The coupon came in at 7.875%. That number is the story — not the $2.276 billion headline, not the 2031 maturity, not the press release.
Nobody hands a triple-B credit 7.875%. That is the price of speculative-grade paper dressed in the vocabulary of infrastructure. Read the deal through that lens and you stop seeing a bitcoin miner raising capital. You see a power company borrowing against concrete it has not finished pouring, and a datacenter operator that has not yet signed a tenant. Two years from now, whichever of those identities survives will justify the coupon. Until then, the coupon is the only verified fact in the room.
I have audited exchange proxy contracts line by line, and I have run automated rebalancing logic through 4,200 executions in ninety days. Both disciplines taught me the same primitive: read the obligation before you read the pitch. The obligation here is fixed and compounding. The pitch is elastic and unverified.
Ledgers do not lie, but liquidity always flees.
So let's audit the obligation.
CleanSpark trades as CLSK on Nasdaq. It runs proof-of-work infrastructure — powered land, large-scale datacenter shells, ASIC fleets — and its flagship development sits in Sandersville, Georgia. The company has spoken in hundreds of megawatts when describing its power footprint. That is the asset base. Not a protocol. Not a token. Not a consensus mechanism. Electricity access, switchgear, transformers, cooling loops, and the legal right to draw a very large amount of current out of a grid that is already congested.
That matters because the industry changed shape in 2024. Post-halving, block subsidies compressed, and the marginal miner's economics tightened to the point where hash rate alone stopped being a valuation anchor. At the same time, AI training demand detonated. Hyperscalers began signing decade-long power contracts. And a handful of miners figured out that the thing they actually own — an energized industrial site with grid interconnection — is the same thing an HPC operator needs, minus the engineering.
That realization became a sector-wide narrative. MARA moved. RIOT moved. IREN moved. Core Scientific signed an actual long-duration compute contract with CoreWeave, which converted a story into a backlog. And CleanSpark, sitting on powered land in a state where the Atlanta metro datacenter market is one of the fastest-growing in the United States, decided to fund its transition with debt rather than dilution.
The instrument: senior secured notes. Seven point eight seven five percent. Two thousand seven hundred sixty million dollars. Due 2031. Registered with the SEC. Filed on EDGAR. That registration is not incidental — it is the compliance backbone of the whole structure, and it gives CleanSpark something an anonymous protocol can never manufacture: a documented, audited, legally accountable capital-markets footprint.
Now the arithmetic. At 7.875% on $2.276 billion, CleanSpark has taken on roughly $179 million in annual interest expense — a fixed, non-negotiable cash obligation that accrues whether or not Sandersville converts a single workload to HPC.
The revenue side of that ledger is not fixed. Bitcoin mining revenue swings with spot price, network difficulty, transaction fees, and the thermal efficiency of whatever ASIC generation is deployed at any given moment. HPC revenue, if it materializes, would be contractual and stable — but it does not exist yet in any disclosed form. So the structure is a fixed liability bolted to a variable, cyclical income stream, with an unbuilt second income stream expected to close the gap. That is not fraud. That is a duration mismatch, and duration mismatches are how credit events are manufactured.
Here is where my audit background changes how I read the language. Six weeks inside the 0x v1 contracts taught me that the vulnerability is almost never in what the documentation says. It is in the gap between what the code does and what the prose implies. The same discipline applies to filings. CleanSpark's disclosure says the Sandersville infrastructure can serve HPC workloads. Look at the verb. Can serve. Not has contracted. Not has signed. Not has delivered.
In the audit, we find the truth that price hides.
The engineers in the room know why that distinction is load-bearing. A bitcoin mining facility and an AI datacenter are not the same building with different occupants. They diverge on three axes. Networking: miners tolerate latent, low-redundancy connectivity; AI clusters demand non-blocking fabric with strict east-west bandwidth guarantees. Cooling: ASICs run hot and loud but tolerate wide thermal envelopes; GPU racks at high density require liquid cooling, and liquid cooling requires a retrofit that touches the floor plan, the riser design, and the power distribution units. Uptime: a mining site that drops offline for an hour loses an hour of hashrate; an HPC tenant with a training run in flight has a contractual service-level agreement, and a violation has a dollar figure attached to it.
None of those gaps are fatal. All of them cost time and capital. And time, in a leveraged structure, is the most expensive input on the sheet.
So let's separate the asset from the aspiration. The asset — hundreds of megawatts of energized, interconnected industrial land — is genuinely scarce. Grid interconnection queues in major North American markets run years deep. A site that already holds capacity is worth a premium because it skips the queue. That is real. That is defensible. That is what the 7.875% is ultimately pricing.
