The AI bottleneck is moving from compute to power
As AI clusters scale, power stops being an operational detail and becomes a system design problem. This deal signals how suppliers are productising energy delivery for AI workloads that swing fast and hit hard.
Why AI data centres are different customers
- AI training and large-scale inference create extreme load patternsrapid changes that traditional power setups aren't built to smooth.
- Reliability expectations are unforgiving: downtime and power quality issues cascade directly into SLA risk and lost revenue.
What 'purpose-built' power looks like here
- Onsite 'behind-the-meter' generation designed to avoid adding stress to public grids.
- Hybrid approaches that pair generation with battery energy storage to absorb spikes and stabilise output.
- Long delivery and build timelinesequipment deliveries in this case are scheduled across 20262027, underscoring how far ahead planning must run.
Why this matters for platform leaders
- Data centre strategy increasingly depends on energy partnerships, permitting, and supply chain coordination.
- Procurement shifts from buying electricity to buying power capability: response time, quality, and scaling path.
- Investors will ask whether your growth assumptions survive real-world grid constraints.
The board-level takeaway
If your AI roadmap assumes 'we'll just add more GPUs,' you're missing the next gating factor. Power is becoming the new capacity planning discipline.