Power Plants Aren't Bedrock. They're Depreciating Assets With a Nice Narrative.

The PJM capacity auction for 2025/2026 cleared at $268.89 per MW-day. The prior year: $28.90. That is not a rounding error. That is a structural shift in the value of something that used to be considered a commodity: existing, dispatchable generation. So when Constellation CEO Joseph Dominguez says existing power plants are the bedrock for data centers, I don't hear an engineering thesis. I hear a balance sheet statement. And as someone who has spent the last decade auditing smart contracts rather than earnings calls, I've learned that the most interesting claims are the ones where the speaker's assets align with the argument. The timing is no accident. Data center power demand in the US is projected to grow two to three times by 2030, reaching 8-10% of national electricity consumption. New generation projects take five to seven years to interconnect. Distribution transformer lead times have stretched from under one year to between two and four years. In this environment, anything that already exists and can spin up quickly becomes a rent-bearing asset. The code doesn't lie, but it doesn't have to when scarcity does the talking. Let me be precise about the technical reality. A hyperscale data center needs 99.99% uptime. That requires continuous power, not intermittent power, not "mostly" power. Solar has an LCOE below nuclear and gas, but that calculation ignores the cost of matching 24/7 load. Once you add storage and firming, the system-level cost of solar is no longer cheaper. The capacity factor of wind is 35-45%, which is nowhere near the availability guarantee a data center demands. The average interconnection wait for new solar in the US is over four years. That is not "immediate." Storage has speed. A lithium-ion battery can respond in milliseconds. It can provide frequency regulation, peak shaving, and maybe four hours of backup. LFP systems now cycle 6,000 to 8,000 times with levelized costs around $0.03-0.06 per kWh. That is transformative for short-duration applications. But it is not a substitute for a 500 MW load running uninterrupted for a week. Flow batteries and compressed air can go longer, but they are still in commercial infancy, with efficiency and integration penalties. The math is simple: storage is a buffer, not a source. Anyone who tells you otherwise is selling a roadmap, not a product. This is why "existing power plants are bedrock" is technically correct in the short term. A combined-cycle gas plant or a nuclear reactor delivers energy with a low marginal cost. Constellation's nuclear fleet runs at roughly $30-50 per MWh marginal cost. But the market is paying far more. In the Microsoft-Constellation deal to restart Three Mile Island, the estimated PPA price was around $115 per MWh. That is a 20-year contract priced at two to three times the marginal cost of the asset. The gap between cost and price is not intrinsic value. It is scarcity rent. I ran a similar stress test back in 2020 when I reverse-engineered Compound's cToken interest rate models. I simulated liquidation cascades under extreme volatility. What I found then was that collateral factors looked safe in normal conditions but became fragile when everyone tried to exit at once. The same principle applies here. Existing power plants look like bedrock during an AI buildout. They become less stable when fuel supply chains tighten, when cooling water is restricted, or when environmental compliance costs rise. The current reliability narrative ignores depreciation. Coal plants running past retirement dates are not new assets. They are accumulated technical debt with a temporary reprieve. Let me push the contrarian angle further. The CEO's framing creates a false binary: existing power plants versus everything else, including renewables plus storage. That binary serves a commercial purpose. Constellation is the largest nuclear fleet operator in the US. It owns gas assets too. If data center customers believe the only immediate option is an existing thermal plant, that locks out future alternatives. It compresses the window for long-duration storage, hydrogen fuel cells, and small modular reactors. It also reframes the policy conversation away from "build new clean generation" toward "pay more to keep old generation running." Notice what is missing from the public statement: any mention of fuel risk. The US still imports roughly 25-30% of its enriched uranium from Russia. The ban on Russian uranium takes effect in 2028, but short-term supply chains remain exposed. Natural gas has its own pipeline and price volatility. When you sign a 20-year PPA tied to an existing plant, you are also signing up for that plant's fuel dependency. The code doesn't read press releases. Neither do commodity prices. There is also a structural irony. Data centers want reliability, but they are pushing the grid into a new regime where reliability itself becomes scarce. The ordinary market response is price signals. The PJM auction is exactly that. But high capacity prices are not an investment in future generation. They are a transfer from electricity buyers to owners of already-depreciated assets. This is not a criticism of Constellation. It is how capital markets work. But calling it "bedrock" implies permanence, when what we are seeing is a temporary bottleneck monetized with surgical precision. The same dynamic happened in DeFi during 2022. Protocols with aggressive lending parameters looked robust during bull markets. They were not. They were just untested at scale. When the retreat came, the fragility showed up in the same place every time: improper risk parameterization. Existing power plants have the same risk profile if you look at their regulatory, fuel, and environmental exposure. They work today. They will work next year. But a 40-year-old asset is not a foundation. It is a bridge. The real question is what happens after the bridge. If data center load hits 8-10% of US electricity consumption by 2030, the demand will be so large that no single asset class can handle it. The winning architecture will be a hybrid: baseload nuclear or gas, plus batteries for instantaneous response, plus renewable PPAs for energy volume, plus demand response to shave peaks. Constellation's CEO knows this. The "bedrock" language is not about excluding storage. It is about establishing who holds the negotiation leverage while the bottleneck lasts. From my own work on verifiable inference oracles and zero-knowledge proof systems, I have learned that every technology stack has an actor that controls the final execution layer. In data centers, that is the power provider. In blockchain, it is the sequencer or validator. The output is only as trustworthy as the layer you cannot easily replace. Right now, existing power plants are that layer. But layers get optimized. Firms get disintermediated. The only constant is that entropy increases without maintenance. Here is my forecast. The next three years will see an aggressive push to sign long-term PPAs anchored to existing thermal and nuclear assets. Prices will stay elevated. The winners will be the generators with fuel-diverse portfolios and the data centers that lock in capacity before the next auction cycle. After that, watch for acceleration in modular nuclear and long-duration storage projects. They will still be late for this cycle, but they will be perfectly timed for the next one. The code doesn't care about narratives. But it does care about incentive structures. And in this market, the incentive structure overwhelmingly favors the seller of an existing, dispatchable, fully-permitted power source. Call it bedrock if you want. I call it a depreciating asset with a transient monopoly on latency. The question is not whether existing plants are the foundation. The question is whether the foundation can hold when the market finally prices in the cost of not building the next one. At some point, the queue clears, new capacity arrives, and the scarcity rent evaporates. When that happens, the fine print in those 20-year PPAs will be the real bedrock. Not the concrete. Not the turbine. The terms.