Seven point eight seven five percent. That number hit the tape before any of the language in the press release did. CleanSpark, ticker CLSK on Nasdaq, has closed $2.276 billion in notes at a 7.875% coupon, and the proceeds carry two stated destinations: build out a data center in Sandersville, Georgia, and repay equity capital already sunk into the project. Attached to that project is a 20-year lease.
Twenty-six years in this industry taught me to read contracts the way I read Solidity. The interesting part is never the function that gets called. It is the branch that does not. So here is the branch. A bitcoin miner does not sign a twenty-year lease. A bitcoin miner signs a three-to-five-year lease, because that is the economic depreciation life of the machine doing the work, and the halving cadence is four years. The fact that this one runs twenty tells you what the asset actually is before any executive says it out loud. That is the whole article. Everything below is arithmetic supporting it.
Context: what CleanSpark actually is
CleanSpark occupies the infrastructure layer of PoW hashrate. No protocol. No consensus mechanism. No cryptographic surface. The business is composed of four inputs: electricity priced in dollars per megawatt-hour, land, interconnection queue position with the regional grid operator, and ASIC fleet efficiency measured in joules per terahash. Everything else is financing.
That matters because the failure modes are financial, not technical. There is no reentrancy vulnerability in a substation. There is only a fixed obligation sitting against a variable revenue stream, which is the exact structure that killed the yield-farming complex in 2020 and the algorithmic stablecoin complex in 2022. I modeled Compound's APY decay curve before the liquidity crisis resolved, and reduced Terra-linked exposure six months before May 2022, using the same method in both cases: isolate the fixed term, stress the variable term, find the price at which the two cross. The method transfers cleanly to a leveraged miner.
The post-halving environment is the variable term. April 2024 cut the block subsidy to 3.125 BTC. Hashprice, revenue per petahash per second per day, has repeatedly tested levels where older fleets go cash-negative on the power line alone. The marginal cost of production for low-efficiency machines is now a printed number that gets compared against spot every week. CleanSpark's fleet is newer than the sector average, which buys margin and time. It does not buy twenty years of certainty.
Which brings the duration problem into focus. ASICs depreciate on a three-to-five-year curve. Power contracts typically run one to five years with renewal options. Bitcoin's issuance schedule resets every four years, and nobody can underwrite hashprice into 2044. A twenty-year lease sits outside every one of those clocks. It can only be supported by a use case whose revenue is not a function of BTC price. That means a compute tenant paying rent, not a mining operation paying itself.
Core: reading the coupon, the arithmetic, and the duration
The coupon is a credit rating in disguise. Duration-matched US investment-grade corporate paper has been clearing in the 5.0 to 5.5 percent area. Extremes reserved for unrated or distressed issuers sit well north of that. 7.875% lands roughly 250 to 300 basis points above the investment-grade line, squarely in high-yield single-B territory, the neighborhood occupied by mid-tier energy and infrastructure issuers with contracted cash flows and real execution risk. Read that as a sentence: the lenders are not underwriting a bitcoin miner. They are underwriting a fixed-obligation credit with a partially verified contract attached. Had the twenty-year lease been executed with a named hyperscaler, with credit support and escalators, the coupon would have tightened. The spread is the price of the missing annex.
The arithmetic is fixed and it is daily. $2.276 billion at 7.875% is approximately $179 million in annual cash interest, or about $490,000 per day, assuming the facility is fully drawn. That is the hurdle the new asset clears before it contributes a dollar to the equity. It is also a constant. The revenue servicing it is variable in exactly one respect: whether the lease carries escalators, and whether the tenant's credit is money-good. If the lease is flat and the tenant is an affiliate, the structure is a levered self-mining expansion wearing a data center costume. If the lease escalates and the tenant is investment-grade, it is closer to a contracted infrastructure yield. Those two worlds sit eight multiples apart, and the disclosure gap between them is currently the width of a footnote.
This is refinancing, not capital formation. The stated use of proceeds includes repaying prior equity capital. The project already existed on the balance sheet, funded by shareholders. The transaction converts a floating, permanent, dilution-prone funding source into a fixed, dated, covenant-bearing one. Three consequences follow, mechanically. No dilution today. Mandatory cash service every quarter until maturity. And the risk migrates from the equity holder to the creditor. The equity is no longer a claim on a growth project. It is a residual claim on a levered asset, which is mathematically a call option struck at the interest coverage line.
