Court documents hit the docket on September 24th. Two dead data centers. One buyer. And a number that should make every Poolin Wallet user physically ill.
Hut 8 won Pyote and Tarbush — Poolin's two Texas sites — for $140 million. Combined. The stalking horse, the floor that was supposed to anchor the whole auction, was Thor CALAP's $52 million. Hut 8 paid 2.7 times it.
Read that again. In a bankruptcy auction, where the entire point is extracting maximum value for creditors, the winning bid came in at nearly triple the baseline. That is not a rounding error. That is the market standing up and admitting it had been mispricing Texas power assets for years.
And yet.
Poolin Wallet users — the retail crowd who parked funds in a wallet and watched their withdrawals get frozen in 2022 — are owed roughly $163.7 million. Total Poolin debt sits around $173.1 million. The auction fetched $140 million. Run the numbers yourself. Before the lawyers take their cut, before the priority claims get paid, before the transaction costs eat the pot, the money does not cover the hole. The gap is about $33 million.
So a facility just sold for triple its floor, and the people it was built on top of are still going to lose money. That is the story. Not the $140 million. The gap.
Let me walk you through it.
CONTEXT: WHO GOT BOUGHT AND WHO DID THE BUYING
Poolin was never a small player. At its peak it was one of the largest Bitcoin mining pools on earth, a Chinese-origin operation that at various points touched double-digit percentages of global hashrate. But mining pools are a business with a fatal structural flaw: they sit between the miners and the block rewards, they hold customer funds in the interim, and when bitcoin's price collapses, the float becomes a liability instead of an asset. That is exactly what happened.
In 2022, withdrawals froze. Users could see their balances. They just could not move them. That is the tell I always look for — a balance that renders on screen but does not clear. It is the same pattern I documented during the FTX implosion, when wallet movements proved insolvency weeks before the mainstream outlets caught up. A frozen withdrawal is a liquidity mismatch with a marketing department.
By July, Poolin was in Chapter 11 in the United States, with two American affiliates dragged into the proceeding. The mining and hosting operations at Pyote and Tarbush had already gone dark. Not sold. Not repurposed. Just stopped. Concrete, transformers, and a spot in the queue.
On the other side, Hut 8. A Nasdaq and Toronto-listed miner with the ticker HUT. It merged with US Bitcoin Corp and has spent the last two years repositioning itself from a pure Bitcoin miner into something closer to a power and compute landlord. Its development pipeline reportedly sits near 8.7 gigawatts, and it added roughly 300 megawatts quarter over quarter. That is not a mining company anymore. That is an energy company that happens to have once mined bitcoin.
And Texas — specifically the ERCOT grid — is the battlefield. ERCOT runs its own interconnect, separate from the rest of the US grid, and it is where the vast majority of large-scale mining landed because power was cheap and the regulators were friendly. Friendly, but not infinite. The interconnection queue in ERCOT is long. Years long. That queue is the actual asset being fought over here, and I will get to why in a second.
The sale hearing is set for September 29th. That is the date that matters. That is the date the whole thing either clears or blows up.
CORE: THE MATH THAT BREAKS THE NARRATIVE
Start with the bid spread. Four buyers showed up, and the spread between them is the most informative data point in the entire filing.
Thor CALAP, the stalking horse, bid $52 million as a combined baseline. DigiPower X bid $36.5 million for Pyote alone. Pecos, working with Fluidstack — an AI infrastructure outfit — bid $100.5 million for Tarbush alone. Hut 8 swept both for $140 million.
Look at the ratio. One hundred and five million for a single site, Tarbush, from an AI-focused buyer. That is 72 percent of what Hut 8 paid for both sites combined. The AI side valued one asset at nearly three-quarters of the package price. That is the entire thesis in one line: the AI buyer was not valuing concrete and ASICs. It was valuing Tarbush's power position, and it valued it violently.
Now do the debt arithmetic, because this is where the hype and the human collide.
Poolin's total obligations run about $173.1 million. Of that, roughly $163.7 million — call it 94.6 percent — is Poolin Wallet user claims, unsecured. The remaining slice, about $9.4 million or 5.4 percent, is other debt. The winning bid is $140 million. On paper that is an 80.9 percent recovery against total debt.
