CleanSpark

IThe business

CleanSpark is a Nevada company, run from Henderson, that until this year did one thing: it mined bitcoin. It owns and runs data centers in Georgia, Tennessee, Mississippi and Wyoming, fills them with machines that guess numbers for the bitcoin network, and gets paid in bitcoin by a mining pool.³ Its fiscal year ends on 30 September. At 30 June 2026 it owned about 326,530 machines, 225,137 of them running, with an average age of 23 months. The machines it runs produced 42.6 exahashes a second in June, 4.36% of the network.³ The pool operator, Foundry Digital, is its only customer and can end the contract at any time.³

It calls itself a data center developer now. At 30 June it had about 1,817 MW of contracted power capacity, up from about 1,027 MW nine months earlier, because it bought land and power rights in Texas.³² A megawatt is the unit that matters in this business. A modern AI data center uses a lot of power in a small space, and electricity connected to the grid is the scarce input. Machines can be bought. Grid connections take years.

The old business, in numbers

In the nine months to June 2026 CleanSpark mined 5,545 bitcoin at its own sites and booked $455.6M of revenue, which is the bitcoin valued on the day it was mined.³ In the June quarter a bitcoin sold, on that basis, for an average of $71,692. Energy cost $44,406 a coin, 61.8% of revenue. Depreciation of the machines added $51,871, so the all-in cost of $96,277 was 134% of revenue C (calculated).³ A year earlier the same quarter showed revenue of $98,753 a coin and an all-in cost of $86,428. The business made money then and does not now.

Two facts matter for what comes next. First, the machines wear out. The company depreciates them over three years, and the average age is nearly two.³ Depreciation of $333M in nine months is what the fleet costs to replace, and the company bought almost none: miner purchases were $32M in the nine months against $382M a year earlier.³ Second, the next halving, in about April 2028, cuts the bitcoin paid per block from 3.125 to 1.5625. Unless the price doubles, revenue per machine halves while the power bill stays the same E (estimate).

The new business: Sandersville

On 10 July 2026 CleanSpark signed a lease for 175 MW of critical IT load at a new campus next to its Sandersville, Georgia site. The term is 20 years, with two five-year extensions for the tenant and a further 12-month option. It is triple net: the tenant pays all costs of operating the building. Rent escalates 3.0% a year.¹⁰⁵ The company says it is worth about $6.6B over the initial term, $11.6B with both extensions, and about $330M a year of net operating income on average, close to 100% of rent.⁴ In September it named the tenant: Anviran, LLC, a wholly owned subsidiary of Meta, with Meta guaranteeing rent and operating expenses.⁵ Rent is to start on 30 November 2027, after the first network hall is built.⁵

That is a strong counterparty. Meta had $117B of revenue in the first half of 2026 and a market value of about $1.7 trillion, according to CleanSpark’s own presentation.⁵ The risk in this lease is not whether Meta pays. It is whether CleanSpark builds on time. The lease sets financing, construction and delivery milestones, and failing them “may result in rent abatements or termination.”¹⁰ A third party leads the construction, equipment is on long lead times, and CleanSpark has given the noteholders a completion guarantee: if the money runs out, CleanSpark has to put in more.³⁷

The company puts the cost at $10M to $12M a megawatt, and its September presentation says about $11.9M, or about $2.08B for the 175 MW C (calculated).⁴⁵ To pay for it, a subsidiary, CSDC Finance I, sold $2.276B of 7.875% senior secured notes due October 2031 at 98.5% of face on 25 September, for about $2.24B before costs.⁶⁷ The proceeds finish the build, repay CleanSpark’s own earlier contributions and fund reserves. The principal amortizes from rent income, so the notes are paid down from the lease, not at maturity.⁷ I read this as a project financed almost entirely with debt, secured on the lease. That is why the next table matters.

The company’s own table in the September presentation shows $2.69B left over the life of the lease after debt service, before tax.⁵ But it arrives late. Using the same net operating income and the actual 7.875% coupon, I get $24M to $66M a year for CleanSpark through 2042, then $176M to $332M a year from 2044 once the notes are repaid E (estimate). Meta’s rent is safe. CleanSpark’s share of it is small for fifteen years.

