The Premise
Hyperscale AI workloads need power on a 12–24 month horizon. Conventional interconnection delivers it on a 4–7 year horizon. Co-location with existing baseload generation closes the gap — if the regulatory architecture allows it.
The Talen / Susquehanna ISA filing has become the proxy fight that will determine the shape of the next 36 months of hyperscale dealflow.
Three Structural Models
1) Pure behind-the-meter — load served entirely from co-located generation, no grid charges. Simplest architecturally; most regulatorily contested.
2) Hybrid co-location — primary supply behind-the-meter with grid backup; cost-allocation questions remain open.
3) Front-of-meter PPA proximity — load takes service from grid but locates adjacent to generation; lowest regulatory risk, highest grid-charge exposure.
What to Watch
FERC's substantive ruling in the Talen/AWS proceeding (and follow-ons) will set the template.
PJM's evolving capacity treatment of co-located load.
State PUC posture on retail vs. wholesale jurisdiction at the meter.
Key Takeaways
- 01Co-location is the fastest path to hyperscale power — and the most regulatorily contested.
- 02Architecture choice is now a primary deal variable, not a back-office structuring detail.
- 03Funds underwriting hyperscale infrastructure need a defensible read on FERC trajectory.
Discuss the implications for your platform.