The $180 Million Power Bill: Reading the Poolin Texas Auction Through Its Payment Structure

Hook: A Number That Is Not the Price

On September 30, the United States Bankruptcy Court for the District of New Jersey cleared the way for a renewed auction of Poolin's Texas mining facility. The reported structure is $180 million. Four out of five headlines will stop there. The number that actually matters is $80 million — the portion of the consideration that is contingent, milestone-linked, and, by explicit contract language, triggered only by leasing or transferring the facility to artificial intelligence, high-performance computing, or cloud data center use.

That clause is the anomaly. A mining facility is being sold under a payment schedule that financially penalizes the buyer if the site continues to mine. The milestone does not merely reward AI conversion. It excludes mining facilities from counting toward it. Same power rights. Same interconnection queue position. Same dirt in Texas. Two valuations, and the market has now written the gap into a bankruptcy filing.

Tracing the capital flow back to its genesis block, this transaction is not a mining acquisition at all. It is a repricing event for a single input: the right to draw electricity from a grid at a known capacity and a known price. Everything else — the ASICs, the containers, the immersion tanks, the substation furniture — is being valued at or below zero.

That is the finding. The rest of this article is the evidence chain.

Context: What Poolin Was, and What It Left Behind

Poolin was, for a period between 2018 and 2020, one of the four or five largest Bitcoin mining pools in the world by reported hashrate. It was founded in China and grew alongside the industrial mining boom that followed the 2017 bull market. At its peak, it was processing a meaningful share of global Bitcoin blocks. The pool was a pure-play mining operation: it aggregated hashpower from third-party miners, distributed rewards, and monetized the spread. It did not, for most of its history, operate a large portfolio of proprietary hashrate on its own balance sheet.

The distinction matters. A mining pool is a software and settlement business. A mining farm is a capital-intensive power and hardware business. Poolin gradually accumulated the latter. By the early 2020s, it was operating or co-investing in physical mining facilities, including in North America, where cheap stranded power, favorable regulatory treatment, and proximity to institutional capital made Texas the default destination for Chinese-founded mining capital seeking a US footprint.

Then the sequence of shocks arrived in order. China's 2021 mining ban forced the physical displacement of an enormous share of global hashrate, most of it into the United States. The 2022 bear market compressed miner margins. The Terra collapse and the subsequent contagion removed the credit lines that had financed expansion. Poolin's own liquidity situation deteriorated to the point where it suspended withdrawals on its internal wallet product — a tell that the operating business was funding something other than operations.

The endgame was formal. On July 22, Poolin filed for Chapter 11 protection. That filing is the legal origin of the asset now under auction. Chapter 11 is a reorganization statute, not a liquidation statute. It allows a debtor to keep operating under court supervision while restructuring obligations. When a Chapter 11 estate decides it cannot reorganize a particular asset, it sells that asset under Section 363, and the sale is governed by the bankruptcy court's procedural rules rather than by ordinary commercial negotiation.

The Texas facility is that asset. What the estate is selling is not a company. It is a site, an interconnection, a power purchase arrangement, and whatever hardware happens to sit on the pad. The buyer is Hut 8, a NASDAQ-listed Bitcoin miner that has spent the last several quarters repositioning itself as an energy and compute infrastructure company. Hut 8's affiliate holds the status of stalking horse bidder.

Stalking horse is a term worth defining precisely. In a bankruptcy auction, the stalking horse is the bidder who negotiates a purchase agreement before the auction and sets the floor. The stalking horse typically receives protections — break-up fees, expense reimbursement — in exchange for having done the diligence and provided the anchor price. Other bidders then compete above that floor. The mechanism exists because bankruptcy estates need a credible opening bid to attract competition, and no rational buyer does the work without protection.

The sequence of events now on the record is this. In early September, a $140 million transaction was selected. Then the process was contested. Then, on September 30, the court approved a restart of the auction. The current structure carries a $180 million ceiling, with $100 million payable at closing and $80 million tied to milestones.

