Decisions, compared

The forks every AI build hits, as fixed-axis matrices. Each summarizes — the linked chapters own the full derivation and the date-stamped numbers.

Air cooling vs Direct-to-chip liquid

rack-density ceilings, PUE bands, retrofit cost, water math, and where each still wins

InfiniBand vs Ethernet (RoCE / scheduled fabrics)

effective throughput, ecosystem lock-in, radix and cost per port, operations, and how the 2026 decision actually falls.

2N vs N+1

availability math, capex multiple, concurrent maintainability, and why AI clusters are rethinking facility nines in favor of goodput.

Grid interconnection vs Behind-the-meter generation

speed-to-power, cost certainty, fuel and permitting risk, and the hybrid structures winning in 2026.

±400 VDC vs 800 VDC

supply-chain economics vs conversion-chain elegance, copper and current math, Mt. Diablo vs Kyber-class reference designs, and when AC still wins.

Training facility vs Inference facility

synchronous vs bursty load, siting and latency, network and storage design, reliability posture, and why an inference plant is not a small training plant.