Texas and the rest

With the lease, the tenant signed a letter of intent and exclusivity covering CleanSpark’s whole Texas portfolio: 718 acres and up to 885 MW of “secured and planned” power. That is about 271 acres and nearly 300 MW at Sealy and 447 acres at Brazoria, where transmission supports an initial 300 MW of demand that could grow to 600 MW.⁴ A letter of intent is not a lease. It sets no rent, no date and no price. The company paid about $65.7M for the Austin County land and option, including 1,788,834 shares, and $38.2M for Brazoria, so CleanSpark has little invested in the land compared with the value it implies.³ The rest of the 1.8 GW sits in Georgia, Tennessee, Mississippi and Wyoming, where today it runs miners.³

The bitcoin on the balance sheet

At 30 June CleanSpark held 12,205 bitcoin, and 1,719 more were lent out as collateral, 13,924 in all, worth $814.9M at the June price of about $58,500.³⁸ At $77,200 on 30 October that would be about $1.07B C (calculated).⁹ The cost was $90,188 a coin, so the company is underwater on its holding, and it posts the swings in its income statement: a $587M loss in nine months.³ It sells covered calls and puts against the holding and booked $25.8M of premiums in nine months.³ I do not know what it holds today. The next count is in the annual report, due in late November.

The notes it owes

CleanSpark owes $1.8B in zero-coupon convertible notes: $650M due June 2030, which holders can make it repay in June 2028, convertible at $14.80, and $1,150M due February 2032, convertible at $19.16.³ It raised each by selling the notes and using part of the money to buy back shares: $145M in December 2024 and $460M in November 2025.³ With the shares at $7.67, neither conversion price is near, so I treat the notes as debt. Separately, the Sandersville notes belong to the project and are secured on it. I treat them there.

IIThe moat

I ask two questions. Can a rival take the customer at the same price, and would the customer want to move? CleanSpark has two different answers, one for each business.

Mining has none

Bitcoin mining sells a commodity at a price the miner cannot set. The business is a race on power cost and machine efficiency. CleanSpark is good at it: its fleet runs at 15.8 joules per terahash and it pays about $0.053 a kilowatt-hour.³ But its share of the network fell from 5.14% to 4.36% in a year, from 45.3 to 42.6 exahashes a second, as it stopped buying machines.³ Efficiency is available to anyone who buys the next machine, and the halving takes the margin from everyone. The whole network ran at about 1,006 exahashes a second in late October, so 42.6 would be about 4.2% of it.¹⁴ Whatever edge it had does not show in the June quarter: all-in cost was 134% of revenue.

Power is the scarce thing

What is hard to copy now is connected power near a grid that will give it. The 10-Q says the company controls about 1,817 MW. Sandersville is an example: rent is to start about 13 months after today.⁵ That is worth something, and the price is visible. Hut 8 signed 245 MW at River Bend with Fluidstack for $7.0B over 15 years, backstopped by Google, which is about $1.9M a MW-year. CleanSpark’s lease is $6.6B over 20 years for 175 MW, also about $1.9M a MW-year C (calculated).¹¹⁴ Cipher’s Amazon lease, as announced in its 8-K, is about $5.5B over 15 years for 300 MW, about $1.2M a MW-year on a gross-capacity basis, which is not the same measure.¹⁵ I read this as a market with several sellers and a similar price.

What protects the lease

Once signed, a 20-year triple-net lease with Meta behind it is a strong asset. The tenant has put its own equipment in a purpose-built building; moving costs a lot. If CleanSpark builds on time, the cash is contracted to 2047. That is the moat, and it is narrow: it covers 175 MW out of 1,817.

Evidence it is narrowing

Hut 8 and Cipher each announced a lease of this kind in the past year, so the supply of powered land for AI is growing.¹¹¹⁵ A moat is something a competitor cannot match. Here the competitors are matching it with the same pitch. The test is Texas: whether the tenant chooses to sign 885 MW of it, on what rent, in what time.