The price moved up 28.6 percent in under a month. That is the second anomaly. Restarting an auction after a selection has been made is a procedural admission that something in the original process was contested — either the price, the process, or both. I have sat through enough distressed processes to know that courts do not reopen clean auctions. They reopen auctions when a creditor committee, a competing bidder, or the US Trustee raises a colorable objection.

The bid deadline is October 1. The auction is October 2. Whatever follows will be on the record.

Core: The Payment Architecture Is the Analysis

Let me dissect the consideration structure directly, because it is the only part of this event that carries original information.

The total is $180 million. Of that, $100 million is cash at closing. The remaining $80 million — 44.4 percent of the headline — is contingent. The contingency is not generic. It is not "future hashrate" or "future Bitcoin price" or "future revenue." It is specifically tied to the leasing or ownership transfer of the site for AI, HPC, or cloud data center purposes. And the language expressly excludes mining facilities from satisfying the condition.

I want to be careful about what can and cannot be concluded from a term sheet. I cannot conclude the facility has signed AI tenants. I cannot conclude the power contract terms are favorable, because they are not disclosed. I cannot conclude the site has the fiber, the cooling capacity, or the substation headroom to host GPU clusters. What I can conclude is narrower and more important: the contracting parties have agreed that the site is worth more as an AI data center than as a Bitcoin mine, and they have written that agreement into the payment schedule.

This is an earn-out. In M&A, an earn-out is a mechanism that defers part of the purchase price and conditions it on future performance. The buyer likes it because it caps downside. The seller accepts it because it captures upside. In this case, the earn-out does something unusual: it does not merely defer value. It redirects the buyer's incentives. If Hut 8 keeps mining, it forfeits $80 million. If Hut 8 converts the site to AI hosting, it recovers $80 million. The contract is a lever aimed at the buyer's operating decision.

That is the first evidence point. The payment structure is not a valuation of the asset. It is an instruction to the buyer about what the asset is for.

The second evidence point is the mining hardware. The milestone excludes mining facilities. Read that carefully. The exclusion does not say "mining revenue does not count." It says mining facilities do not count. The physical plant — the containers, the racks, the ASIC fleet, the supporting infrastructure — is carved out of the value-capture definition. If those assets had positive residual value, a rational buyer would want them to count, because they generate cash while the AI conversion is being planned. The fact that they are excluded implies the parties treat them as a liability or a rounding error.

This is consistent with the broader state of the ASIC market. Post-halving, older-generation machines — the S17, S19, and their contemporaries — have compressed margins to the point where many operate at or below the marginal cost of power in most jurisdictions. In a site with premium power, running old hardware is not just unprofitable. It is an opportunity cost, because every megawatt-hour consumed by an S19 is a megawatt-hour not sold to an AI tenant at a higher rate.

The third evidence point is the 28.6 percent price increase. Between the early September selection at $140 million and the current $180 million ceiling, the market repriced the asset upward by roughly $40 million. There are two explanations, and they are not mutually exclusive. Either a competing bidder entered and forced the number up, or the original process was contested on price and the estate was forced to test the market again. Both explanations point the same direction: the asset was initially underpriced, and the correction came from the AI-usage valuation, not the mining valuation.

The fourth evidence point is Hut 8's position. Hut 8 is a listed company. It reports to the SEC. Its balance sheet is public. Its strategy is public. Over the last several reporting periods, Hut 8 has been explicit about shifting its identity from a pure Bitcoin miner to an energy infrastructure and compute company. This auction is a live test of that strategy. If Hut 8 converts the site and triggers the milestone, it validates the thesis. If it cannot, the $80 million never materializes, and the effective purchase price reverts toward $100 million.

That last point is the crux. The headline valuation is $180 million. The floor valuation is $100 million. The gap is a bet on execution.

Let me quantify the asymmetry. If Hut 8 executes the AI conversion, it pays $180 million and owns a facility whose value is supported by AI hosting economics. If it fails, it pays $100 million and owns a mining site in a post-halving margin environment. The difference between those two outcomes is not $80 million of cash flow. It is the difference between two entirely different asset classes. One is a data center. One is a stranded industrial site.