The aspiration is everything downstream of the electrical meter. No disclosed HPC tenant. No disclosed contract duration. No disclosed power usage effectiveness figures. No disclosed liquid-cooling retrofit timeline. No disclosed network redundancy tier. Against competitors who have already published signed compute agreements, CleanSpark is narratively ahead and commercially behind. Exit liquidity is a courtesy, not a right — and so is a signed tenant.
Now the contrarian read, because the consensus interpretation of this deal is the one I trust least.
The consensus says: CleanSpark just funded an AI pivot without diluting shareholders. Bullish. Debt is cheaper than equity when the equity is trading on a forward story.
That reading is correct about the mechanics and wrong about the risk. Avoiding dilution is not the same as avoiding loss. Senior secured notes sit ahead of equity in the capital structure. If Sandersville underperforms, the noteholders have a priority claim on the pledged assets, and common shareholders absorb the residual — which, in a downside scenario, can approach zero even while the business keeps operating. Debt financing does not eliminate risk. It relocates it onto the people who own the bottom of the stack.
The second thing the consensus misses is what the financing structure itself reveals about management's conviction. You do not borrow $2.276 billion at a speculative-grade coupon to hedge. You do it to commit. That scale implies management believes the AI-to-power thesis is structural, not cyclical — that energized industrial land in a growing datacenter corridor will keep appreciating as hyperscaler demand compounds.
I arrived at the same conclusion from a different door. In 2021, I held ten Bored Ape Yacht Club tokens, bought as liquid positions rather than collectibles. When the tape showed overheating in November, I liquidated everything inside seventy-two hours and took a 110% return. My peers called it disloyalty. I called it a rule. Profit-taking is arithmetic, not sentiment. The lesson transfers directly: when a sector re-rates on a story rather than on contracted cash flow, the re-rating itself becomes the position — and re-ratings reverse faster than they build.
We trade the code, not the culture.
And the culture right now is euphoric about miners becoming datacenter operators. The valuation framework has already migrated. A few years ago, analysts modeled hash rate, breakeven power cost, and production per share. Today they model megawatts of HPC-serviceable capacity and compare it against hyperscaler kilowatt pricing. That is a genuine structural shift in how the market prices these companies, and it is happening across the entire listed mining cohort simultaneously.
Which raises the question nobody is asking loudly enough: what happens to that framework when the first miner publicly fails to convert? The moment a single peer discloses an abandoned retrofit, a cancelled tenant negotiation, or a timeline slipping past twenty-four months, the sector's new narrative anchor snaps. Miners are highly correlated in drawdowns already. Add synchronized leverage and you get synchronized credit stress. The 2022 collapse taught me that correlation is not a theory — it is a schedule. I moved eighty percent of my book to stablecoins within four hours of the Terra unwind, documented the exact sequence, and published it as a protocol. Not because I predicted the bottom. Because I refuse to own a position whose exit I have not pre-defined.
Apply that standard here. Strategy is the bridge between chaos and profit, and there is no strategy without a pre-committed exit.
What actually needs to be verified before this credit deserves a constructive read? Four things, in order of weight. First, a named HPC counterparty — a signed lease, an 8-K, a duration. That is the single line item that converts narrative into backlog. Second, quarterly interest coverage from the 10-Q: at $179 million annually, coverage below a two-times multiple starts to matter. Third, construction disclosure on the Sandersville retrofit — liquid cooling, network topology, PUE targets, and a completion date that is not aspirational. Fourth, secondary market pricing on the notes themselves. If the paper starts yielding meaningfully above its coupon in the secondary market, the credit market is telling you something the equity market has not admitted yet.
The sideways tape matters here too. Chop is not dead time. Chop is positioning time. Price compression forces projects to reveal whether they are financed on fundamentals or on momentum, and it separates operators with contracted cash flow from operators with a deck. CleanSpark just answered half of that question by proving it can access institutional credit. The other half — whether the credit buys revenue or merely buys time — is still unwritten.
One more observation, and it is the one I keep coming back to. The largest miners are no longer mining companies. They are hybrid entities — part power developer, part datacenter landlord, part proof-of-work operator — and capital markets are slowly, unevenly repricing them as such. Trust the protocol, verify the exit. The protocol here is a registered debt instrument with a payment schedule. The exit is a tenant roster that has not been published.
Watch the Sandersville leasing announcements. Watch the coverage ratio. Watch the secondary yield on the notes. And watch whether the first peer to stumble takes the whole re-rating down with it.
I watched the ape sell; the code still audits.