IIIThe competitive advantage

What CleanSpark has is speed and an early start. It owned land, power rights and a skilled crew before the AI buyers came looking. In July its chief executive said the company had always believed in a “second-mover advantage”: grow the portfolio as the market matures, then execute on excellent terms.⁴ I take that as a fair description. The lease terms look no worse than Hut 8’s.

Three real things

First, power cost and quantity: about 5.3 cents a kilowatt-hour across its sites, and about 1.8 GW contracted.³ Second, a hyperscaler with a guarantee: Meta stands behind rent and operating costs.⁵ Third, a treasury that can fund equity. At the 30 October bitcoin price the 13,924 coins are worth about half the company’s market value C (calculated).⁹¹ Developers that lack that have to issue shares to build.

Against the alternatives

The table puts the price of CleanSpark’s power next to what the market pays for similar leases. It is what the buyer pays for the power, building and the tenant’s credit, not what the stock is worth.

The lesson: CleanSpark is not getting a better price than peers. It is getting similar terms and, in the guarantee, similar credit. Its edge is in how soon it can build and how little equity it must put in. Both have to be proved by delivering rent in November 2027.

IVManagement

S. Matthew Schultz is chairman and chief executive. Gary Vecchiarelli is president and chief financial officer. Zachary Bradford, who ran the company through its mining years, left as CEO on 10 August 2025.¹² The same people signed the lease, raised the notes and ran the buybacks, so I follow what they are paid to do.

How they are paid

Schultz’s pay for the year to September 2025 was $44.9M: $0.86M in salary, $5.5M in bonus and $37.1M in stock awards. The median employee earned $91,263.¹² Each executive also gets bitcoin every month, 1.2 BTC for Schultz, which stops if the company stops mining.¹² That is a small point today and a large one later: the company plans to end mining at Sandersville in fiscal 2028.³

In March 2026 the board made two new grants. The 2026 long-term plan gave 5,411,148 awards to 22 leaders, 40% time-based units and 60% performance units. Half the performance units pay on megawatts under executed AI leases by 20 March 2027: nothing below 600 MW, half at 700 MW, all at 800 MW or more. The other half pay if the 20-day average share price reaches twice $9.40.³ Then a one-time award of 7,698,000 units went to three executives. Half pay on AI capacity ready for service by 30 September 2030, from 25% at 1.0 GW to 100% at 2.5 GW. The other half pay only if the shares reach five times $9.40, $47.00, with full payout at $94.³ The company expects half, 3,849,000 units, not to vest.³

What does that pay them to do? To sign and energize megawatts, quickly, and for the highest paid, to make the share price much higher than today’s. It does not pay for cash left to shareholders per share. Sandersville is a lease whose notes take most of the cash for 15 years. It counts toward the 800 MW target in full E (estimate). I do not say the board is wrong to want leases. I say the pay rewards the thing that is easiest to do and does not penalize the thing that matters, the return on each dollar.

What they own, and what they have done with it

At 9 September 2026 Schultz held 3,039,089 shares, about 1.2% of the common stock and $23M at $7.67.¹³ At January 2026 all current directors and officers together held 2.08%.¹² The Forms 4 I read from October 2025 to October 2026 show share awards, tax withholding and sell-to-cover sales. They show one director sale, 85,315 shares at $11.69 on 23 December 2025. I found no open-market purchase by any insider.¹³ Insiders have chosen not to add to their holdings as the price fell from $23.20 in mid-October 2025.¹

One more fact about control. The 1,750,000 Series A preferred shares carry 45 votes each, 78.75M votes, about 23.5% of the vote at the record date, and are held by Schultz, a director (Larry McNeill), their company and Bradford.¹⁶¹² Until March the preferred earned a quarterly dividend of 2% of earnings before interest, tax and amortization. Schultz received $3.07M of it in fiscal 2025.¹² On 20 March 2026 the board, without the two interested directors, ended that dividend in return for a one-time payment of $17.14 a share, $30.0M in all, paid on about 24 March.¹⁷³ The preferred still votes with the majority of its own holders.¹⁷

