CleanSpark Deep-Dive Value Research

On the "can it be understood" question, this company is in fact not hard to understand: it essentially uses electricity and ASIC hashrate to produce Bitcoin, while trying to convert existing power campuses into AI/HPC campuses. The difficulty is not "what business is this" but "is it a good business for the long term." This is not a classic branded consumer product, nor high-switching-cost software; it is closer to a capital-intensive, fast-iterating, commodity-type industry whose returns are heavily exposed to external variables. If the stock market closed for five years, I would only grudgingly accept holding it if I treated it as a "Bitcoin-infrastructure position + AI option" rather than a stable cash cow.

At the industry level, Bitcoin mining remains a growing cyclical industry, not a mature steady-state one. Long-term demand depends on Bitcoin network activity and the coin price, but industry profitability is strongly shaped by halvings, network-wide hashrate competition, equipment refresh cycles, and power policy. The main competitors the company lists in its 10-K include MARA, Riot, Core Scientific, Bitfarms, IREN, Cipher, and TeraWulf; the company also acknowledges that, in AI/HPC expansion, it faces more mature, better-capitalized players such as Equinix, Digital Realty, and CoreWeave.

Combining the facts above, I assign the following scores:

The view behind these scores is: this is more "a relatively well-executing company in a bad industry" than "a great company in a good industry."

Moat and Management

If you take the moat apart, CleanSpark's advantage lies not in brand, network effects, or switching costs, but in power procurement, site development, operating execution, and a degree of scale. The company had 1,027 MW of contracted power at the end of FY2025, expanding to 1.8 GW by April 2026; hashrate rose from 45.6 EH/s in September 2025 to 50.0 EH/s in April 2026. This shows it genuinely has execution ability in acquiring land, securing power connections, building, and racking machines.

But by a stricter Buffett-style standard, this moat is not deep. The company itself discloses that it has no patents related to its existing and planned Bitcoin-mining business; in AI/HPC it faces incumbent data-center players with long-term power commitments, customer relationships, and far larger capital resources. It has no network effects, switching costs are essentially negligible, and brand confers no decisive advantage on per-unit hashrate economics. Its "edge" is closer to relatively low cost and relatively faster execution than to a structural monopoly that rivals could not replicate in a decade.

My judgment on the moat is therefore: localized advantages, no deep moat. The moat most resembles "redeployable power and campus-development capability" rather than "a business model that can keep raising prices." In an inflationary environment, the company cannot simply pass through cost via price increases, because it does not set mining revenue; in a downturn or a coin-price decline, it cannot prop up margins with subscriptions and high renewals the way a consumer brand or software firm can. A large part of its historical margins came from changes in the fair value of Bitcoin — accounting volatility rather than a steady-state operating competitive advantage.

On management, I assign a neutral-to-cautious assessment. The positives: management is clearly stronger than some smaller miners at expanding its own infrastructure, improving per-machine efficiency, and building larger power reserves; it also conducted sizable buybacks in 2025 and 2026. After issuing the 2030 convertibles in December 2024, it repurchased about 11.76 million shares for roughly $145 million, equivalent to about $12.33/share; after issuing the 2032 convertibles in November 2025, it repurchased about 30.60 million shares for roughly $460 million, equivalent to about $15.03/share. On price, neither buyback is unreasonable, nor is either an obviously reckless buyback at a high.

But the problems are equally clear. First, historical share dilution is very heavy: common shares grew from 55.66 million at the end of September 2022 to 270.9 million at the end of September 2024, and further to 284.3 million at the end of September 2025; in FY2023, FY2024, and FY2025 the company also raised net equity proceeds of roughly $384 million, $1.222 billion, and $187 million respectively. In other words, the company first issued a great deal of stock to raise money, then used convertibles in 2025/2026 to fund buybacks — capital allocation is not purely the textbook "buy low, raise sparingly," but carries a strong dependence on cyclical capital markets.

Second, the alignment between shareholders and management in common stock is modest. The 2026 proxy statement shows Zach Bradford's beneficial common-stock ownership at about 4.737 million shares, or 1.85%; current executives and directors together hold about 5.3299 million shares, or 2.08%. This is not "no ownership," but it does not amount to deep alignment with common shareholders either. Add FY2025 stock-based compensation expense of $45.33 million, with the equity-incentive portion of compensation alone running about $22.22 million in the first half of FY2026, plus heavy reliance on RSUs/PSUs in long-term executive incentives, and "per-share value growth" will keep being constrained by dilution.

