OP-ED: Private Miners Have an Evidence Problem

This op-ed was written by Pierre Corbin, Founder of Flash, a financial reporting platform for bitcoin miners. The views expressed are his own and do not reflect those of TheEnergyMag.

Public miners report cost per bitcoin on different bases, yet lenders work with them. Private miners rarely get the same hearing. The difference has little to do with their sites or sizes.

Three public miners reported electricity costs per bitcoin between April and June 2026 of $38,690, $44,317 and $46,313. Those figures look like a ranking. But differences in what they cover help explain why private miners struggle to borrow.

MARA (NASDAQ: MARA)'s $38,690 covers only its owned and operated sites, according to its Q2 shareholder letter. Third-party hosting and other energy costs appear separately, totaling $69.2 million. Power averaged 4 cents per kilowatt-hour at those owned and operated sites. The figure is clearly labeled; it does not describe the whole operation.

CleanSpark (NASDAQ: CLSK)'s $44,317 covers energy at its owned facilities. Its Form 10-Q says it exited hosted mining by March 31, 2025. Its average power price was 5.3 cents per kilowatt-hour. Its fiscal calendar also differs: April through June is its third quarter, not its second.

Riot's $46,313 is before curtailment credits earned by pausing operations under Texas grid demand-response programs. Those credits totaled $6,335 per bitcoin, according to its Q2 investor presentation. Riot records them outside cost of revenue, then deducts them when reporting its cost to mine. The electricity figure can be reconstructed from $73.5 million in self-mining power costs divided by 1,587 bitcoin mined.

Depreciation adds another layer. Riot reported a cost to mine of $49,912 per bitcoin excluding miner depreciation and $90,631 including it. Its equipment is depreciated over an estimated three years. Change that assumption and reported costs move, even if the machines and power bills stay the same.

None of this is a criticism. Each figure is defined and reconciled. My point is narrower: “cost per bitcoin” can mean electricity alone, site cash costs or a broader measure including depreciation. Each company defines its scope. Riot excludes depreciation from its headline measure because it is a non-cash expense that curtailment cannot avoid.

What makes public numbers usable is that someone else can recalculate them. Filings disclose power expenses, credits, depreciation and bitcoin mined. An analyst can adjust what a measure includes and reach a number they trust. That is an advantage in discussions with lenders. The value lies in the ability to check the calculation, even when readers disagree with the accounting choices behind it.

The same questions, without the footnotes

Public miners' financing is also visible. CleanSpark carried $1.78 billion of debt, almost all in convertible notes due in 2030 and 2032. Riot disclosed a secured facility of up to $200 million, which it repaid and terminated in September. Lenders can examine the filings, footnotes included.

A private miner starts elsewhere. The business may be just as good, and records usually exist, from mining-pool payouts to hosting invoices. What is missing is the path from those records to reported figures.

That path tends to break in three places. First is revenue by site. Pool payouts reach wallets or exchange accounts, but exchange exports often omit deposit addresses. When several sites feed one account, matching bitcoin to sites becomes a manual task. Company totals are easier than verifiable site totals.

Second is the site profit-and-loss statement. Operations teams may estimate it from hashrate, machine efficiency and power consumption. That helps run the site, but does not reconcile with wallet and bank records. The spreadsheet rarely tells lenders which kind of figure they are reading.

Third is scope. Does hosting include power? Are curtailment credits deducted from costs or shown separately? What useful life underlies depreciation? Public miners explain these choices in footnotes. In private reporting packages, they may go undocumented.

AI tools can turn an exchange export into a tidy table resembling a financial statement. But if assumptions remain undisclosed, the table cannot answer a credit analyst's basic question: Where did this number come from? A company total alone cannot resolve that question. The lender needs to see how individual transactions became site revenue and how the resulting figures match the books.

Lenders must reconstruct the figures, ask operators to start over, price uncertainty into loan terms or decline to lend. Each costs time or money. That verification gap can raise financing costs regardless of site quality.

The issue extends beyond mining as operators convert sites to host artificial intelligence computing. The question remains: Is the hardware generating revenue, and can anyone outside the company check?

Monthly figures someone else can verify

The fix applies the logic of public filings to one private company's monthly reporting. Tie inflows to sites and reconcile totals with the books. Separate energy, other operating costs, depreciation and financing. Document assumptions. Show power credits and hosting fees separately, making any netting explicit. Distinguish mining proceeds from bitcoin sold, held, lent or pledged. Record the steps so someone else can reproduce the figures from the source files.

The test is simple: give the reporting package and source files to someone outside the company. If they reproduce the numbers and understand every choice, it works. If they need the preparer to explain it, more work remains.

This is not a valuation, forecast or pitch. It shows what happened during a month and leaves judgment to the reader. I would expect a careful lender to value that more than a confident story.

Editor’s note: TheEnergyMag welcomes op-ed submissions reviewing 2026 and offering forecasts for 2027 across digital asset mining, AI compute and energy infrastructure. Send articles to news@theenergymag. Submissions must be under 1,200 words and must not promote company products or services.