This is why I care about the payment structure more than the headline. The structure is the only place where the two valuations are separated and made legible.

Why Power Rights Are the Only Scarce Input

Step back from the specific transaction. The general pattern is now well established in the North American mining sector. The scarce input in both Bitcoin mining and AI compute is not silicon. It is firm, dispatchable, reasonably priced electricity delivered to a site with an interconnection agreement.

In mining, that input is monetized by running ASICs. The revenue is Bitcoin-denominated, the cost is dollar-denominated, and the margin is a function of the hashprice — the revenue per unit of hashrate per day. Post-halving, the hashprice has compressed structurally. The block subsidy was cut in half in April 2024. Transaction fees have not compensated at scale. The result is that the same megawatt produces less revenue than it did a year ago, while the cost of that megawatt has not fallen.

In AI hosting, the same input is monetized by renting rack space and power density to tenants who need GPUs. The revenue is dollar-denominated, contracted, and typically locked for multi-year terms. The tenant is a creditworthy counterparty — a hyperscaler, a neocloud, or an enterprise. The margin is a function of power cost, cooling efficiency, and lease rate.

The comparison is not subtle. A megawatt sold to an AI tenant can command several times the revenue of a megawatt spent mining, depending on the site, the lease terms, and the power price. The reason the AI multiple is higher is not that AI is fashionable. It is that AI tenants pay in dollars under contract, while miners pay in Bitcoin under volatility.

So when a bankruptcy estate sells a Texas site, it is selling a power option. The option has two exercise modes: mine or host. The market has now priced the hosting mode substantially higher. The milestone clause is that pricing, expressed as a contract term.

The 44 Percent Signal

Let me isolate the number that should be in every headline and will not be.

Eighty million dollars is 44.4 percent of $180 million. Nearly half the consideration is contingent. In a normal asset sale, a contingent component of that size would be unusual. It would signal deep uncertainty about the asset's value. Here, the uncertainty is not about the asset's existence or its physical condition. It is about whether the AI conversion can actually be executed — whether leases can be signed, whether the site can be retrofitted, whether the grid can deliver what the AI workload requires.

The earn-out is a risk-transfer device. The seller accepts a lower certain price today in exchange for upside if the buyer succeeds. The buyer accepts a higher contingent price to reduce its downside if it fails. The structure is rational on both sides. But it also tells you what the parties actually believe. If they believed the AI conversion was near-certain, they would not need an $80 million earn-out. If they believed it was near-impossible, they would not bother structuring one.

The 44 percent number is a confidence interval written in dollars. It says: the conversion is plausible, not probable, and the parties disagree on the odds.

What the Auction Restart Actually Signals

A bankruptcy auction restart is not a neutral event. It is a procedural remedy for a defect. When a court reopens a sale process, it is responding to a credible objection. The most common objections are price inadequacy, process irregularity, or both.

Price inadequacy is the more likely explanation here, given the 28.6 percent jump. If the original $140 million figure was challenged as too low — presumably by creditors who believed the AI-usage valuation supported a higher number — then reopening the process is the estate's way of testing that claim. The $180 million ceiling is the market's partial answer. The bid deadline and the auction will provide the rest.

There is a second-order signal here that is easy to miss. Restarting a process costs time. In a Chapter 11 estate, time is a cost because the estate is burning administrative expenses and the creditors are accruing interest. An estate does not spend that time unless it believes the expected recovery is higher than the cost of delay. The restart is therefore a vote of confidence — by the estate, and presumably by the creditor committee — that the asset is worth more than the first offer.

That is the fifth evidence point. The restart itself is a price signal, independent of the final number.

The Cross-Border Dimension

Poolin is a Chinese-founded entity. Its assets are in the United States. Its bankruptcy is being administered in New Jersey. That combination carries structural implications that the news coverage will skip.

Cross-border bankruptcies are procedurally complex. Creditors in different jurisdictions have different claims and different priorities. Currency controls in one jurisdiction can impede distributions in another. Recognition of the US proceeding by a foreign court is not automatic. The estate's ability to move value across borders is constrained by rules that have nothing to do with crypto.