Third, the AI/HPC direction is for now more vision management than proven capital-allocation success. The 2025 annual report and the April 2026 operating update both emphasize the AI platform, a new Texas campus, the Submer partnership, and pursuing a hyperscale customer, but as of May 2026 the verifiable information is still "pursuing the first customer." By contrast, Riot had already disclosed Data Center revenue in Q1 2026 and an AMD expansion to 50 MW of critical IT capacity; this shows CleanSpark's AI narrative is at an earlier stage that still needs evidence.

On balance, I assign the following scores:

Financial Quality and Owner Earnings

Start with the overall judgment: headline profit is highly unreliable, and true cash earning power is far weaker than accounting profit. In FY2025 the company recorded $766.3 million of revenue and $364.5 million of net income, but this included $425.6 million of Bitcoin fair-value gains and $92.2 million of Bitcoin-collateral fair-value gains; in the first half of FY2026, with the Bitcoin price falling from about $114,100 to about $68,200, it recognized $470.9 million of Bitcoin fair-value losses and $142.5 million of Bitcoin-collateral losses, widening the six-month net loss to $757.1 million. This shows net income is to a large extent "recording coin-price swings," not "recording distributable operating cash flow."

I prefer to treat 2022–2025 as a "comparable but still flawed" sample window, for two reasons: first, around 2022 the company was still transitioning from its old energy business toward Bitcoin mining as its core, and the 2023 10-K still includes discontinued operations; second, from 2024 the Bitcoin accounting basis changed, and the 2025 annual report shows the cumulative effect of the accounting-policy change. So the historical trend can be examined, but should not be mechanically extrapolated.

The table below shows only what I consider the most critical and relatively comparable data:

Note on the data basis: 2022 and 2023 data come from the 2023 10-K; the restated/audited 2023–2025 data come from the 2025 10-K. The "direct capex proxy" above uses "rig payments/including prepayments + property and equipment purchases" as a rough representation, and excludes all cash outflows from acquisition transactions; the "reported-basis FCF proxy" = operating cash flow minus that proxy. Because miners first record mined Bitcoin as an asset, and sales often show up in investing activities, operating cash flow and free cash flow are systematically distorted relative to true economic earnings, so this can serve only as a conservative reference, not as a directly usable FCF for a traditional industrial company.

On revenue growth, the company has indeed grown fast: from 2022 to 2025, revenue rose from $131.5 million to $766.3 million. But this growth is capital-heavy, and each round of growth came with higher depreciation and amortization, higher equipment-refresh needs, and higher financing needs. FY2025 depreciation and amortization reached $348.3 million, nearly 45% of revenue; in the first half of FY2026, D&A rose further to $222.2 million. This is the classic capital-intensive trait of "growth that does not automatically convert into stable owner earnings."

On earnings quality, the three most critical facts are these. First, the 2025 annual report discloses that the estimated useful life of rigs has been changed to 3 years, which itself shows the economic life of the equipment is not long. Second, the company states clearly that, to hold its competitive position, it must keep adding new rigs, both to replace worn equipment and to chase network-wide hashrate growth. Third, in FY2025, rig and equipment payments plus property and equipment purchases together exceeded $560 million. So even if accounting shows net income in a given year because the coin price rose, over the long run this business still devours large amounts of capital.

I estimate owner earnings with a more conservative method. Take FY2025 as the example: Facts: net income $364.5 million; Bitcoin fair-value gains $425.6 million; collateralized-Bitcoin fair-value gains $92.2 million; depreciation and amortization $348.3 million; stock-based compensation $45.33 million.

My conservative estimation method is as follows. Step one, strip out of net income the coin-price-related, non-stably-distributable fair-value gains; the resulting "adjusted earnings" roughly turn into a loss. Step two, add back depreciation and amortization, but do not treat stock-based compensation as a fully ignorable non-cash expense, because for common shareholders it is a real dilution cost. Step three, deduct the maintenance capex required to hold the current competitive position. Here the company does not directly disclose maintenance capex, so I can only make an inference: given an average rig age of about 15 months, a 3-year life, and the company's own admission that it must keep refreshing rigs, I believe FY2025 maintenance capex of at least $250 million to $350 million is a reasonable, conservative range.

Under this framework, my conservative Owner Earnings estimate for FY2025 is about $0 to $100 million, with a midpoint of $50 million. This is not a management-disclosed figure but my observational inference. The implication matters a great deal: even in a year of a sharply rising coin price and very flattering accounting profit, the company's true distributable cash flow may not be abundant. At the current price of about $14.69/share, the market's valuation of that $50-million-scale owner earnings is still on the expensive side; and once the first-half-FY2026 coin-price drawdown and heavy depreciation are factored in, owner earnings could well turn negative again.

