INSIGHTS

Insights from the owner's side of the table

Analysis grounded in deployments, not press releases

These are not trend pieces. Each article traces back to twenty years of building high-density infrastructure and more than fifty data-center deployments, plus the public record on how the AI buildout is actually being financed and built. Two threads run through everything published here: heat is the real constraint on AI capacity, and the fastest new capacity comes from power that already exists.

$5.2Tprojected AI data center infrastructure spend by 2030 (McKinsey)

82 -> 219 GWglobal AI-ready capacity, 2025 to 2030

4+ yearsaverage grid interconnection queue standing between capital and power

The scale of what these articles are about

Heat Is the Product

Every watt entering an AI data center leaves as heat. The operators who design for that fact first are the ones whose economics work at scale, while the ones who treat cooling as a facilities afterthought are the ones writing change orders. This piece walks through why air stops working, what liquid actually changes, and what that means for anyone about to commit capital to a rack density they have not stress-tested against their building.

Air zoneTransitionLiquid zone

10 kWlegacy enterprise rack

40 kWpractical ceiling for disciplined air containment

140 kWmodern AI training racks, still climbing

The density curve this article is built around

The Smelter Play

The fastest AI data center is the one that does not have to wait for power. This piece explains why dead industrial sites, smelters especially, carry exactly the electrical infrastructure an AI campus needs, why the load profile of a smelter matches an AI training campus almost exactly, and where the diligence separates a real deal from a disaster.

US average interconnection queue4 years

4+ years

Constrained markets10 years

up to 10 years

Retrofit on a site with live power rights0 years

0 years of new queue time

Why the queue is the whole argument

Next step

Read the analysis, or put it to work

Read either piece in full, or if the thesis applies to a project you are already working, start with a readiness assessment or send the details directly.

5 minute readHeat Is the Product

Every watt that enters an AI data center leaves as heat, and the operators who design for that fact first are the ones whose economics work. Rack density has broken the air-cooling playbook, and liquid cooling turns the cooling budget from a facilities line item into the business model.

6 minute readThe Smelter Play

The pace of AI capacity is set by grid interconnection queues, not by capital or hardware. Dead industrial sites with live power rights are the fastest route around the queue, and a $200M idle smelter becoming a $19B twenty-year campus lease shows the scale of the repricing.