Waymo signed a deal to push hundreds of megawatt-hours of retired robotaxi battery capacity into grid-scale stationary storage. Not a pilot. Not a press release about reuse. A contract big enough to act as a reliability layer for the local power systems those batteries plug into.

I keep coming back to this because it changes what power availability looks like on a balance sheet. A depreciating fleet asset, the kind of line item that usually shows up in disposal economics, becomes a delivery-timed power input. That is a different kind of object. It has a volume, a duration, a dispatch profile, and a counterparty who needs the electrons to show up on a known schedule.

The temptation is to file this under green corporate behavior. That framing misses what is actually happening. Waymo is not selling a sustainability story. It is monetizing physical capacity that the grid, and by extension the operators running AI-adjacent compute on that grid, are willing to contract for.

What procurement-grade actually means

The word matters. Procurement-grade storage is not a battery sitting somewhere being useful. It is contracted volume, a delivery timeline, a duration of discharge, and dispatch obligations the buyer can plan around. Those four properties turn capacity into something a finance team can underwrite and a reliability engineer can build a load profile against.

Most second-life battery conversations have failed that bar. They have stopped at the chemistry and the carbon math. What changed is that the contracts themselves now look like the contracts utilities sign with peakers and gas generators: a volume, a window, a duty cycle, and a price for showing up.

My read is that this is the inflection. When reuse becomes a thing you can book on a delivery date, it stops competing with sustainability narratives and starts competing with grid expansion timelines. That is a much more interesting market.

The pattern widens past Waymo

Crusoe just signed for 12 gigawatt-hours of Form Energy's 100-hour iron-air batteries, with deliveries starting in 2027. The same announcement added 8 megawatts of repurposed EV battery capacity from Redwood Materials, layered onto a microgrid Crusoe already runs with a 12 megawatt, 63 megawatt-hour second-life installation that was the largest of its kind when it went live.

For context: Form is already building a 30 gigawatt-hour system for Google in Minnesota, reportedly worth about a billion dollars. Two of the largest compute-adjacent buyers in the market are now booking long-duration storage on multi-year delivery windows. That is what scheduled infrastructure procurement looks like when it has graduated out of pilot category.

This is the shift to notice. The contracts have timing. The timing is far enough out that operators are clearly treating these batteries as planned capacity rather than opportunistic supply. A 2027 delivery window means somebody at Crusoe has already drawn the 2027 load curve and decided storage is a load-bearing input, not a hedge.

Power that ships when promised becomes the marketing surface

For years the AI infrastructure pitch has been GPU count, network fabric, and how many megawatts a campus could theoretically pull. The interesting move is that once long-duration storage becomes a booked reliability layer, the differentiator shifts from how much compute you can light up at peak to how reliably you can deliver scheduled compute over a window the customer cares about.

That is a real change in how AI capacity gets sold. Hyperscalers and neoclouds will start positioning around contracted uptime profiles backed by storage they actually own, not just grid interconnects they are waiting on. Enterprise buyers, especially those running long-horizon agentic workloads and inference services with strict latency and uptime SLAs, will start asking harder questions about what is upstream of the rack.

This connects directly to a point I made in Compute Geography Is Becoming the Routing Layer for Agentic Workloads: siting decisions are no longer procurement trivia, they are reliability primitives. Storage procurement is the same logic one layer down. If geography sets the available pool of power, storage decides which hours of that power are actually yours to spend on customer workloads.

The operators who treat circular and long-duration capacity as a contracted input now will be selling a fundamentally different product in two years than the ones who keep treating storage as a sustainability sidebar.

A liability becomes a primitive

The Waymo move is worth lingering on because it inverts a category. A robotaxi fleet is a depreciating mobility asset. The batteries inside it carry a tail of usable life that has historically been written off in fleet accounting. Reframing that tail as contractable grid storage turns the back end of fleet economics into a power product.

That is a template. Any operator running a meaningful battery fleet, delivery vans, last-mile logistics, ride-hail, eventually consumer EV trade-in flows, has the same hidden capacity. The question is whether it gets organized into a procurement-grade product or stays in the recycling waste stream. The companies that build the contracting layer between fleet depreciation and grid procurement will sit on a real margin.

For AI infrastructure operators, this opens a buy path that is shorter than greenfield storage manufacturing and orthogonal to whatever happens with the iron-air or sodium-ion supply curve. You do not have to bet on a chemistry. You have to bet on contracts.

The pricing question markets have not solved

There is a governance echo here worth one paragraph. SpaceX filed for what looks like the largest listing in history, valuing the company at up to two trillion dollars on a multiple that only makes sense if you believe in uncapped infrastructure appetite and founder-driven physical buildout. Public markets are being asked to price growth bets that depend on physical capacity asymmetry. AI infrastructure is the same shape of bet, just less theatrical. The operators who quietly contract for storage now are accumulating the kind of capacity advantage that is invisible until it is decisive. The market has not figured out how to price that yet. It will.

What this asks of operators

The operating discipline is unglamorous: treat power like a workload, not a utility bill. Map your compute roadmap against the delivery windows of the storage contracts available to you. Decide which megawatt-hours you want under contract by which quarter. Stop assuming the grid will expand into your demand curve.

This is the same shape of advantage I described in Agents Are Deployable. The Trust Budget Is What Compounds the Advantage. A trust budget is what you get from designing legibility and audit into agentic workflows early. A reliability budget is what you get from contracting storage early. Both compound. Both look like overspend at the moment of decision and look like a moat eighteen months later.

The operators worth watching will not be the ones publishing the most ambitious data center renderings. They will be the ones whose 2027 power schedule already has names on it.

The question worth holding

The thing I would ask any operator running AI-adjacent compute right now: does your capacity plan treat storage as a contracted input with a delivery date, or as a background utility you assume will be there?

If the answer is the second, you are not really planning capacity. You are hoping. And the operators booking 12 gigawatt-hour windows two years out are quietly making sure your hope is more expensive than their contract.

That is the move. Not green, not clever, not futuristic. Booked power, on a schedule, before the auction clears.

Sources