for liquidating a treasury. If the market capitalization sat above it, the company sold a scarce asset to buy an expensive one.
The missing number is the loan principal. Without it, the net asset bridge has a hole wide enough to drive the entire valuation through. It also determines the margin-call price. Suppose the lender runs a 50% loan-to-value covenant. At a $60,000 BTC price, 1,117 pledged coins represent about $67 million of collateral, supporting roughly $33.5 million of debt. At a $50,000 BTC price, the same collateral supports $28 million. If the actual principal is materially higher than those figures, the covenant is tighter than the market assumes and the next drawdown tests it. This is the same structure I modeled in 2020, when I built a Python framework to simulate liquidation cascades across Aave and Compound under a 30% flash crash. The output was unambiguous: the trigger price, not the average price, determines survival. That framework applies here without modification. A borrower with pledged collateral does not get to hold through a dip. It gets a phone call.
And there is an operational chokepoint. Canaan counted paused Ethiopia mining as nearly 35% of its July operating hashrate total. A single jurisdiction hosting a third of operational capacity is a structural failure mode, not an operational inconvenience. I have written before about systems marketed as distributed that quietly route through one node. A sequencer is one example. A national power contract is another. The failure mode is identical: when the chokepoint goes down, the redundancy turns out to have been a diagram.
Now the runway arithmetic. Third-quarter revenue guidance of $11 million to $15 million is a top line, not a cash flow, so I will not present it as one. But set it against $66 million of cash and $34.5 million of identified noncash charges in the prior quarter, and the shape is clear. If cash operating burn runs anywhere near $20 million per quarter, the runway is roughly three quarters before the next financing event. If the burn is lower because mining contributes cash before depreciation — management says it did — the runway extends. The most important forward-looking number in the next filing is not revenue. It is the secured loan balance and the covenant terms attached to it.
The bull-market consensus on treasury companies is a simple template: coins on the balance sheet make the equity a leveraged proxy, so buy the equity during drawdowns. That template breaks on encumbrance. A pledged coin and a free coin have the same ticker and completely different utility. The correlation between "BTC held" and "BTC exposure per share" is not a constant. It degrades with every lien attached to the treasury.
Here is the contrarian inversion, and I will take it seriously because it cuts against my own instinct. The revenue collapse and the crypto sales are correlated in time, and correlation is not causation. The August sale may have been a deliberate capital-allocation decision rather than a liquidity scramble — a company choosing to retire discounted equity instead of holding a volatile reserve. If that is the correct reading, the buyback was the most disciplined action in the entire release. And I cannot verify which reading is correct, because the loan principal is undisclosed. That is the honest position. Neither bullish nor bearish. Unrateable, pending a document.
Watch three lines in the next filing. The secured term loan principal and its loan-to-value covenant. The Q3 print against the $11 million to $15 million guide, with product revenue isolated from mining revenue. And any further disclosure of digital-asset impairment, which the current release does not break out. If product revenue stays near $13.6 million or below while mining supplies the majority of the top line, then Canaan is a mining company with a hardware division attached, and it should be valued on hashprice, power cost, and uptime. Not on a Bitcoin headline that has already been spent. The 10-Q is the document. Everything else is the narrative, and the narrative is the part of the ledger that has never once balanced.