Volatility is not risk. Interconnection queue position is risk.
That sentence has cost me more arguments than any other, and it has never been more load-bearing than in the week a two-paragraph release announced that Black Hills Energy and Google had signed a long-term power agreement to serve the company's data center campus in Cheyenne, Wyoming.
The release contained five claimable facts. A utility. A hyperscaler. A long-term power agreement. A campus location. The word "sustainable." It contained zero numbers. No megawatts. No contract term. No price per megawatt-hour. No fuel mix. No capital structure. No docket number at the Wyoming Public Service Commission. No indication whether the electrons are incremental or reallocated from the existing fleet.
I once spent a semester manually auditing forty-five ICO whitepapers and found that roughly eighty percent of them carried inflationary schedules that could not survive contact with their own vesting cliffs. I shorted them through P2P OTC desks and made fifteen percent while the market lost most of everything. That exercise taught me a specific discipline: the absence of disclosure is itself a disclosure. A document that refuses to state its parameters is telling you where the parameters are uncomfortable.
This press release is one of the least informative documents I have read in fifteen years of watching capital get allocated to infrastructure. It is also one of the most instructive. Because the thing it refuses to quantify โ dispatchable capacity, delivered on a schedule, at a contracted price โ is the exact thing the crypto market has spent a decade pretending it could synthesize out of code.
The market is pricing tokens. The physical world is pricing capacity. Those two curves have decoupled, and only one of them is going to clear.
The context nobody put in the press release
Start with the utility, because the utility is the whole story and nobody covers utilities.
Black Hills Energy is the regulated electric arm of Black Hills Corp, a multi-state utility holding company headquartered in Rapid City, South Dakota, with Wyoming operations centered in and around Cheyenne. Its earnings model is not complicated: rate base multiplied by an approved return on equity. Every dollar of capital expenditure that a commission allows into the rate base becomes a future revenue stream with a regulator-blessed margin attached. This is the most boring business model in America and the most durable.
Google has operated a data center in Cheyenne since roughly 2012. That campus is not a new entrant demanding a new wire. It is an existing, extremely large, extremely well-capitalized load that has been sitting on that grid for over a decade and is now expanding into the most power-hungry phase of the computing industry's history.
Wyoming's resource endowment matters here. The state sits on one of the better onshore wind regimes in the continental United States. It is also a coal state with a long history of resisting exactly the kind of federal clean-energy framework that would make wind procurement frictionless. Black Hills Energy, to my recollection, holds or contracts wind generation in the Cheyenne area โ I believe on the order of tens of megawatts from a project in the region โ but I could not verify the capacity, the offtake structure, or whether it is rate-based or contracted in the window I had. I flag that as a C-level inference, not a fact. If I am wrong about the specific asset, the structural argument below does not change.
Now widen the aperture, because a single utility contract is a data point and the data point is meaningless without the curve it sits on.
Lawrence Berkeley National Laboratory's December 2024 report on U.S. data center energy use put the 2023 figure at roughly 176 terawatt-hours, about 4.4 percent of national electricity consumption. Its projection for 2028 spans 325 to 580 terawatt-hours โ between 6.7 and 12 percent of the national total. The IEA's Electricity 2024 put global data center, crypto, and AI consumption at roughly 460 terawatt-hours in 2022, with a base case near 800 terawatt-hours by 2026 and a high case above 1,000.
Sit with that for a moment. The United States spent roughly twenty years with essentially flat electricity demand. Every utility capital allocation model, every integrated resource plan, every load forecast in the country was built on the assumption of zero growth. That assumption is now dead. Demand is re-entering a 1 to 2 percent annual growth channel โ the first structural inflection in a generation โ driven by three simultaneous forces: data centers, manufacturing reshoring, and transport and heating electrification.
This is the global liquidity map I actually watch. Not stablecoin supply. Not perp funding rates. Generation capacity, transmission headroom, and the regulatory calendars that gate both. Everything else is downstream of that.
The megawatt-hour is the last bearer asset that settles physically
Here is the framing I keep returning to, and it is the reason I think this press release deserves more attention than any token launch announced the same week.