What they did with capital

Since 2021 the company has raised about $2.2B from selling shares, $271M, $125M, $376M, $1.23B and $187M in fiscal 2021 to 2025, and about $1.66B of it has gone into new miners.¹⁸²¹⁹ Weighted shares went from 102.7M in fiscal 2023 to 282.2M in fiscal 2025.² In fiscal 2024 it wrote off $197M of fixed assets.²²⁰ Then it sold zero-coupon convertible notes twice and used part of the proceeds to buy back stock: 11.76M shares for $145M in December 2024, about $12.33 a share, and 30.6M for $460M in November 2025, about $15.03.³ That is $605M for 42.4M shares worth about $325M at $7.67 C (calculated). The buybacks were funded by borrowed money that does not have to be repaid until 2028 and 2032, and by the sale of convertibles that now sit far out of the money.

My read: this is a management that raises capital easily and spends it quickly. The test of the new plan is a different one, whether it earns more than its cost on each lease. The Sandersville structure suggests it may: it put in little equity, kept a long contract and got a guarantor. The pay plan suggests it will be paid for more of the same whether or not it does.

VFinancials

CleanSpark began mining bitcoin in December 2020. Before that it sold energy software and microgrids, with revenue of $10M in fiscal 2020 and less before.¹⁹ So there are six full fiscal years and the twelve months to June 2026, not ten, and only the last five are the mining company. Revenue rose from $39M in fiscal 2021 to $766M in fiscal 2025, then fell to $679M over the twelve months to June 2026 C (calculated).¹⁸²³

Quality of earnings

Net income is a poor guide here. In fiscal 2025 the company reported $364M of net income. It included a $425.6M gain from valuing bitcoin at its year-end price and a $92.2M gain on bitcoin lent out as collateral.² Take the first out and operating income was about −$107M C (calculated). In the nine months to June 2026 the same two items were losses of $587.2M and $159.0M.³ Without the first one, the operating loss was $307M against $41M a year earlier C (calculated).³ So the earnings are the price of bitcoin, with a business under them that was losing more.

What an owner could really take out

Nothing so far. Over the twelve months to June 2026, operating cash flow was −$529M, because mined bitcoin is not cash until sold. Counting the bitcoin as cash gives +$151M. Capital spending was $160M and stock pay $73M, so owner cash flow was −$82M C (calculated).²³ Over fiscal 2021 to 2025 the total was −$1.65B C (calculated). The reported net income in those five years added up to about +$2M, a gap of $1.65B between the accounts and the cash an owner put in.

How much of the capital spending is needed just to stand still? Depreciation was $442M over the twelve months, and $333M of it in nine months was on miners.³ If the fleet were to be kept at the June size, the cost would be near that, not $160M. At that spending owner cash would be about −$365M E (estimate). This is why I do not value the mining business as a going concern. It is a fleet running down.

The balance sheet

At 30 June 2026 the company had $202.6M of cash, $814.9M of bitcoin and collateral receivable, $1.81B of debt at face, almost all zero-coupon converts, and $761M of book equity.³⁸ Book equity was $2.18B nine months earlier. The fall is the $997M net loss and the $463M of buybacks.³ It has no coupon to pay on the converts. It does have two dates: June 2028, when holders of the 2030 notes can demand $650M, and February 2032, when $1,150M falls due.³

The new debt came after the quarter. The $2.276B notes belong to a subsidiary, are secured on the Sandersville lease and are not guaranteed by CleanSpark, except for the completion guarantee.⁷ A company-wide stress would not trigger them, but a failed build would reach the parent. Other claims: the Customs agency has billed tariffs on miners and if it loses, the total bill could rise to about $130M, plus interest. It has not set aside any money.³ A securities class action filed in 2021 is still open, though the court on 30 June 2026 excluded the plaintiffs’ expert on loss and damages.³

VIValuation

First, how this goes badly wrong

Before any number, I ask what would most likely hurt an owner, and work back from there. I see three ways.

