The stalking horse was $52 million. Hut 8 paid $140 million.
That is a 2.69x premium over the floor bid for two shuttered West Texas data centers — Pyote and Tarbush — formerly operated by Poolin for bitcoin mining and hosting before the company wound them down. The public record contains no megawatt figure, no power purchase agreement terms, no equipment schedule, and no stated conversion budget. What it contains is a number, a buyer, and a September 29 sale hearing on a New Jersey bankruptcy docket.
I have watched enough distressed infrastructure auctions to read that multiple correctly. It is not a valuation. It is a bid for optionality, placed by a buyer that believes the marginal energized megawatt in ERCOT is becoming scarcer than the market currently admits.
Optionality does not pay interest. The $140 million does. That is the part of the story the tape is not pricing yet.
Context
Poolin's estate is not large, but its shape matters. Total liabilities run to roughly $173.1 million, and $163.7 million of that — about 93% — is unsecured notes owed to Poolin Wallet users. Those are not trade creditors or secured lenders with collateral claims. They are retail custody customers holding paper at the back of a Chapter 11 waterfall.
The auction drew more than one serious bidder. Among the competing offers was an entity designated by Fluidstack, an AI cloud infrastructure provider. That detail is the most informative line in the entire disclosure. It tells you who else wanted these sites, and more importantly, why. Fluidstack does not mine bitcoin. It rents compute. Its interest in idle Texas mining shells is a statement about what those shells are worth once you stop measuring them in hashes.
Hut 8 won anyway, at a price nearly three times the floor. The company's Q2 disclosure puts its total development pipeline at 8.7 gigawatts, up 300 megawatts quarter over quarter. Hut 8 is a listed operator with an existing mining fleet, existing hosting contracts, and an explicit public ambition to reposition as a hybrid compute infrastructure company rather than a pure miner.
So the transaction is straightforward to describe and difficult to underwrite. A listed operator is buying brownfield power at a premium, inside a bankruptcy, and intends to convert or redeploy it. Everything that matters sits in the execution, and almost none of it is public.
Core
Start with the number you do not have. No disclosed megawatts.
So bracket it. Brownfield Texas mining campuses built during the 2020 through 2022 buildout typically landed between 100 and 300 megawatts each, depending on substation capacity and land. Assume Pyote and Tarbush total somewhere between 200 and 400 megawatts combined.
$140 million divided by 400 megawatts is $350,000 per megawatt. $140 million divided by 300 megawatts is $467,000 per megawatt. $140 million divided by 200 megawatts is $700,000 per megawatt.
Now hold that against the alternative. Greenfield energization in ERCOT stopped being a construction problem years ago. It became a queue problem. Interconnection studies, transmission upgrades, curtailment exposure, and local opposition add years, not months, between site selection and first electron on a large load. In constrained pockets of the state, that timeline can run past three to five years.
That is what Hut 8 is actually buying: calendar time. Not land. Not transformers. Not steel. Calendar time, purchased at a discount to the cost of waiting.
The harder question is what a megawatt is worth in three years. There are two revenue curves, and they do not look alike.
The first is bitcoin mining. Revenue per megawatt is a function of hashprice, network difficulty, and fleet efficiency. Post-halving, that curve is compressed, fleet-dependent, and exposed to a single volatile asset price. It is a good business at scale. It is not a stable one.
The second is AI hosting. Revenue per megawatt is contractual, dollar-denominated, and tied to a counterparty's credit rather than to an asset's market price. Deals in this category — the Core Scientific and CoreWeave template being the obvious precedent — convert a mining site into something closer to a REIT.
The spread between those two curves is the entire trade. And the market has already voted. The presence of a Fluidstack-designated bidder in the auction is the vote. Someone who rents compute, not hashes, looked at two idle Texas mining campuses and decided they were worth bidding on. That is a compute bid, not a mining bid. Hut 8 then outbid it.
Which is bullish if you believe the AI curve holds. And it is a leveraged bet on a narrative if you do not.