The twenty-year duration is the tell, and it is not subtle. Self-mining cannot support a twenty-year commitment. Hashprice is a competitive equilibrium, not a contract, and it has never held stable across a single halving epoch, let alone five. The issuance schedule is a four-year clock. ASIC economics cap at three to five years. Site-level power agreements renew on short cycles. A twenty-year lease therefore commits the site to a use that must survive at least four generations of silicon and one full halving cycle. That is only coherent if the use is not silicon-dependent. It points to colocation or a built-to-suit HPC lease. This is inference. But it is inference drawn from duration arithmetic rather than from the press release, and that distinction is material.
If that reading holds, the frame changes, not just the multiple. A bitcoin miner is valued on hashrate, fleet efficiency and power cost, a cyclical equity with a discount rate to match. A contracted data center is valued on capitalized rent plus residual optionality, an infrastructure asset with a lower discount rate. Moving a cash flow stream from the first bucket to the second is worth more than any hashrate expansion. That is the entire bull case. It rests on one document that nobody outside the deal has read.
I have seen this shape before, in different clothing. In 2017 I audited an ERC-20 contract line by line before mainnet launch. The vulnerability was not in the visible functions. It was an integer overflow that surfaced only when a specific input range was exercised, a branch nobody calls during testing but every attacker eventually finds. The team patched it, and roughly $12 million in ICO proceeds stayed where they belonged. The lesson generalizes: security is a property of the entire state space, not of the code path you demonstrated. The state space here includes tenant default, a covenant breach, a collateral dispute, and a refinancing window that may not open. None of those branches appear in the headline. All of them are priced into 7.875%.
Preemption discipline applies the same way. When I built the Compound short, the work was not forecasting a crash. It was modeling the decay curve and locating the date on which the incentive structure crossed from sustainable to reflexive. When I cut exposure to anything Terra-adjacent, the work was structural: the peg depended on a market maker of last resort representable as an extrapolation of reflexivity. In both cases the covering branch was disclosed in the code, not in the community updates. Same method here. The document to read is the note indenture, not the thread.
Where the tape and the filings diverge. Markets reprice on the announcement and reprice again on the disclosure. The announcement says AI infrastructure. The disclosure will say whether the thing is contracted or aspirational. Between those two events, the equity trades on a story and the notes trade on a covenant package. Those instruments will not stay correlated.
Contrarian: the equity was re-risked, not de-risked
The reflexive read is that financing closed, therefore the project is de-risked, therefore the equity re-rates. The correct read is that financing closed, therefore the project is de-risked and the equity is re-risked. The creditor received the contract. The shareholder received the residual above a $179 million annual cash hurdle. In a bear market, residuals get discounted aggressively and contracts get bid. That asymmetry is not a flaw in the structure. It is the structure.
A crowded narrative is not a catalyst. Miner pivots to AI is now a sector-wide positioning theme, and themes do not clear a coverage ratio. The re-rating, when it arrives, will be triggered by a name, a credit rating attached to that name, and an escalator schedule. Until those exist in a filing, the market is capitalizing a duration with no counterparty attached.
The tail case is priced, not ignored. If the twenty-year lease runs to an affiliate, a shell, or a single-tenant startup with no balance sheet, what remains is a $2.276 billion levered self-miner operating into post-halving hashprice. That scenario is not zero-probability. It is precisely why the coupon is 7.875% and not 5.5%. The desks ran the branch. Retail is reading the headline.
And liquidity depth has migrated. In 2021, exiting NFT positions across multiple OTC desks over three weeks taught me that the lesson was never that the assets were worthless. It was that the exit prices at whatever depth exists at the moment of need, not at the floor quoted on a screen. The tradeable depth of a CleanSpark position now runs through a covenant package rather than an order book. If coverage deteriorates, the equity gaps before the notes do. Anyone modeling exit liquidity off the prior range is modeling the wrong instrument.
Takeaway
Track the filing, not the tape. Five items resolve this: the tenant's identity and credit rating; whether the notes are secured against the site and its lease cash flows; the covenant ladder, specifically any maintenance test on interest coverage; escalator and renewal terms; and the maturity wall. A practical trigger: if adjusted EBITDA to cash interest prints below 1.5x on a quarterly basis, the tradable expression is the notes, not the equity. The notes have contract. The equity has amplification.
The forward question is not whether CleanSpark becomes a compute company. It is whether the signature on a twenty-year lease is worth 7.875%, and whether the people holding the residual understand they are now second in line. That is the balance sheet's immutable logic. It does not negotiate, and it does not read press releases.