Three AI Rivals Called for a Slowdown. Here Is What Our Side of the Industry Owes.
The frontier safety debate reaches beyond the lab door. Energy responsibility, failure testing and fault containment belong on the infrastructure agenda now.
Long-form writing on power, liquid cooling and heat reuse for AI infrastructure. Published first on Substack and mirrored here, with links into the evidence and tools each piece points at.
The frontier safety debate reaches beyond the lab door. Energy responsibility, failure testing and fault containment belong on the infrastructure agenda now.
Manufacturing, delivery, and service life: one word—“leak”—hides three clocks, three owners, and three bodies of evidence.
The ten terms behind certification, interoperability — and the white space where project risk actually lives
Models Can Be Downloaded. The Compute, Power, and Cooling Needed to Make AI Broadly Accessible Cannot.
It is a reordering of roles across the AI infrastructure industry
The AI industry is obsessed with the data centers it has not built yet. The bigger story may lie in sites where the power position is already secured, but the compute capacity has yet to be unlocked.
Not for lack of capital — but because capital could not turn into usable megawatts fast enough.
A water-and-chemistry giant just bought a cold-plate company. The number says less about who removes heat most aggressively — and more about where the real moat is moving.
Liquid cooling isn't one technology — it's three, with completely different temperaments. Let me explain them through a decision I still don't love talking about.
CapEx, OpEx, Reuse. How many times can one kilowatt-hour be sold? In the old low-density world, liquid cooling looks like an added cost.
The AI water panic gets the scale wrong, but the local constraint is real. What we learned moving from open cooling towers to warm-water heat reuse.
How warm-water AI data centers turn cooling from a cost into a revenue line — and what we learned heating real buildings with it.
The industry argues about two numbers: dollars per GPU, and tokens per second.
Hotter than a hot tub — and one of the biggest efficiency leaps in the history of the data center.
AI compute creates heat. Smart systems give it a second life.