The build slips or costs more, and the parent pays. CleanSpark has promised the noteholders it will cover any shortfall in finishing Sandersville, and the lease lets the tenant cut rent or end it if milestones are missed.⁷¹⁰ A 15% overrun on $2.08B is about $312M, or $1.12 a share C (calculated). A longer delay does more harm, because the notes accrue interest from the day they were sold and rent has not started.

The cash from the lease is spoken for. In my base, Sandersville leaves $37M in 2029, against overhead and stock pay of $113M in FY2029, while mining winds down E (estimate). So until Texas signs, the company lives on its bitcoin and cash. That is $1.28B at today’s price, against $1.8B of convertible notes that come due in 2028 and 2032. If Texas never turns into a lease, the treasury runs down and the notes need a new lender.

Bitcoin falls further or stays here. The treasury is about half the market value. At $55,000 it loses $0.3B. And mining at today’s price loses money once the machines wear out, so no help comes from there.

What an owner gets today

Nothing. Over the twelve months to June the reported net loss was $998M, but owner cash was −$82M, because most of the loss was bitcoin marks and depreciation, not cash. Treat that as luck: the fleet is being run without replacement. If it were replaced the cost would be near $440M a year E (estimate). In my base, owner cash is −$119M in FY2027 and −$327M in FY2028, before the convertible notes, because the first Texas leases need their equity put in before they pay.

What I can’t judge

Five things. What bitcoin the company holds today, since the last count was 30 June and it sells coins to fund itself. What rent Texas will pay: the letter of intent sets none. What the extensions pay: the lease says “annual escalators” and I assumed the 3.0% in the presentation continues. What overhead will be after mining ends: if it is half my number, the base is higher by $1.52 a share C (calculated). And how much of Sandersville’s power today runs miners, so how fast the fleet shrinks. I use ranges. The wide gap between bear and bull is the honest answer.

How I value it

I add up the parts, because the company is now several things at once. Each is the present value of cash after stock pay and after the capital needed to stand still, discounted at 13% for mining, overhead and new leases, 10% for the Meta-guaranteed Sandersville cash and 9.5% for the convertible notes, which I subtract at present value. Years are fiscal, to September, from 2027 to 2057, the end of the extensions. Taxes use 15-year depreciation on the buildings, which keeps cash taxes near zero until the notes are paid. The odds are mine and are not used in the values. I do not forecast bitcoin: base holds it at today’s price, bear at $55,000, bull at $110,000.

Read the base column down. The new leases are worth $3.26 a share, Sandersville $1.71, mining $0.62. Overhead takes −$3.04. The bitcoin and cash, $4.57, nearly offset the notes, −$4.37. So almost all of the base value is the lease business after its own costs, and most of that is leases not yet signed E (estimate).

What the price says

At $7.67 the market is paying for much more than my base. Holding my other base inputs, the price needs about 1,170 MW built and rented on Sandersville’s terms, or about $101,000 bitcoin with all 885 Texas MW built E (estimate). The Texas letter covers 885 MW at most. The company would need to find another 285 MW or more beyond it, and also put in $2.1B of equity at 15% of cost, from a treasury of $1.1B of bitcoin and $0.2B of cash.⁴³

How much the answers move

The discount rate matters, but not enough to explain the price. Cutting every rate by three points gives $4.64, below the price. The bigger swing is how many megawatts get leased, and what bitcoin does. The grid is below.

The margin of safety

My margin-of-safety price is $1. The bear case is zero in the model, so I set it where the market would pay only about $0.3B for a company with a signed Meta lease, 1.8 GW of power and a treasury that nearly covers its notes. In my base, the treasury net of the notes is worth almost nothing, so at $1 the market is paying for part of Sandersville and giving Texas away. That is not a floor. If Sandersville fails the equity can go to zero.

The price against the range

At $7.67 the price is 193% above my base value of $3 (the model gives $2.62 before rounding) and 45% below my bull of $14. The market is paying for a good part of the bull case. Its case is that Meta’s tenant signs all of Texas soon, that CleanSpark builds at Sandersville’s cost, and that debt stays available at 8%. Each has happened once, at one site. I give the full set a 7% chance together, in the table below. The market may know more than I do about the tenant’s plans. I cannot see them.