On financial risk, I see no direct evidence of financial fraud. The audit and internal-control opinions accompanying the 2025 10-K are unqualified. The real problem is not that "signs of fraud are obvious," but that the financial statements are extremely hard to read and very easily distorted by Bitcoin and collateral fair-value swings. So with this kind of company, the most dangerous thing is not "whether there is profit on the books," but investors mistaking accounting profit for steady-state cash profit.

Intrinsic Value and Margin of Safety

The current market price is as follows:

First, my core view: buying at the current price does not offer an adequate margin of safety. The reason is not that the company is bound to fail, but that the market has already priced in a fair amount of "AI monetization succeeds + mining competitiveness stays in the lead," and those expectations have not yet been adequately validated by contracted cash flow.

Discounted Owner Earnings

The valuation below is not a precise answer but a range judgment of "roughly what the business is worth today under different operating realities." All scenarios are based on the Owner Earnings I defined above, not GAAP net income.

What most needs emphasizing here is that these figures are extremely sensitive to maintenance capex and to whether the AI business can be monetized. If AI/HPC stays at the story level while the mining business must keep investing in equipment to chase network hashrate, then even the neutral scenario may be too optimistic. Conversely, if over the next two or three years the company signs high-quality long-term contract customers, letting the power campuses earn higher and more stable returns than pure mining, the optimistic scenario has room to hold.

Based on this model, I give three ranges:

The current price of $14.69 is already above what I consider the more solid "fair-value range," yet below the "blue-sky scenario after AI/HPC delivers smoothly." This means that buying today is not buying "a very-high-certainty bargain," but more like paying for an option that has not yet been commercially proven.

Relative Valuation

Looking at this sector strictly through PE, EV/EBITDA, and P/FCF tends to produce distorted conclusions, because most miners are distorted by Bitcoin fair-value accounting and high capex. Even so, on a rough "operating assets/reserves/market cap" comparison, CLSK is far from cheap. CleanSpark's current market cap is about $3.35 billion, with operating hashrate of 50.0 EH/s and Bitcoin holdings of 13,453 coins; MARA's market cap is about $4.73 billion, holding 35,303 coins of Bitcoin at quarter-end, with Q1 2026 revenue of $174.6 million and total hashrate of 72.2 EH/s; Riot's market cap is about $7.87 billion, with Q1 2026 revenue of $167.2 million, of which $33.20 million was data-center revenue, and AMD has already expanded its contracted capacity to 50 MW.

The inference from this comparison is that the market's pricing of CleanSpark already reflects, to a considerable degree, expectations of its "AI data-center option." Otherwise it would be hard to explain why, with Bitcoin reserves far smaller than MARA's and AI commercialization lagging Riot's already-disclosed data-center revenue, it still commands a not-low asset premium. Put differently, the fact that peers are not cheap does not mean CLSK is cheap; more likely it means the entire sector is capitalizing future narrative ahead of time.

Asset Value

From an asset angle, CleanSpark is supported by three parts: cash, Bitcoin/collateralized Bitcoin, and hard assets such as data centers/power/land/rigs. As of the end of FY2025, the company had $42.97 million of cash on its books, Bitcoin (current + non-current) of about $1.189 billion, Bitcoin-collateral receivables of $294.6 million, and net property and equipment of $1.364 billion. But note that this is only a snapshot at September 30, 2025; by the first half of FY2026, the company's net loss had reached $757.1 million and it had carried out another $460 million of buybacks, so book equity has been noticeably eroded.

For this kind of company, book net assets are not a solid floor. The inference is: cash and Bitcoin are relatively easy to value; the real trouble is the rigs and some of the supporting equipment. Rigs have short useful lives, technology advances fast, and second-hand residual values are unstable, so in a downturn or a coin-price slump the markdown could be large. I therefore would not assign a high liquidation score to book property and equipment. Instead, I prefer to understand CleanSpark's asset value as: a portion of monetizable Bitcoin assets + a portion of power/campus options that may have value for AI/HPC + a large block of specialized equipment that depreciates fast.

Combining the three methods, I give the following price ranges:

This is not to say $14.69 is bound to fall, but rather: buying at today's price, you need many good things to happen together to earn a sufficiently good long-term return.

