Inflation as a Down Payment on Deflation

Renewable power plants are strange assets. Almost the entire cost is paid upfront — the panels, the turbines, the interconnection — and then, for two or three decades afterward, the fuel is free. Wind and sun don't send invoices. A gas plant is the opposite: cheap to build, expensive forever, because someone has to keep buying the gas. That structural difference — pay once versus pay perpetually — is not a minor accounting detail. It's the entire argument for why the current spending spree on AI and clean energy might be better understood as a down payment than a permanent cost increase.

The spending is real and it is large. Hyperscaler capital expenditure is now running above a trillion dollars annually, most of it flowing into power-hungry data centers and the transmission, transformers, and generation needed to feed them. Add the parallel buildout of renewable generation and grid modernization, both drawing on the same copper and skilled labor, and you get an investment cycle large enough to move commodity prices and stretch equipment lead times across the entire economy. This is, by any reasonable definition, inflationary. The question worth asking isn't whether it's inflationary — it obviously is — but what, structurally, that spending is buying.

Central banks tend to treat inflation as a malfunction to be corrected. But a capex cycle is not the same thing as a monetary shock; it's closer to an investment with a payback period, and the payback here has a name economists have been arguing about since 1987. Robert Solow, watching computers spread through every office in America without productivity statistics moving an inch, wrote that you could see the computer age everywhere except in the economic data. The lag between spending and payoff turned out to be real — roughly a decade before enterprise IT investment showed up cleanly in national productivity numbers — and AI appears to be repeating the pattern almost exactly. Federal Reserve surveys this year found the large majority of firms reporting negligible measured productivity gains from their AI investment; Goldman Sachs and JPMorgan both cut their productivity forecasts in response. That's not evidence the thesis is wrong. It's evidence the lag is happening on schedule, if you believe the electrification and IT precedents are the right ones to trust.

The clean energy side of the ledger has a cleaner, faster-moving version of the same story, because the mechanism is simpler than productivity statistics: it's just marginal cost. Once a wind or solar asset is built, the cost of the next unit of electricity it produces is close to zero, and wholesale electricity markets dispatch the cheapest available power first. That single fact, replicated across enough installed capacity, physically displaces expensive gas generation from the price-setting position. The empirical relationship is already measurable — European studies have found that each percentage point of added renewable penetration reduces average wholesale power prices by roughly half a percent, and that pushing penetration to 50% could cut wholesale prices by a fifth. Unlike the AI productivity question, this isn't a hypothesis awaiting confirmation. It's a mechanism already running, at smaller scale, inside grids that have partially made the transition.

Where this gets complicated is that the deflationary payoff and the inflationary spending aren't sequential — they're happening at the same time, in the same markets, which makes them easy to mistake for contradictory trends rather than one process at two different stages. A wind developer today is absorbing higher transformer costs and multi-year interconnection queues to build an asset that, once operating, will spend the next twenty-five years pushing prices down. NextEra, Enel, Iberdrola, and EDP Renováveis are all, structurally, doing exactly this trade simultaneously across different geographies — paying the current inflationary toll to become, eventually, the deflationary force in their local markets. Vestas and Enphase sit further upstream, selling the hardware that makes the trade possible; their order books are a reasonable leading indicator of how much future price suppression is currently under construction. Bloom Energy's behind-the-meter model is a related but distinct bet: bypass the inflated, congested grid-connection queue entirely and let the customer pay a premium now for power certainty, rather than wait years for the same electron at a lower price.

Our prediction is based on a simple long-term process: the same companies currently absorbing elevated input costs and long lead times — the grid builders, the renewable developers, the transformer and switchgear suppliers like Eaton and GE Vernova sitting at the center of both the AI and clean energy buildouts — are structurally positioned to become cheaper, not more expensive, to operate against once the current wave of capacity comes online. That's a different bet than "AI will be productive" or "the war in the Middle East will end." It's a bet that physical infrastructure, especially clean energy, behaves the way physical infrastructure has always behaved: expensive to install, cheap to run. If that holds, today's inflation isn't the economy's steady state, it's closer to a bridge loan the physical world is taking out against its own future marginal cost.

Previous
Previous

The Grid Doesn't Know What Kind of Load You Are

Next
Next

AI Supply Chains: One Copper Mine, Two Systems