Against the alternative

The next best use of the same dollar is a business that already collects rent on power for data centers. The closest on this site are Constellation Energy, whose base value implies +7.7% a year to year 10, and Vertiv at +3.7%. My base for CleanSpark gives +1.1% a year on the same method, below both, and it needs more things to go right E (estimate).

Quality and price, separately

The quality question is whether the business earns more than the capital it uses. Mining does not at today’s price. The Sandersville lease does, on a project basis, with a strong tenant. Whether the whole company does depends on Texas and on how much of the lease income it pays out in overhead and stock. The price question is separate: $7.67 sits above my base and below my bull. A good lease at an unclear price is not a good purchase by default. I leave the decision to the reader.

How the odds multiply

The price needs several things at once. Multiplying my guesses gives 7%. The odds are mine and are not used in the values.

Where this could be in 5 and 10 years

These are my estimates of value, not forecasts of the price.

The table gives the model’s value per share at each date: the cash and bitcoin kept by then (bitcoin at today’s price), plus the present value of owner cash after it, less the notes still owed, per 279M shares today, in each year’s dollars. The model pays nothing to owners before 2036: cash stays in the company to meet the notes and fund new leases. The bear falls below zero in both years because the treasury is spent before the notes fall due, so I show $0.

For the base to be right at five years, Sandersville must be paying rent, 400 MW of new leases must be signed and under construction, and the 2028 notes must have been repaid from the treasury. It rises from $3.72 at five years to $8.53 at ten because the new leases come into rent and their debt is paid down. It needs rents near $1.45M a MW-year, 3% escalators, debt at about 8% and a discount rate near 13%. The most likely way it is wrong is that Texas is smaller or later, or that the rent is lower; each 100 MW is worth about $0.65 a share today, $1.39 at five years and $2.59 at ten.

The bull is the same business with 1,285 MW leased and bitcoin at $110,000. The bear is Sandersville alone and bitcoin at $55,000. A reader’s guide: the market price can sit above or below these values for years.

VIIThe case for the price

What most people miss

Three things cut in CleanSpark’s favour and are easy to overlook. First, the tenant is not a start-up renting from a miner. Meta, through a subsidiary, signed, and Meta guarantees rent and operating costs.⁵ Second, CleanSpark has put in little of its own money. The September notes pay for the remaining build, repay its earlier contributions and fund reserves, so the company says the equity for Sandersville is fully funded.⁷⁸ Third, the convertible notes cost nothing in cash. They carry no coupon, and the first date that matters is June 2028.³ So there is time to sign Texas.

The bull case

In plain words: AI buyers are short of powered land, CleanSpark has it, and one signed lease proves buyers will pay a fair price for it. The chairman and CEO said on 14 July that the lease validates the land-and-power strategy, and that the company had always believed in a second-mover advantage: grow as the market matures, then execute on excellent terms.⁴ If Texas follows on the same terms, the company is a landlord with 1,285 MW and 3% rent escalators, and 20-year contracts. That is my bull, $14 a share.

Holders say it with money more than words. Situational Awareness LP, the fund run by Leopold Aschenbrenner, held 12,276,139 CleanSpark shares, worth $178.6M, at 30 June 2026, according to its 13F-HR filed on 14 August 2026.²¹ That is a dated fact. The 13F does not show shorts or holdings outside the United States, and I do not read it as saying what the fund holds now or will do. Not everyone agrees: no insider bought shares in the market in the past year.¹³

Against reality

Today, one 175 MW lease is signed and 885 MW of exclusivity is not a lease. If CleanSpark builds Sandersville on time, Texas converts at about $1.9M a MW-year, and bitcoin holds, then a price near $8 is possible only if it also finds a further 285 MW of customers. My base is 400 MW and $3. The distance between them is 2.9 times the Texas I have in my base, or $101,000 bitcoin C (calculated).