None of this is disclosed in the source material, and I will not speculate on the specifics. But the pattern is well documented from prior mining bankruptcies. When a Chinese-founded miner liquidates US assets, the process tends to be slower and the recovery tends to be lower than a comparable domestic case. The Texas auction is not just an asset sale. It is a test of how efficiently the US bankruptcy system can dispose of Chinese-origin mining infrastructure.

Contrarian: The Narrative That Mining Is Dead Is Premature

The consensus reading of this transaction will be that Bitcoin mining is being repriced downward and AI is repricing upward. That reading is directionally correct and tactically dangerous. Let me separate the two.

The direction is real. Power rights allocated to AI hosting command a higher multiple than power rights allocated to mining. The milestone clause proves it. The 28.6 percent price increase supports it. The exclusion of mining facilities from the milestone reinforces it. I am not disputing any of that.

What I am disputing is the inference that this makes mining assets worthless. That inference confuses the value of a power option with the value of its current exercise mode. A mining site is not a failed data center. It is an unexercised power option. The option has value because the power rights have value. The hardware is the depreciating part. The interconnection is the appreciating part.

The correct frame is this. A site with firm power, a good interconnection, and a substation with headroom is worth roughly the same whether it is currently mining or currently hosting, minus the cost of conversion. The current mining use is a placeholder. The conversion cost is the spread between the two valuations. The 28.6 percent price increase is a rough estimate of that spread, as priced by the auction.

There is a second contrarian point, and it concerns causation. The news cycle will attribute this transaction to the AI boom. The attribution is partly wrong. The transaction is happening because a bankruptcy estate needs to sell an asset, and the estate will accept the highest bid available. The AI boom is what makes the highest bid higher. But the sale would be happening regardless. Poolin's bankruptcy is the cause. The AI repricing is the price.

Confusing the two leads to bad forecasts. If you believe the sale is caused by AI demand, you will expect more mining sites to convert. If you believe the sale is caused by bankruptcy, you will expect conversions only where the estate is forced to sell and the buyer has the capital to convert. The second frame is more accurate and less exciting, which is usually a sign it is correct.

A third point. The milestone excludes mining facilities, but it does not exclude mining revenue from the site's cash flow during the transition. The site can presumably keep mining while the AI conversion is planned and permitted. That interim cash flow is not captured in the headline number, and it may be material. A 100-megawatt site mining for twelve months while the conversion is engineered is not a stranded asset. It is a bridge.

Finally, I want to flag the execution risk that the headline hides. The AI data center conversion is not a software upgrade. It requires cooling systems rated for GPU density, which is a different engineering problem from ASIC cooling. It requires fiber capacity that a mining site may not have. It requires tenants, and tenants require leases, and leases require creditworthy counterparties willing to commit for years. Each of those is a failure point. The earn-out structure exists because the buyer knows this.

The data does not lie, only the narrative does. The data here says the asset was repriced upward by 28.6 percent on the strength of an AI-linked milestone. It does not say the conversion will succeed. It does not say mining is dead. It says the market is willing to pay for the option, and it is unwilling to pay for the current exercise mode.

I have seen this pattern before. In 2017, I spent twelve weeks auditing ICO whitepapers against their on-chain deployments. The pattern then was that the narrative ran months ahead of the code. The pattern here is the same shape. The AI narrative is running ahead of the leases. The difference is that here, the narrative is anchored to a real asset with a real power contract, and the payment structure forces the buyer to deliver on the narrative or forfeit 44 percent of the consideration.

That is a better structure than 2017 had. It is not a guarantee.

The Ledger View: What the Auction Actually Tells Us About the Sector

Let me zoom out and place this transaction in the larger map of the mining sector's current cycle.

Bitcoin mining is in a consolidation phase. The April 2024 halving compressed block subsidies. Transaction fees have not filled the gap. The marginal miner is unprofitable at current hashprice. The result is a shakeout in which weaker operators sell assets, and stronger operators with low power costs or diversified revenue acquire them. Poolin's bankruptcy is a data point in that shakeout. It is not the first and will not be the last.

