Energy
Power Is the New Bottleneck for AI
The constraint on AI infrastructure has moved from chips to electricity. Where the grid can deliver decides where intelligence gets built.
6 min readDemo
Topic · Artificial Intelligence
AI Infrastructure
Chips, data centres, networking, power and the capital cycle behind compute.
Parallax Index
82 +2
AI Infrastructure
Why now?
- Capital expenditure on AI data centres is at a historic scale
- Inference demand is becoming a larger share of compute than training
- Power, land and grid connections are now gating factors
- Sovereign and enterprise buyers are entering the market beyond the hyperscalers
- Why it matters
- Infrastructure decides who can deploy AI at what cost, and where. It is the most capital-intensive layer of the stack.
- What changes
- Data centre location is increasingly decided by power availability. Utilities, chipmakers and cloud providers are becoming interdependent.
- Who benefits
- Chip designers and foundries, power equipment makers, cooling specialists, and regions with surplus clean power.
- Who is at risk
- Regions with constrained grids, and buyers who over-commit to capacity before demand is proven.
- What happens next
- Expect more co-located generation, long-term power agreements and a sharper split between training and inference sites.
Companies
NVIDIA
Semiconductors & AI computing · Santa Clara, California, US
The dominant supplier of accelerated computing for AI training and inference, and increasingly a platform company spanning chips, networking, software and robotics.
OpenAI
AI research & products · San Francisco, California, US
Frontier model developer behind ChatGPT, with a growing focus on agents, enterprise deployment and consumer scale.
Schneider Electric
Energy management & industrial automation · Rueil-Malmaison, France
Global provider of power distribution, cooling and automation systems, and a critical supplier to the AI data centre build-out.
Technology profile
Where capital becomes compute.
AI data centres are purpose-built facilities that house dense clusters of accelerators, with power, cooling and networking designed for training and serving large models.
- Grid or on-site generation delivers power through substations and switchgear.
- Racks with far higher density than traditional servers require liquid cooling.
- High-bandwidth networking links thousands of accelerators into a single training fabric.
Related topics
AI Data Centers
The buildings, cooling, networking and grid connections that turn capital into compute.
AI Models
Frontier and open-weight models, reasoning systems, small models and the economics of intelligence.
Energy for AI
The electricity demand of AI, and the generation, storage and grid investment racing to meet it.
The Parallax Brief
5 things worth knowing today.
A five-item daily brief: what happened, why it matters and what to watch. Written by the desk, verified by editors.