Flexible Data Centres, not flexible enough?

The industry’s obsession with data-center flexibility as a grid savior is a dangerous distraction. While massive loads can flex, they can’t solve the local feeder-level failures that actually break the system. Real grid stability in the AI era won’t be built on a few giant batteries—it requires a foundation of massive, distributed DER deployment.

Joti Mangat

12/4/20251 min read

A technician installing solar panels on a residential rooftop under a clear blue sky.
A technician installing solar panels on a residential rooftop under a clear blue sky.

“Data-centre flexibility will save the grid.”

This is a dangerous, wishful thought at the center of the AI energy conversation. 🔥

🤖 — AI data centres can flex.
👍🏾 — that’s good news.

But let’s be clear:

👉 A 200 MW “flexible” data centre is still a single point load sitting at the wrong voltage level to solve the problems that actually break grids.

It cannot fix:

• neighbourhood transformer overloads
• evening feeder peaks
• EV clustering pressure
• local voltage collapse
• hosting-capacity limits

And real operators know the truth no one wants to say publicly:

🤖 AI workloads don’t always flex when the grid needs them to.

Training overruns, inference spikes, replication cycles — all of it can block curtailment.

So here’s the reality:

The more we rely on big flexible loads, the more essential small distributed flexible loads become.

To make “big flex” safe, you need thousands of micro-flex nodes:

🔋 home batteries
☀️ rooftop PV
🚗 shiftable EV charging
🔄 V2H/V2G
🔥 flexible heat pumps
🧠 feeder-level orchestration

DER-as-a-bonus? 👎🏾

DER-as-the-foundation for AI-era grid stability ✅

Without massive DER deployment, data-centre “flexibility” is one firmware update away from becoming a liability.

DER + VPP engineers — what failure modes are people still underestimating as AI load explodes?