🌿 Friday's Climate Infra Brief: Tiny Data Centers and Organized Batteries

News packed 2 weeks! You cannot even make this up:

• Commonwealth Edison recently tried to cancel a $20B, 1.8 GW data center interconnection down in Joliet, Illinois, over inadequate credit support. The credit backing this expansion is literal a $1 LOC, which a FERC commissioner noted was “less than the price of a cup of coffee.” FERC blocked the cancellation over process reasons but declined to rule on whether a single dollar was adequate, kicking the fight to the courts.

• Oracle sent a force majeure letter to its lender, Blue Owl Capital, over the New Mexico piece of its $165B Stargate campus. Oracle’s credit default swaps also hit a record high, because half of its $664B revenue backlog rests on one counterparty: OpenAI.

GW scale data centers are having some credit and procedural issues right now. Occam Edge, an infrastructure project execution risk analytics firm, has more coverage here. Happy to intro if you want to talk to those guys.

Given that BIG headache, the obvious move is to go small - tiny data centers and organized batteries.

Tiny Data Centers: You tuck tiny data centers into the distribution grid (you can read the WSJ article about Crusoe’s next “tiny” plan), or move them physically next to curtailed renewable power instead of fighting for interconnection. Rune is a live example. On September 16, they raised a $40M Series A led by Spark Capital for their “RELIC” modules. These are GPU servers, liquid cooling, and power electronics packaged in one box, plugged in behind the inverter directly at solar farms (DC to DC!). They run fully off grid on clipped solar or wind power that would otherwise go to waste. It is an undeniably interesting emerging asset class, with roughly 1MW deployed across a 200MW Texas solar farm, plus sites in California and Massachusetts. But these small sites face similar issues as all distributed assets do (O&M, remote control reliability, insurability, etc.). Going tiny is not automatically going easy.

Organized Batteries: Another interesting mechanism showed up in a joint white paper Piq Energy and Base Power posted. They modeled a hypothetical 100MW data center sitting at Oncor’s Burleson Switch near Dallas-Fort Worth. Their finding: about 80MW of batteries, placed strategically across several specific grid buses, clears essentially every transmission constraint that data center would create. You get zero curtailment required. The mechanism works because a battery sited exactly where a transmission line is strained relieves that constraint close to 1:1 per MW. Curtailing the distant data center itself only gets you about 0.5MW of relief per MW cut, because the data center’s own bus has less effect on that specific line. Base Power says this spatial approach covers 80 to 85 percent of all constraints. I appreciate the paper uses grid modeling to turn a vague curtailment obligation into a shopping list of megawatts needed at specific nodes.

Here is the pitch, entirely my own half baked business idea. The white paper proves the physics/mechanics work, but there is no direct market paying batteries for this specific transmission relief role yet. But it can be through a bilateral contract with large load itself. Today, battery fleets behave as a grid reserve (paid by all grid users via utility’s capacity program) and a trading asset (buy low and sell high, following power market price movement).

This specific market structure needs an integrator, not a power trader and not a software company. Coordinated battery dispatch software is necessary, but it is not scarce. Voltus, Tesla, Fluence, and fleet owners’ control systems already do this well. The real scarcity is about controlling fleet supply at the specific grid buses that actually matter, which is location specific and cannot be pooled region wide. It is about offering a firm MW guarantee with real financial penalties attached, requiring an actual balance sheet. And it is about getting utility or ISO acceptance for the metering and controls, a slow case by case slog that goes to whoever actually does the grid studies and persuasion work. It is development work and it is not easy. Most asset owners keep dispatch inhouse, and large loads prefer a creditworthy counterparty. Onsite storage behind the data center’s own meter seems a simpler path, even though it delivers less grid relief per MW. The integrator has to be genuinely better, and own the contracts and performance promises in the contracts. One thing I am not sure is how much of the battery fleet capacity must be “on call” all the time to provide relief and how that cannibalizes grid services revenue a typical battery fleet would earn.

Look at the competitive landscape. Piq does the brilliant grid modeling but not the hardware fleet. Base and Branch own real fleets (residential and C&I). Voltus has the closest precedent, a 100MW deal supplying capacity for Google’s data centers in PJM, but they pool capacity system-wide rather than at specific buses. Software players like GridCARE, Camus, and Emerald AI offer deep modeling and technical capabilities but do not own the assets or contracts (I think). Nobody is doing the whole integrator job yet.

The “AI Energy Management Alliance”—featuring Google, NVIDIA, AES, Constellation, NRG, National Grid, (and Generate) launched in mid September is a signal that the big players are working on grid and power solution for data centers (via flexibility design, compute orchestration, onsite power, or other paths). This could be one of the many paths. I am thinking out loud in public here, and I would love to hear from you if you are building in this space.