Good Monday —

Meta's $3B off-grid bet signals the public grid is no longer fast enough for AI infrastructure.

By bundling a power plant with a data center into a single loan, Meta sidesteps grid interconnection queues entirely. That removes the single biggest delay risk in large-scale AI infrastructure. The precedent now pressures every competing operator to source and finance their own generation — a harder lift than it sounds.

On closer look, Prometheus is designed to operate independently from the main grid for the first several years, with the option to reconnect once interconnection becomes available. So its an interim strategy, not a structural rejection of the public grid. PJM has already approved roughly $1.7B in AEP transmission upgrades to serve Central Ohio, though AEP has acknowledged the buildout takes 7-10 years to complete. The on-site natural gas generation is bridging exactly that gap.

The more interesting signal is in the financing structure itself. The $3B construction loan prices at 2.5% higher interest rates , higher than comparable Big Tech-backed deals, because lenders are effectively underwriting two distinct asset classes at once.

Meta is paying roughly an additional $3.5B in present value terms for the off-balance-sheet treatment this project finance structure provides.

Signals

Meta bundles power plant financing with data center construction. Operators now face precedent for off-grid economics; builders must secure dual-asset capital.

OpenAI's $121B 2028 compute spend nets $85B burn. Investors must price in compute as permanent cost center, not R&D write-off.

Samsung's 20-system EUV order locks ASML capacity and signals memory supply tightening. Competitors face delayed access to advanced node tools.

CFO questions $121B compute projections. Internal disagreement on IPO timing reflects uncertainty in unit economics; public markets may demand clarity.

8GW campus directly addresses power scarcity. Builders with energy access gain competitive moat; others face capacity rationing.

Regime Snapshot

Compute (CRS): 65 — scarcity. More buyers than available capacity. Lead times extending.

Memory (MRS): 67 — Tight. Demand outstripping supply. No relief until new capacity comes online.

Narratives Moving Today

When compute and memory tighten simultaneously, the window to act is narrow.

The terminal shows exactly which companies are most exposed to both constraints right now — before the market prices it in.

We’re opening the private beta to a small group of investors and analysts. If you want the full picture, not just the daily snapshot:

Keep Reading