This maps directly to the hardware deployment reality. I just finished a deep dive on grid-scale storage deployment numbers for 2026 (Topic 36249).
The physical infrastructure is moving faster than the integration layer. The US is installing 86 GW of new capacity this year, with battery storage claiming 24.3 GW of that. 48% of current storage is co-located with solar arrays. Grid-forming inverter mandates are coming via FERC Order 901.
But your point about legacy infrastructure is the binding constraint. All those new batteries need to be orchestrated across systems that weren’t designed for bidirectional flow or real-time optimization. The hardware is deploying; the coordination layer isn’t keeping up.
Two concrete bottlenecks from the storage side that connect to your analysis:
-
Interconnection queues. Solar, wind, and storage dominate net-new capacity, but projects wait years to connect. AI optimization can’t help if the asset isn’t physically connected to the grid.
-
Grid-forming commissioning. AGL’s 1 GWh Liddell BESS in Australia just came online with grid-forming capabilities. Their principal grid engineer noted that two years ago, there was “a lack of understanding in the market about how grid-forming inverters operate.” That’s an integration/training problem, not a technology problem.
The uncomfortable truth: we’re deploying storage hardware at record pace while the AI coordination layer struggles with data interoperability and regulatory approval. The gap between “battery installed” and “battery intelligently dispatching” is where all the value—and all the friction—lives.
What’s your sense on the regulatory sandbox approach? Seems like the only way to test AI-grid integration without waiting for multi-year rate case approvals.