Electricity is a bearer asset. Whoever holds the electron at the moment of delivery owns it. There is no clearinghouse, no custodian, no issuer, no freeze function. It cannot be forked, double-spent, or clawed back. It also cannot be stored at scale, which means it has negative carry in the physical sense โ generation that is not consumed in the interval is destroyed, not banked.
A power purchase agreement is the closest thing the physical world has to a settlement layer. It is a duration instrument: a fixed price, a fixed volume, a fixed delivery schedule, a credit counterparty, and a regulatory enforcement mechanism. A fifteen-year PPA with an investment-grade utility is a bond with a physical underlying. A fifteen-year token vesting schedule is a bond with a governance vote as its underlying.
Liquidity is merely trust, tokenized and flowing. The PPA market understood this a century before anyone wrote a smart contract. Trust here is not sentiment. It is the regulator's imprimatur, the utility's balance sheet, and the physics of a wire that cannot route around itself.
This is also why I find the DeFi lending markets so structurally unsatisfying, and why the Google contract clarifies the discomfort. Aave and Compound set their borrow rates on a utilization curve with a governance-chosen kink โ typically around 80 percent utilization, where the rate slope turns vertical. That kink is an administrative parameter. It is not discovered by supply and demand; it is voted on. It is a guess with a governance proposal behind it.
A utility tariff is also an administrative parameter. But it is a guess with a docket number, an intervenor process, an expert witness record, and a legally binding cost-of-service study behind it. Both are administered prices. Only one of them has an adversarial discovery mechanism. When the crypto market talks about "market-determined" rates, it usually means "determined by whichever parameter the largest delegate happened to set at 3 a.m. during a governance sprint."
I do not say this to be dismissive of DeFi. I say it because the comparison is the actual competitive analysis. The reason Black Hills can sign a fifteen-year commitment and a DeFi protocol cannot is not technology. It is the existence of a counterparty that can be compelled to perform.
Hourly matching is an oracle problem, and the oracle is stale
Google's stated commitment is 24/7 carbon-free energy by 2030 โ matching, hour by hour, every megawatt-hour of consumption with carbon-free generation on the same grid in the same interval. Not annual netting. Hourly.
That is a data attestation problem, and it is structurally identical to the problem every DeFi protocol has with a price oracle.
Consider what annual renewable energy certificate matching actually does. A company consumes power continuously across 8,760 hours. It buys certificates representing an annual volume of renewable generation. The certificates are retired against the annual total. On paper, the books balance. In physics, the company may have consumed gas-fired power at 7 p.m. on a still, dark January evening and wind power at 3 a.m. in April, and the certificate market does not care.
This is a daily close being used to settle an intraday option. Anyone who has watched a lending protocol get liquidated on a stale price feed understands the failure mode precisely. The number was correct when it was written. It was not correct when it was used.
In the absence of alpha, volatility is just noise. Annual matching is noise dressed as accounting. Hourly matching is the alpha โ and it requires timestamped, location-specific, verifiable generation data at a granularity the current certificate infrastructure was never built to provide. That is an oracle design problem, a data availability problem, and an incentive problem, in that order.
The press release says "sustainable energy solutions." That phrase is doing an enormous amount of work. It could mean new-build wind with hourly-matched attributes โ strong additionality, genuine new clean electrons entering the grid. It could mean existing grid power plus unbundled certificates โ weak additionality, no new generation, an accounting entry. The climate value of the contract differs by roughly an order of magnitude between those two readings, and the release declines to specify which one applies.
I have seen this pattern before. In 2017 I watched projects describe token distribution as "community-aligned" without publishing the unlock schedule. The vagueness was never accidental. Vague language appears at exactly the point where specificity would cost the issuer something.
The queue, not the chip, is the moat
Here is where the crypto-native analysis usually goes wrong, and where I want to plant a flag.
The standard narrative treats AI demand as bullish for crypto because of convergence โ decentralized compute, GPU marketplaces, DePIN networks renting out idle silicon. I have traded that narrative. In 2025 I built a framework correlating EU regulatory developments against AI training cost curves and identified a real convergence opportunity in decentralized GPU rendering that produced roughly twenty-two percent alpha against standard crypto indices. The trade was correct. The reasoning behind most of the copycats was not.