Here is where the physical detail starts to bite. A bitcoin mining rack draws roughly 10 to 20 kilowatts and tolerates air cooling. An AI training rack draws 40 to 120 kilowatts and does not tolerate much of anything. Retrofitting from one to the other means new switchgear, new transformers, liquid cooling loops, coolant distribution units, higher floor loading, and fiber routes that are often the binding constraint rather than the power. None of it is exotic engineering. All of it is slow. Two to three quarters is optimistic for a single site. Two sites, sequenced, with permitting, is longer.
And these sites were shut down, not idled warm. Restarting a cold site is not the same as expanding a live one. Maintenance deferred during a wind-down does not restore itself.
There is one structural advantage the Poolin sites retain over a pure AI data center, and it deserves stating plainly. ERCOT is an energy-only market with violent wholesale pricing. A facility that can flex its load — curtailing mining when power prices spike, resuming when they collapse — monetizes volatility in a way a hyperscale AI tenant contractually cannot. Demand response revenue in Texas is real and has been material to mining margins for years. If Hut 8 preserves the ability to arbitrage between mining load, AI load, and grid curtailment on the same interconnection, it owns a rare optionality stack. Pattern recognition precedes profit realization, and the operators who survived 2022 were the ones who treated flexibility as a product rather than a fallback.
I spent two weeks in May 2022 reverse-engineering the UST stabilization mechanism, building a model to locate the exact liquidity buffer below which the system could not survive stress. The lesson I took from that work was not about Terra specifically. It was procedural: model the failure threshold before you model the upside. So apply it here. At what AI hosting rate, at what utilization, does $140 million plus conversion capex clear Hut 8's cost of capital? If you cannot answer that, you do not have a valuation. You have a headline.
The financing structure is the second unknown. The disclosed consideration is $140 million in cash and other consideration. If the non-cash portion is equity, the premium is paid in dilution. If it is debt or a convertible, it is paid in future claims on the balance sheet. Neither is disqualifying. Both need to appear in the Q3 filings before anyone calls this accretive. Risk is the price of admission — but only if you know which game you bought a ticket to.
Contrarian
The consensus read writes itself: miner pivots to AI infrastructure, buys scarce power, wins. That is the version circulating.
The contrarian read is about who set the price. In a bankruptcy auction, the clearing price is determined by the second-highest bidder. The second-highest bidder here was an AI cloud company. That means the asset's value was established by AI demand economics — not by mining economics — while the winning buyer carries a miner's balance sheet and, in practice, a miner's cost of capital.
That is a duration mismatch. Hut 8 is underwriting a long-duration, contract-backed infrastructure asset against a comp set that trades on near-term compute scarcity. If GPU rental rates keep compressing in the mid and lower tiers, the implied value per converted megawatt falls, and the premium paid at auction starts to look less like foresight and more like peak narrative. History repeats, but the signature changes. The 2021 cycle overvalued ASIC fleets. This cycle may be overvaluing brownfield megawatts.
The second blind spot is Poolin's creditors. $163.7 million of $173.1 million in liabilities is unsecured notes owed to wallet users — roughly 93%. There was no exploit here. No consensus failure, no reorg, no broken cryptography. A centralized ledger with a counterparty simply stopped being able to meet claims, and the holders of those claims now sit at the back of a court-supervised line. Proceeds from this sale will not make them whole. Verify the code, trust the ledger — and if you cannot see the ledger, what you are holding is not an asset. It is a claim on someone else's balance sheet.
The third blind spot is timing. Anyone treating this as a closed deal is mispricing it. The September 29 hearing is the gate. An overbid is possible. Creditor objections are possible. Cure costs, unpaid taxes, and environmental liabilities attached to the sites are possible and, in a wind-down estate, probable. Every one of those is a claim on the $140 million before it reaches the estate, and every one of them raises the effective purchase price.
Takeaway
Watch three things, in order.
The structure of the other consideration in Hut 8's Q3 filings. That tells you whether the premium was paid in cash, in dilution, or in debt.
The first named AI counterparty contract on either site. That is the moment an 8.7-gigawatt pipeline stops being a marketing number and starts being cash flow. No contract, no thesis.
And the dollar-per-megawatt clearing price at the next distressed Texas sale. That is the only comp that matters. If a similar campus clears above $500,000 per megawatt with an AI buyer in the room, Hut 8 bought cheap. If it clears below $300,000, Hut 8 bought the top.
The hearing is September 29. Silence before the volatility spike.