On paper.
In reality, that 80 percent is fiction until you subtract administrative costs, professional fees, priority claims, and the transaction friction that always shows up in Chapter 11. Everything senior to the wallet users gets paid first. Wallet users are unsecured. They are last in line, standing behind a stack of claims that will happily eat the pot before anyone at the front desk sees a check. The headline recovery number is a ceiling, not a floor, and the actual distribution to unsecured creditors will very likely land far below it.
This is not speculation. This is how Chapter 11 works, and the data point I keep coming back to is the presence of Hut 8's non-cash consideration. The reported bid structure is not pure cash. When a buyer pays in something other than straight cash — assumed debt, asset swaps, revenue-sharing, environmental remediation obligations — the effective burden on the buyer is higher than the headline number, and the effective cash available for distribution can be smaller or structured over longer horizons. You cannot look at the $140 million and assume $140 million lands in the estate on day one.
t check.
Now the part nobody is pricing: the interconnection agreement.
A data center in Texas is not valuable because of the building. Steel is steel. Cooling is cooling. What is scarce — genuinely, structurally scarce — is a live position on the ERCOT interconnection queue with transferable capacity. Getting a new large load connected in Texas is a multi-year process. If you already hold the queue position, you hold something that cannot be manufactured on demand. Hut 8 did not buy two mines. It bought two live seats at a table with a years-long waiting list. Allocate that to the 8.7 GW pipeline story and the acquisition stops looking like a mining deal and starts looking like a land grab.
That is the AI premium. Fluidstack's bid on Tarbush tells you the premium is real and that an AI buyer was willing to pay $100.5 million for one site to capture it. Hut 8 paid $140 million for both, which, if you believe the Tarbush pricing, means it got Pyote for something in the neighborhood of $40 million. If the AI valuation of Tarbush is the honest number, Hut 8's package price looks almost sensible. If the AI valuation is froth, Hut 8 paid a strategic premium to win a bidding war.
The truth is probably both. And that is the risk embedded in the deal: Hut 8 is paying an AI-flavored price for assets whose near-term cash generation still comes from Bitcoin mining. Same physical shell. Two completely different revenue models. And only one of them has a customer contract signed.
The dual-use flexibility is the technical underbelly nobody talks about honestly. An ASIC mining hall and a GPU training cluster are not the same facility. AI workloads want higher power density per rack, different cooling — often liquid cooling at the high end — and dramatically more stable power than a bitcoin miner needs. Miners are famously flexible loads: they can curtail instantly when ERCOT prices spike, and that flexibility is a revenue stream in itself. AI training clusters are far less tolerant of curtailment. You cannot pause a multi-day training run every time the grid gets tight without torching your SLA.
So the conversion is not a flip of a switch. It is a capital project. Power distribution gets rebuilt. Cooling gets retrofitted. The curtailment economics that made these sites attractive as mines actually work against them as AI facilities. Poolin's Texas sites already went dark, which means Hut 8 inherits restart risk on top of conversion risk. Permits, environmental review, grid re-engagement, and the engineering reality of turning a mining hall into an HPC hall — none of that is disclosed, and all of it costs money that has not been spent yet.
Pump, dump, debug. Repeat.
I have audited enough infrastructure deals to know the pattern: the headline is the power capacity, the reality is the capex required to make that capacity productive, and the gap between them is where the disappointment lives. When I was digging through Solidity contracts during the 2017 ICO rush, the exact same tell showed up in code instead of concrete — a beautiful white paper, a clean front end, and a token contract that quietly routed value to the team. Here, the white paper is a press release about 8.7 GW, and the fine print is a retrofit bill nobody has quantified.
One more piece of the core, because it matters for the broader sector. This auction is a pricing event. Thor set the floor at $52 million. Hut 8 cleared it at $140 million. That is a 169 percent surprise against the anchor. In any asset market, a 169 percent overshoot of the reference price is called a re-rating. The market had been valuing distressed Texas power assets as mining assets. The auction just revalued them as AI infrastructure assets. That is the single most important line in the entire filing.
And that re-rating does not stay contained to Poolin. Every listed miner holding ERCOT capacity — Riot, Marathon, Core Scientific, and the rest — now has a fresh comparable. If a bankrupt, offline, two-site package cleared at $140 million, what is a live, operating, interconnected facility worth? The market will start asking, and the answer will flow straight into share prices before a single new AI contract gets signed.