Risks, the Bear Case, and Opportunity Cost

What I care about most is not short-term share-price volatility but permanent loss of capital. For CleanSpark, the most important risks, in order, are: First, competition and cyclical risk. Bitcoin mining is a network-wide competition, with every miner chasing the same block rewards; the company itself states that, to hold its competitive position, it must keep buying new rigs. After the halving, if network hashrate keeps growing faster than the company, the return per unit of capital will be continuously compressed. Second, AI/HPC execution risk. As of May 2026, the company is still pursuing its first hyperscale customer, while Riot has already disclosed data-center revenue and the AMD capacity expansion. The strongest bear view would say: the market has already granted CleanSpark an AI premium, but actual signing and cash collection have yet to appear. Third, leverage and dilution risk. The company issued $650 million and $1.15 billion of 0% convertibles at the end of 2024 and the end of 2025 respectively; although the coupons are low, they bring future dilution and refinancing complexity, and the historical scale of equity issuance has been large. Fourth, customer/supply-chain concentration risk. The current sole pool customer, the key rig suppliers, and multi-site utility power all constitute concentrated dependencies. Fifth, accounting and statement-readability risk. When profit is heavily affected by Bitcoin fair value, collateral fair value, and derivatives, investors very easily mistake book volatility for operating improvement or deterioration.

The strongest bear case can be summed up as: this is a "high-capex exposure to the Bitcoin price + an unfulfilled AI narrative," not a compounding business. If no substantive AI tenant lands within the next two or three years, and the Bitcoin price and network difficulty also fail to cooperate, then CleanSpark may repeatedly display a pattern: revenue grows, hashrate grows, assets grow, but the cash actually distributable to shareholders is small, and it may even keep borrowing or diluting. In that case, what shareholders bought is a more complex, more operationally risky leveraged substitute for spot Bitcoin, rather than a high-quality business.

Which facts would overturn my judgment? Put another way, which facts would make me admit I am being too conservative. First, if the company signs a verifiable, large-enough, long-enough AI/HPC customer contract and discloses contracted power, the rent/return framework, capex, and the collection cadence, I would re-rate the moat and valuation upward. Second, if over the next several quarters it proves it can keep generating positive Owner Earnings without relying on high coin-price fair-value gains, its quality score would improve markedly. Third, if the company stops large-scale dilution and puts per-share value growth ahead of scale growth, I would reassess management.

Which facts would make me admit I am wrong, or even that I should sell? If the following combination appears, I would be very cautious: AI business with no customer for a long time, Bitcoin holdings shrinking continuously without being exchanged for high-return projects, another round of large-scale equity issuance, a continued decline in network-hashrate share, rising per-unit power cost, or new regulatory/power-connection constraints. The worst permanent-loss scenario is that the company ultimately neither becomes a high-quality AI-infrastructure provider nor forms a sufficiently low-cost edge in mining, leaving only fast-depreciating specialized equipment and high-volatility digital-asset exposure.

Compared with other opportunities, my conclusion does not side with CLSK. MARA has larger Bitcoin reserves and higher hashrate; Riot has already disclosed data-center revenue and the AMD contract expansion; and for most balanced long-term investors, SPY offers a diversified claim on the earnings of 500 large U.S. companies, while the U.S. 10-year Treasury yield had already reached 4.67% on May 19, 2026 — meaning CLSK must offer a long-term certain return well above 4.67% to be worth tying up capital. By my current judgment, it cannot deliver "well above, with certainty."

If my portfolio could hold only 5 assets, CLSK does not currently qualify to be among them. Not because it lacks upside elasticity, but because its upside looks more like the return of "correctly betting on a theme" than the return of "buying an excellent business and compounding for the long term."

Investment Checklist and Final Verdict

Investment Checklist

Final Verdict

【Final Rating】 Watch

【One-sentence Investment Thesis】 CleanSpark is a strongly executing, fast-expanding Bitcoin miner, but it currently looks more like a combination of "high-volatility digital assets and an AI-campus option" than a long-term value stock driven by predictable cash flow.

【Core Bull Case】

• The company executes strongly on its own sites, power procurement, and hashrate expansion, completing an expansion from 45.6 EH/s to 50.0 EH/s and from 1,027 MW to 1.8 GW between the end of 2025 and April 2026.

• The buybacks are not entirely ill-judged: after the two convertible financings, it conducted large buybacks at about $12.33/share and $15.03/share respectively.

• If the AI/HPC campuses are successfully contracted and the power reserves convert into high-quality rental cash flow, the valuation ceiling would clearly move up. This upside is not fantasy but an option built on existing power/campus assets.

• The company is clearly focused on Bitcoin and does not diversify across multiple coins, making its strategy relatively pure.

【Core Bear Case】

• Mining is a commodity-type business with no pricing power, requiring continuous capital deployment to hold its competitive position.

• The flattering 2025 profit was driven mainly by Bitcoin fair-value gains and cannot be treated as steady-state earnings.

• The AI/HPC business had still not disclosed a first hyperscale customer by May 2026, so commercialization is under-validated.