The consolidation is happening at the same time as a capital inflow into AI infrastructure. Data center REITs, neoclouds, and hyperscalers are competing for power. In markets where power is constrained — and Texas is increasingly constrained — the competition for interconnection capacity is intensifying. Mining sites with existing interconnections are therefore attractive to AI capital, because they bypass the queue.

That combination — mining consolidation plus AI power demand — produces the repricing we are observing. The repricing is not uniform. It applies most strongly to sites with three characteristics: firm power, large capacity, and proximity to fiber. Sites without those characteristics will not reprice, regardless of the AI narrative. The distinction between "real power rights" and "concept packaging" is the most important analytical filter for the next six months.

Yields are temporary; the ledger remains eternal. The yield here is the AI hosting rate. The ledger is the power contract. When the AI cycle turns — and it will turn — the sites with real power rights will still have value, and the sites with narrative will not.

What to Watch, and Why It Matters

The next data points are scheduled and public.

October 1 is the bid deadline. October 2 is the auction. The final price and the identity of the winning bidder will resolve the near-term uncertainty. A price above $180 million would confirm the AI-usage valuation and likely trigger a revaluation of comparable mining assets. A failed auction would do the opposite, and would raise questions about the estate's ability to realize value.

Beyond the auction, the metric that matters is not the transaction price. It is Hut 8's lease signings. The $80 million milestone is triggered by leasing or ownership transfer for AI, HPC, or cloud data center use. That means the milestone is realized not when the deal closes, but when a tenant signs. Every quarter that passes without a lease is a quarter in which the earn-out remains contingent. The correct tracking variable is therefore the lease pipeline, not the acquisition price.

I have built trackers for this kind of thing before. In 2020, during DeFi Summer, I ran a Python scraper across more than 100 liquidity pools, aggregating APY, TVL, and unlock schedules daily. The lesson from that exercise was that headline yields were noise and emission schedules were signal. The same principle applies here. The headline price is noise. The milestone trigger is signal. Follow the trigger, not the number.

A third variable is the bidder composition. If the final bidder list includes pure AI or data center capital — not just miners — that is evidence the repricing is broader than the mining sector. If the bidders are all miners, the repricing is internal to mining, and the AI premium is being captured by a narrow set of operators. The distinction matters for how the trend propagates.

A fourth variable is Bitcoin price. A sharp decline in Bitcoin would compress mining margins further, increase the pressure on miners to sell or convert, and accelerate the trend this transaction represents. A sharp rise would relieve that pressure and slow the conversion wave. Neither outcome changes the direction. Both change the speed.

Takeaway: The Option Is the Asset

Strip away the headlines and the transaction is legible. A bankrupt mining company is selling a Texas site. The buyer is a listed miner pivoting to energy infrastructure. The payment structure splits the consideration into a certain $100 million and a contingent $80 million. The contingency is tied to AI conversion and explicitly excludes mining. The price rose 28.6 percent between the first selection and the current ceiling.

Read together, those facts say one thing. The market is pricing the power rights, not the mining. The mining hardware is a depreciating placeholder. The interconnection is the asset. The AI conversion is the option.

That is not a story about Bitcoin's decline. It is a story about the industrialization of compute. The same megawatt that was worth one thing in 2021 is worth another thing in 2025, and the difference is not sentiment. It is the counterparty. A miner pays in Bitcoin under volatility. An AI tenant pays in dollars under contract. The ledger does not care which is more exciting. It only records which is more creditworthy.

Due diligence is the only alpha that compounds. The diligence here is not about the auction price. It is about the lease. Watch the lease.

And when the October 2 auction closes, note what the final number is. Then note what the milestone clause excludes. The gap between those two facts is the most honest valuation of Bitcoin mining hardware that the market has produced this cycle. It is not zero. It is not negative. But it is not where the value is.

The silence between the blocks reveals the true intent. The intent here is written in the contract, and the contract says: mine if you must, host if you can, and we will pay you $80 million to prefer the latter.

That is the signal. Everything else is noise.