PowerCompute Mined 23.9 BTC and Earned Less Money. Now It's Spending $1M on 600 Miners
PowerCompute produced 35.8% more Bitcoin year over year and still booked lower revenue. Its answer is a $1 million cash order for 600 Antminer S21++ units it says will add 10% hashrate without new power capacity. The mining margin looks better. The operating loss didn't go anywhere.
PowerCompute mined 23.9 Bitcoin in the quarter ended September 30. That's 35.8% more than the same quarter a year earlier. And it brought in roughly $300,000 less.
Revenue came in near $1.9 million, down from $2.2 million. The coins were there. The dollars weren't. That's what happens when a higher coin count meets a lower average Bitcoin price, and it makes a mining income statement read like it belongs to two different companies.
The company published results on November 13. Attached to them was an equipment order: 600 Antminer S21++ units, about $1 million, paid in cash, deployed by mid-December at the Mississippi site. Management expects the swap to add roughly 80 to 85 petahashes per second of active hashrate without one extra megawatt of capacity.
Mining margin improved to 50.7% from 29% the quarter before. That's a genuine jump and it deserves credit.
But I'd slow down before reading too much into it.
Three Numbers That Don't Agree
Mining margin covers a narrow slice of operating cost. It doesn't pay salaries. It doesn't pay professional fees, the audit bill, or the depreciation sitting on the balance sheet. So a miner can get measurably better at running machines and still lose money as a company.
That's exactly what the quarter shows. Operating loss came in around $0.9 million. Net income was positive, near $0.4 million, and the gap between those two lines is worth understanding before anyone gets excited. The release also carried a going-concern disclosure, meaning management itself flagged substantial doubt about the company's ability to keep operating.
Those two facts sit in the same document. A 50.7% mining margin and a going-concern paragraph. Does a margin figure mean much when the company running it's telling you it might not be running in a year?
Here's my read. Mining margin is the metric you publish when the metric above it's ugly. It's not a lie. It's a frame. And frames are chosen.
Why Replacement Beats Greenfield, Mostly
The order itself is the more interesting decision. Building a new site means land, permitting, transformers, and a grid interconnection queue that in parts of the US can run longer than the useful life of the machines you're waiting to plug in. Swapping miners inside an existing electrical envelope skips all of that.
The specification is as follows. 600 units. Roughly $1 million in cash. Mid-December deployment. About 10% more active hashrate on the same power draw. Management's estimate, dependent on delivery and installation going to plan.
PowerCompute ran a version of this play before. Its September fleet-refresh announcement described an earlier batch of replacements running around 21.5 joules per terahash against roughly 34 for the retired equipment. Those numbers belong to that earlier program, not the newly ordered S21++ units. Still, they explain the logic: the same rack space and the same amps can produce meaningfully more hashing if the silicon is newer.
This is why machine counts mislead. Two fleets of identical size can draw wildly different power and deliver wildly different work. The number that matters combines electricity price, real joules per terahash, uptime, and the network's total difficulty. Nothing else.
And the cash element matters. A $1 million wire is money that can't cover payroll next quarter or a transformer failure in February. The investment has to clear a real bar, not a spreadsheet bar.
My take: the replacement route is the right call. It's also riskier than it reads. Installation means downtime. Nameplate efficiency means little if the machines run hot in a Mississippi summer. The 80 to 85 PH/s figure is a target, not a measurement, and targets have a habit of arriving late.
There's a second angle here that gets ignored. Curtailment. Sometimes the best-paying job for a miner in a given hour isn't mining. Shutting down and selling the power back can beat running every rig flat out, and a fleet refresh changes that calculus in both directions. Newer machines earn more per megawatt-hour mined, which raises the opportunity cost of curtailing. That's a trade-off, not a free win.
The Checkpoints That Actually Matter
Forget the annual production comparison. It's the least informative number in the release because it says nothing about cost or price realized.
Watch three things instead. First, delivered machines on site by mid-December, or a credible explanation for why they aren't. Second, sustained hashrate thirty and sixty days after installation, not the day-one figure. Third, cash on hand at the next reporting date after the $1 million leaves the account.
This change affects any forward model built on the September fleet efficiency numbers. Analysts who carried 21.5 joules per terahash into 2026 assumptions need to recheck which machines that figure describes. Backward compatibility is maintained except where noted below: it isn't.
Compare the strategy to peers routing capital into AI hosting and high-performance compute contracts. That path trades Bitcoin price exposure for contracted revenue. PowerCompute is doing the opposite. It's doubling down on Bitcoin output from sites it already controls, which means its results stay welded to hashprice and the network difficulty curve.
That's a defensible bet for a small operator with existing infrastructure. It's a terrible bet if the balance sheet can't absorb a bad quarter while the new units ramp.
So here's the concrete takeaway. The upgrade is cheap, incremental, and intellectually honest. The going-concern language is the part that decides whether it matters. A miner that produces 10% more hashrate and runs out of cash doing it hasn't improved anything. It's just mined a slightly larger hole.
23.9 Bitcoin is a good quarter for a company this size. But production isn't revenue, revenue isn't margin, and margin isn't solvency. PowerCompute just showed all four numbers in one document, and only three of them look healthy.
Explore More
Key Terms Explained
The first cryptocurrency, created in 2009 by the pseudonymous Satoshi Nakamoto.
Borrowed money used to increase trading position size.
Using computational power to validate transactions and create new blocks on proof-of-work blockchains.
In the context of restaking and EigenLayer, an operator is an entity that runs infrastructure to validate AVSs (Actively Validated Services).