• Historical dilution is heavy, and the company was still reshaping its capital structure through convertibles in 2026, with complex future dilution/refinancing.

• Both operations and statements are highly sensitive to the Bitcoin price, making it unsuitable as a "rest-easy long-term core" asset.

【Key Assumptions】 For the investment logic to hold, at least four conditions must be met: first, the company can convert a meaningful share of its 1.8 GW power reserve into AI/HPC contracts on a high-return basis; second, the mining business at least maintains industry-leading per-unit power and per-unit hashrate efficiency; third, it no longer relies on large-scale equity issuance in the future; fourth, the Bitcoin price and network-hashrate growth do not crush unit economics over the long run. The first two lean toward operating facts; the latter two lean toward external variables.

【Fair Buy Price】 I consider a more comfortable buy range to be $6–9/share. This range roughly corresponds to either the market assigning almost no value to the AI/HPC option, or the company having used subsequent data to disprove part of the operating risk while the share price has not yet fully reflected it. The current $14.69 is still a distance from the value-investing entry point I would endorse.

【Target Holding Period】 If held, you should allow at least 3–5 years to wait for AI/HPC commercialization to be validated; and on the 10-year-plus horizon you have set, I would rather treat it as a watch item than as a "ten-year core position" today. Because over a ten-year horizon what truly matters is business quality, not the size of a thematic opportunity.

【Expected Annualized Return】 This is based on my scenario model, not management guidance:

• Conservative scenario: if a decade from now it remains mainly a capital-intensive miner, the expected annualized return is roughly -16%/year.

• Neutral scenario: if mining stays competitive and AI delivers only partially, the expected annualized return is roughly +2%/year.

• Optimistic scenario: if the AI/HPC option is fulfilled smoothly and significantly lifts Owner Earnings, the expected annualized return could reach about +13%/year. These returns are very sensitive to future dilution; if the share count keeps expanding, actual returns will fall below the model. The base price uses the current $14.69.

【Maximum Loss Risk】 In the worst case, the loss could be very large — above 50% is not an exaggeration; if "weak coin price + intensifying network-hashrate competition + AI customers persistently failing to land + continued refinancing/dilution" occurs, the extreme case could even approach a loss of most of the principal. Not because the company is bound to go bankrupt, but because the market may ultimately be willing to price it only as a low-quality mining asset rather than an AI-infrastructure story.

【Tracking Metrics】 What is most worth tracking going forward is not the share price but these facts:

• The number, power scale, contract term, and return rate of signed AI/HPC customers.

• Operating hashrate, average hashrate, per-unit efficiency (J/Th), and changes in network-hashrate share.

• Quarterly Bitcoin output, holdings changes, and the share of Bitcoin pledged/receivable.

• Cash costs, especially per-unit power cost and per-unit output cost.

• The relationship between depreciation and amortization and new rig/infrastructure investment.

• Whether there is another large-scale equity issuance or new high-value convertibles.

• Whether reported net income and my defined Owner Earnings keep diverging.

• Whether data-center revenue starts to become a verifiable line in the financials.

• Whether the sole pool customer, key suppliers, and utility relationships change.

• Whether the regulatory, tariff, and power-connection/permitting environment deteriorates.

【Signals That Trigger a Reassessment】 Any one of the following would make me reassess immediately:

• Disclosure of a first substantive hyperscale AI/HPC customer contract.

• Positive Owner Earnings, net of maintenance capex, across multiple consecutive quarters.

• Another clear dilutive equity issuance.

• A stall in power-reserve expansion or obstacles to power connection/permitting.

• Bitcoin holdings shrinking without being exchanged for provably high-return projects.

• Depreciation continuing to surge while unit economics fail to improve.

• Another major change in management or a clear shift in capital-allocation policy.

【Final Recommendation】 If you insist on the framework of a "long-term business owner" rather than that of "catching the next leg of the Bitcoin-and-AI narrative," then my recommendation on CLSK is: keep watching, and do not rush to buy. Wait for a lower price, or for stronger operating evidence. A truly reassuring investment is not "it could rise a lot" but "even if the market closed for years, you still know clearly what you hold and how it keeps creating cash for you." On today's CleanSpark, that certainty is not yet enough.

【Open Questions and Limitations】 This report has tried its best to use the company's 10-K, 10-Q, proxy statement, investor-relations announcements, and official market/Treasury data; but two points still need your re-verification before you actually place an order: first, the latest precise figures for the full share count and book net assets in the first half of FY2026; second, the specific return model, customer-signing progress, and capex budget for the AI/HPC project. For a company like CleanSpark, these two are not minor details but core variables that determine the upper bound of valuation.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.