What a 10x would take. Ten times $7.67 is $76.70. At that price the convertible notes would convert and the count would rise to about 377M shares, a market value near $29B C (calculated). On my model that needs about 15,138 MW rented on Sandersville’s terms, 8 times CleanSpark’s whole contracted capacity of 1,817 MW, and about $29B a year of rent E (estimate).³ The shares have done it before. They rose 13 times from $1.78 on 19 December 2022 to $23.40 on 25 March 2024, and the highest close ever, $40.39, was on 7 January 2021.¹ That was off a low base. A tenfold move from here is a new, much larger company, and I would not build a case on it.

Where I come down

The Meta guarantee and the way the notes were structured moved me: Sandersville needs little of the company’s own cash. The buybacks at $12 and $15 and a pay plan that rewards megawatts did not. A signed Texas lease at or above $1.9M a MW-year, with debt near 8%, would raise my values most, and I would need a lower overhead to move the base toward the price.

VIIIRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest.

• One tenant. All the contracted rent comes from one Meta subsidiary. The guarantee covers rent and operating expenses, but the lease still lets the tenant cut rent or end it if CleanSpark misses its milestones.⁵¹⁰
• Construction and equipment. A third party leads the build, equipment has long lead times, and the project needs regulatory approvals and power.³
• Dilution and financing. The company says it will need to raise substantial additional capital for Sandersville and any other site, perhaps equity or equity-linked.³ It sold $1.23B of stock in fiscal 2024 and $187M in fiscal 2025, and could sell more if the treasury ran short.²
• Bitcoin. The treasury and the mining margin both depend on it, and the company has lent coins out as collateral.³
• Mining wind-down. The company is still assessing impairment on the Sandersville miners and cannot estimate the charge.³ If the company stopped mining altogether, the executives’ monthly bitcoin pay would end.¹²
• Legal. A securities class action from 2021 is open, a shareholder derivative case is open, and Customs says the total tariff bill on its 2024 miners could rise to about $130M.³
• Control and pay. About 23.5% of the vote sits with insiders through the preferred shares.¹⁶ Pay is built on megawatts and high share-price hurdles.³
• Demand. The 10-Q lists a change in AI infrastructure needs as a risk to the value of the project.³

Culper Research short reports on CleanSpark

Who and what. Culper Research, a short seller, says in both reports that it is short CleanSpark, so it gains if the price falls.²⁵²⁶ The reports are “Cleanspark (CLSK): Back to the Trash Can” (14 January 2021) and “CleanSpark, Inc. (CLSK): Mining Operations Don’t Add Up” (18 June 2021). Culper’s site blocks automated access, so I read a court copy of the first and an Internet Archive copy of the second.²⁵²⁶

What it alleges. The January report says ATL Data Centers, a College Park, Georgia site CleanSpark bought in December 2020, was the rebranded assets of a bankrupt operation, and that Marathon dropped a bid in September 2020 because ATL’s “subsidized” power rate would end in three years, undisclosed by CleanSpark (pp. 2 to 6).²⁵³² It also says CleanSpark overstated or fabricated customers and contracts in its energy business and had undisclosed related-party deals (pp. 7 to 12).²⁵ The June report says reported mining costs for the March 2021 quarter, $611,863, do not reconcile with $693,144.70 of ATL power bills, that city emails call out a rate change “starting in 2024”, and that a “strategic contract” is spread over five years, not three (pp. 2 to 6).²⁶

The response. On 21 January 2021 CleanSpark filed an 8-K press release calling the accusations false, saying it “stands behind its prior statements and disclosures” and that the publisher was hidden and short the stock. It names no claim and does not address ATL’s power rate. I found no reply to the June report.²⁷²⁹

Established, and not. Marathon said on 17 September 2020 that it withdrew its bid for Fastblock Mining over a subsidized power rate ending in three years; the link to ATL is Culper’s.³² CleanSpark’s filings call its ATL “strategic contract” a critical-input supply “at significantly low prices compared to market”, amortized over five years¹⁸ to nil by 30 June 2026.³