That is the bullish read, and it is legitimately strong on the asset side.
The bearish read on the same data is that the AI premium is being capitalized into asset prices before it is being realized in revenue. A rent check is not a bid. A bid is a signal of demand. Demand is not the same as a signed lease with a hyperscaler paying monthly. Hut 8 has the assets and the narrative. It does not, as of the filing, have a disclosed long-term AI tenancy converting that 8.7 GW into contracted cash flow. Until it does, the premium is a bet, not a yield.
I want to be precise about that distinction, because the two get blurred constantly in bull markets. The AI data center buildout is real. Power is genuinely scarce. The interconnection queue is genuinely a bottleneck. All structural, all true. But the specific question for Hut 8 is narrower: can it convert this asset base into recurring revenue faster than the capital markets lose patience with the capex? That is an execution question, and execution questions do not get answered by auctions. They get answered by quarterly filings, and Hut 8's next few will be worth reading line by line rather than skimming the pipeline number.
CONTRARIAN: THE ASSET APPRECIATED. THE PEOPLE DID NOT.
The consensus take on this deal is simple and comfortable: miners are becoming AI landlords, Texas power is scarce, Hut 8 made a smart strategic buy, and the sector is re-rating upward. Every part of that is arguably true. Which is exactly why it should be distrusted.
The unreported angle is that asset appreciation and creditor recovery are not the same variable, and they never were. The auction makes this vivid. Poolin's assets got bid up nearly threefold over the floor. That is the best possible outcome for the assets. And the wallet users — 94.6 percent of the debt stack, the actual humans — still face a shortfall, because the appreciation went into the price the buyer paid, not into a surplus that flows to the bottom of the capital structure. In a solvent sale, appreciation benefits shareholders. In a bankruptcy, appreciation benefits the senior claims and the estate's professional fees first, and the unsecured ranks only see what is left.
You can watch the same divergence in the secondary market for distressed claims. Bankruptcy claims trade at deep discounts to face — typically in the range of 20 to 50 cents on the dollar — precisely because sophisticated buyers know that the headline recovery rarely survives contact with the priority waterfall. If the market thought wallet users were getting 80 cents, claims would not be trading at a fraction of that. The claim market is the honest number. The press release is not.
There is a second blind spot, and it is bigger than Poolin. If this deal is the template — distressed mining assets getting repriced as AI infrastructure — then the industry has an incentive to convert hashrate capacity into HPC capacity. Which means bitcoin's own physical security budget could quietly shrink as power migrates to workloads that pay more. That is not a 2026 problem. That is a 2027 problem. But the auction just took a step toward it, and nobody is modeling it.
The third blind spot is regulatory. Texas has been friendly to miners, but a mining load and an AI training load are not viewed the same way by grid operators. Miners are valued for their interruptibility — they absorb surplus power and drop off on command. AI facilities are the opposite: they demand firm, stable power. The moment enough ERCOT capacity converts from interruptible to firm, the grid's flexibility math changes, and the regulatory posture toward large loads changes with it. That is a policy risk sitting on top of every AI-conversion thesis in Texas, and it is invisible in the $140 million headline.
Gas fees higher than the yield. Typical. The yield on this deal is a narrative. The cost is concrete, and it is being paid now.
TAKEAWAY: WATCH THE 29TH, THEN WATCH THE LEASES
Two dates matter going forward. The first is September 29th — the sale hearing. If a losing bidder, most plausibly Pecos and Fluidstack given how aggressively they priced Tarbush, files an objection to the auction process, the deal can stall, reopen, or get re-cut. A high bid that loses is a motive, and a motive plus a courtroom equals delay.
The second, and the one that actually resolves the thesis, is Hut 8's first disclosure of a long-term AI or HPC tenancy with real contracted revenue behind it. Until that lands, the 8.7 GW pipeline is a promise, and the $140 million is a bet that the queue position converts before the capital runs out.
Watch the assets get repriced upward. Then watch whether the people underneath them ever got made whole. Those two lines are going to diverge, and the size of the gap between them is the real measure of what just happened in Texas.
