BitFuFu's (FUFU) cloud-mining customers spent a quarter of what they spent a year ago. That collapse in repeat business, rather than the fall in the Bitcoin price, is what drove the company to a 43% revenue miss on 17 Aug and knocked a fifth off its share price before the US market had opened.
Revenue was $43mn against the $75mn analysts expected and $115mn a year earlier. Shares fell from Friday's close of $1.43 to $1.15 in pre-market trading in the hour to 12:00UTC, a drop of 20% and their weakest level in two months. They stood at $1.28 by 14:00UTC, half an hour after the US market opened, down 10% on the day, according to TradingView.
The Singapore-based company sells cloud mining, where customers buy a share of computing power BitFuFu operates for them and keep the Bitcoin ($BTC) it produces, rather than buying and running machines themselves. That business was 82% of revenue a year ago. It brought in $24.9mn this quarter, down 74% year to year and 57% from the first three months of 2026.
CFO Calla Zhao told the earnings call that the net dollar retention rate, which measures what existing customers spend against a prior period, came in at 24%. She said the fall reflected customers cutting order volumes rather than leaving, driven by lower Bitcoin prices and weaker mining economics.
Revenue below the cost of producing it
Cost of revenue was $44mn against revenue of $43mn. BitFuFu sold mining services for less than it cost to deliver them, before any overhead. Zhao attributed the pressure to high-cost computing power the company bought in late 2025, which she said would weigh on margins through the first half.
Mining equipment sales, worth $5.2mn a year ago, were nil. Hosting, where customers own machines and BitFuFu runs them, grew to $3.9mn from $1.1mn. Self-mining held roughly flat at $14mn from $15mn, which took a 47% increase in computing power allocated to the company's own account to offset a 28% fall in the average Bitcoin price to $71,600 from $98,800.
What the reported loss conceals
Net loss was $21mn against net income of $47mn a year earlier. Adjusted EBITDA, BitFuFu's preferred measure of operating performance before financing and accounting charges, was negative $18mn against positive $61mn.
Both quarters were dominated by fair value movements, the accounting requirement to mark Bitcoin holdings to market whether or not any coin is sold. Zhao said that excluding a $17mn fair value loss, adjusted EBITDA would have been about negative $1.5mn.
On the same basis, the first quarter of 2026 produced roughly positive $1.2mn and the second quarter of 2025 about positive $17mn. So the reported loss narrowed 41% from the first quarter, while the underlying operating result crossed from profit into loss.
Capacity halved, and the reason matters
Computing power under management fell 58% to 15.3 exahashes per second from 36.2 a year earlier. Power capacity fell to 273 megawatts from 728, a 62% cut. Customers mined 255 $BTC through the platform against 917.
Leo Lu, chairman and CEO, described this as "not a simple contraction" but a deliberate effort to improve the quality and economics of capacity, saying the company would "never sacrifice unit economics simply to maintain headline hash rate" and had walked away from third-party contracts whose economics no longer worked. He said monthly self-mined production nearly tripled during the quarter, from 32 $BTC to a peak of 90, and that capacity would return to about 20 EH/s by mid-August.
Both things can be true at once, and the release gives no way to weigh them. Retention at 24% suggests demand was doing much of the work.
Selling coin, buying stock
BitFuFu held 1,671 $BTC at 30 Jun, down 6.8% year to year. Zhao said operations were funded through operating cash flow, Bitcoin sales and a credit facility, and that the company repaid $10mn of Bitcoin-backed loans, ending the quarter with $5.4mn outstanding and 54 $BTC pledged as collateral, down from 357 $BTC in March.
Its own July operational update, filed 7 Aug, put holdings at 1,314. That is 357 $BTC gone in a single month, a 21% reduction and larger than the full-year decline the results disclose, in a month when the company mined 112. Both figures are BitFuFu's; the results release mentions only the smaller one.
The board has also authorized a buyback of up to $5mn of Class A shares, which Zhao presented as confidence in long-term prospects. The company is therefore selling Bitcoin to fund operations while preparing to buy back stock.
The bet on staying in mining
Across the sector, miners spent this earnings season moving capital into artificial intelligence. Keel Infrastructure shut its US mining operations outright, while MARA, CleanSpark, TeraWulf and Hut 8 pressed>Bitdeer, also Singapore-based, nearly tripled self-mining revenue and made a gross loss doing it.
Lu argued the opposite case with some specificity. As US-listed miners convert power and infrastructure to AI data centres under long-term contracts, he said, that capacity will not return to the Bitcoin network. He put global network computing power roughly 20% below its peak by late June and 14% below this year's high, leaving a less crowded network for those who stay. BitFuFu bought about 3,200 S21 XP machines in the quarter, reported fleet efficiency of 17.8 to 18.1 joules per terahash, and said curtailment arrangements in Oklahoma cut power costs to about $0.03 per kilowatt-hour in June.
That is a coherent thesis rather than inertia. It also leaves the company more exposed to the Bitcoin price than peers building a second revenue line, with a customer base spending a quarter of what it did.
BitFuFu took no analyst questions on the call, directing them to its investor relations address instead.