The suit is Hasthantra v. CleanSpark, Inc., No. 1:21-cv-00511 (S.D.N.Y.), filed 20 January 2021 as Bishins.²⁸ It alleges CleanSpark misled investors about ATL and cites the January report as a corrective disclosure.³⁰ The motion to dismiss was denied on 5 January 2023, taking the complaint’s facts as true.²⁹ A class of CleanSpark buyers was certified on 24 September 2025.³⁰ The plaintiffs’ loss-causation and damages expert was excluded on 30 June 2026 and an amended report refused on 14 October 2026.³¹³ The docket shows no settlement, judgment or appeal; I could not read entries 128 to 130: two letters and a memo endorsement, 28 October to 9 November.²⁸ CleanSpark says the claims are without merit and it cannot estimate a loss; its 10-K says they appear derived from a short seller report.³² None of these rulings decides whether a Culper claim is true; I found no finding that does.

In my model. Only the power claim reaches an input, my 5.3 cents a kilowatt-hour; existing mining is $0.62 of my $2.62 base. ATL drew about 10 MW in 2021 on Culper’s bills, about 1.5% of today’s 670 MW fleet C (calculated). Doubling its rate takes about 2.5 cents off the base; if College Park is now 5% of the fleet, about 8 cents E (estimate). No filing I read gives its size. The class action has no line in my base; each $100M paid is about $0.36 a share C (calculated). The customer and related-party claims concern 2017 to 2020 deals, mostly in the energy business held for sale in 2022, so sit outside it.¹⁸

So what

These are 2021 claims about one site and an old business; my value rests on the Sandersville lease, Texas and bitcoin. Culper is right if College Park’s power cost rose sharply from 2024 and 2021 costs were understated; CleanSpark is right if the costs reconcile through items outsiders cannot see and College Park’s power stayed cheap. Culper gains if the price falls; management and insiders are paid in the stock. The court record and filings are knowable now; what College Park pays, why the 2021 figures differ and what the parties told the court after 14 October are not. Watch the fiscal 2026 10-K, due by 30 November, for its legal note and owned-facility power price, and the docket.

Geopolitical risk, and what it means for the value

CleanSpark has no foreign plants, customers or currency. Its exposure is to American power rules, American trade policy and the bitcoin price. I rerun my base for each case below and show the value per share.

Power. The Texas land is in Austin and Brazoria counties. In July Austin County put a temporary moratorium on data centers and battery storage, and on 3 August Governor Abbott told ERCOT and the Public Utility Commission to audit every data center in the interconnection queue before it moves on. The company says it is still weighing both.³ Texas law since June 2025 also lets ERCOT cut large new loads in an emergency.²² In Georgia, a new Georgia Power customer above 100 MW can be given longer contracts and minimum bills, and the contract goes to the state commission.²³ I could not confirm which utility serves Sandersville, so I do not know whether that rule applies. The Meta guarantee covers rent and operating costs, but the lease still lets the tenant cut rent or end it if CleanSpark misses a milestone, and a late power connection misses one as easily as a late builder.¹⁰ Tennessee matters less: those sites mine, and my base already runs mining down.

Tariffs and rules on mining. Customs says some miners CleanSpark imported in 2024 came from China, not where the seller said; if it wins, the total could reach about $130M.³ In August 2025 the gap between origins was large, 57.6% against 21.6% for Southeast Asia, and I did not check today’s schedule.²⁴ My base charges $30M. Tariffs on new machines matter little because the base buys almost none; tariffs on electrical equipment would raise build cost while rent stays fixed. The filing calls mining largely unregulated, expects more rules, and describes Georgia, Mississippi and Tennessee as favorable.²

Odds, with my basis. Texas two years late or smaller: about one in three, because two government actions arrived within four months. Sandersville rent a year late: about one in five, with equipment already paid for but a third party building. Meta ending the lease: under one in ten, because it is the tenant and the guarantor. Customs winning in full: about one in four, because some protests were approved. I do not forecast bitcoin. These are judgements, not forecasts of events.

The price needs about 1,170 MW, so it assumes none of these delays. My base holds only the $30M Customs charge, so each case below widens the gap between value and price.