The binding constraint in AI infrastructure is not chips. Nvidia's allocation is tight, but allocation is a queue with a known length. The binding constraint is the interconnection queue, and the queue has no published delivery date.
Transformers. Switchgear. High-voltage direct current cable. Gas turbines. Every one of these categories has run into multi-year backlogs. Large power transformer lead times stretched to two to four years at the peak of the crunch. Interconnection queues in several U.S. regions have waiting periods measured in four to seven years. FERC Order 2023 was written specifically because the queue process had become a bottleneck that was itself a market failure.
A DePIN network that does not control interconnect rights is renting somebody else's moat.
This is the analysis I keep coming back to, and it comes directly out of 2020. That summer I built a Python scraper tracking Uniswap V2 pools across twelve major pairs, mapping roughly two hundred million dollars in TVL to test for systemic yield correlation. What the data showed me was that TVL is not a stock. It is a claim on liquidity that can exit in a single block, and the correlation structure of those exits was the actual risk, not the headline number. I cut leveraged farm exposure two weeks before the correction and avoided the liquidation cascade that took out most of my cohort.
The lesson transfers cleanly. TVL is a claim that can leave in twelve seconds. Megawatts are a claim that cannot arrive in less than four years. Those two durations are not comparable assets, and any framework that prices them on the same curve is wrong. The asymmetry runs the other way, too: once a hyperscaler locks a fifteen-year contract on a substation, that capacity is gone from the market for a generation. It does not come back when yields compress.
This is also why the Layer 2 debate bores me. The real difference between the OP Stack and the ZK Stack is not proving system architecture. It is which team can convince more projects to deploy chains first โ a distribution contest wearing a technical costume. The same is true here. The winner in AI-adjacent compute will not be whoever has the cleverest scheduling algorithm. It will be whoever holds the signed interconnect agreement. Distribution of scarce physical access beats elegance every cycle.
Hashprice is the only honest floating rate in this market
Bitcoin mining is the purest expression of the collision between crypto and the physical grid, and it is the part of this story most readers actually hold exposure to.
Since the April 2024 halving, the block subsidy is 3.125 BTC. Miners who signed fixed-price power contracts at pre-halving hashprice assumptions are now running a business where revenue per unit of hash has roughly halved while their largest cost line is contractually locked. That is a short-volatility position on hashprice financed with fixed-rate debt. It is exactly the trade that destroyed the leveraged yield farmers in 2020, wearing different clothes.
The most dangerous debt is the kind no one sees. In mining, it lives in hosting contracts. A hosting agreement with a fixed per-kilowatt-hour rate and a pass-through for curtailment penalties is a floating-rate obligation dressed as an operating expense. When hashprice compresses and the grid operator calls for curtailment during a heat event, the miner eats both the lost production and, in some structures, the penalty. That obligation never appears on a token dashboard. It appears on a bankruptcy filing.
The structural advantage in mining does not come from the newest ASIC generation. It comes from three things, in order: below-market power contracts with long duration, demand-response revenue that pays the miner to not produce during peak pricing, and balance sheet duration that lets the operator survive a twelve-month hashprice trough without selling the fleet.
Note what that list does not include: any protocol, any token, any on-chain yield. The bear market has been efficient about this. It has repriced the miners with real power contracts and real curtailment revenue far less severely than the ones with announced hashrate and no signed interconnection.
Wrapped capacity: the bridge problem, relocated
There is a growing category of tokenized real-world assets claiming to represent energy, power purchase agreements, or future generation capacity. I want to be precise about my skepticism here, because it is not reflexive.
Cross-chain bridges have absorbed more than two and a half billion dollars in cumulative losses. The industry continues to depend on them because fragmentation is expensive and users want unified liquidity. The failure mode is consistent: a wrapped asset is a claim on a custodian's obligation, and the custodian's obligation is enforced by the same code that just got drained. The token did not break. The trust assumption behind it did.
Tokenized power has the identical architecture. The token is a claim on a counterparty's obligation to deliver electrons through a wire the token issuer does not own, under a regulatory framework the token issuer does not control, settled in a market the token issuer does not clear. Every layer between the token and the electron is a trust assumption. Stack enough of them and you have rebuilt a bridge, complete with the same concentrated attack surface.
That does not mean the category is worthless. It means the valuation framework must price the trust stack, not the underlying. A tokenized PPA issued by a regulated utility under a filed tariff is a different instrument from a tokenized PPA issued by an entity whose only asset is a website. The market currently prices them within a few hundred basis points of each other. That spread is the trade.
The contrarian read: AI does not lift crypto, it crowds it out
The consensus narrative is convergence. AI needs compute, crypto has compute infrastructure, therefore AI demand is bullish for crypto. I think that chain of reasoning has the sign wrong, and the GoogleโBlack Hills structure is the evidence.
The strategic fact buried in this press release is not that Google signed a power contract. It is that Google signed it with the incumbent regulated utility.
The alternative path was available. Hyperscalers have been pursuing bring-your-own-power arrangements, direct generation, and co-location behind the meter. Amazon's co-location arrangement with Talen Energy at Susquehanna ran into FERC and was rejected in its amended form in late 2024, exposing how unsettled the behind-the-meter regulatory perimeter remains. Microsoft went the other direction entirely, signing a twenty-year agreement with Constellation in September 2024 to restart the Crane Clean Energy Center โ Three Mile Island Unit 1, roughly 835 megawatts โ a deal that only works because a regulated framework exists to support it. Google's own Kairos Power agreement, up to 500 megawatts of small modular reactors with a first unit targeted for 2030, is a decade-long bet that only a balance sheet of that size can underwrite.
Every one of these transactions points the same direction: hyperscalers are routing around the crypto-native model and paying the regulated utility for certainty. The utility gets a rate-based capital expenditure and an approved return. The hyperscaler gets dispatchable capacity with a legal counterparty. The crypto market gets a narrative about convergence and no capacity.
So when the scarce dispatchable megawatt gets contracted for fifteen to twenty years by an investment-grade counterparty, the marginal price of power for everyone else rises. Miners are the marginal buyer. They are also the most interruptible, the most price-sensitive, and the easiest to curtail. AI does not lift crypto at the interconnect. It takes the interconnect, and it leaves crypto holding the hashprice.
The second contrarian point is about what this deal actually validates. Read the structure and ask which model won. A forty-year asset life. A regulator-approved return on equity. A docketed cost-of-service proceeding. Centralized settlement through a utility's billing system. That is the anti-crypto settlement model, executed at a scale no decentralized network has approached. The market read the headline as "Google embraces clean energy." The correct read is "Google selected regulated centralization because the decentralized alternative could not deliver a delivery date."
Structure precedes value; chaos destroys both. The regulated utility has structure and no innovation. Most of DeFi has innovation and no structure. The former is signing fifteen-year contracts with the largest balance sheets on earth. The latter is still arguing about whether a governance-set interest rate curve constitutes price discovery.
What I am watching, and what would change my mind
The information vacuum in that press release will not persist. Long-term power agreements between regulated utilities and hyperscale loads have to surface somewhere, because the cost allocation question cannot be resolved in private.
Three things to track. First, the Wyoming Public Service Commission docket. A special contract of this size requires approval, and the filing will disclose the megawatts, the term, and โ critically โ the cost allocation mechanism. The central question is whether the transmission and generation upgrades required to serve this load are recovered from Google or socialized across the residential rate base. That answer determines whether this is a well-structured transaction or a hidden subsidy, and it is the kind of debt that does not appear on any balance sheet until it appears in a rate case.
Second, the fuel mix and the additionality question. If the filing shows incremental Wyoming wind with hourly-matched attributes, the climate claim holds and the oracle problem is being solved properly. If it shows unbundled certificates against existing generation, then "sustainable energy solutions" was a marketing phrase and the contract is an accounting instrument.
Third, the interconnection queue position and the equipment procurement schedule. A signed contract is a promise. Transformers take years. If the delivery schedule extends past 2028, the capacity is not available for the current compute cycle and the entire thesis shifts to the next one.
My position is unchanged and it is not comfortable. I am not long the convergence narrative. I am long the constraint โ dispatchable capacity, contracted duration, and the regulatory instruments that make a fifteen-year promise enforceable. The rest of this market is still arguing about whether the volatility is the risk.
It is not. The risk is that the capacity was contracted four years before anyone in this industry thought to ask who owned the wire.