The transition from 48V to 800V DC distribution for AI infrastructure. Architecture, vendors, economics, regulation, supply chain.

Direct-to-chip liquid cooling, immersion cooling, and the co-emergence with 800V DC at 100+ kW rack densities.

Wide-bandgap (SiC, GaN) and specialty power semiconductors reshaping AI infrastructure power delivery.

Retrospectives on consequential deals of the AI infrastructure cycle. PE take-privates, industrial consolidation, hyperscaler PPAs.

Debt structures, equity vehicles, and capital sources underwriting the ~$500B-1T AI infrastructure buildout.

The ten regulatory workstreams determining whether AI capex gets delivered. FERC, PUCs, EPA, zoning, CFIUS, export controls, and more.

How hyperscalers are securing baseload nuclear capacity for AI data centres. Restart, SMR, PPA structures.

The FERC + ISO grid-queue constraint that determines when AI capex gets delivered. Post-Order 2023 state of play.

10-layer BOM, semiconductor $/MW, conversion boundary between sidecar and compute rack, and what could kill the sidecar.

$34M of silicon per MW of AI compute, decomposed across compute, storage, networking, and power layers. 1 MW Blackwell reference.

ECAP silicon capacitors, FinFast switching, the strategic logic behind the July 2026 acquisition.

Shortage is not pricing power. Ten commercial positions across the AI infrastructure chain, sorted by the two conditions that determine rent.

48V incumbents face 20-40 percent revenue-at-risk. Vertiv / Eaton / Schneider win through content-dollar expansion 2-3x per rack.

The architectural transition. Loss, copper, density, retrofit cost, when each wins.

Direct-to-chip liquid cooling vs single-phase vs two-phase immersion. When each architecture wins at >100 kW rack density.

Voltage class, switching frequency, cost per kW, qualification depth. The complement vs the substitute.

The two 800V DC specifications under contention. Topology, protection, grounding, vendor ecosystem.

TMI restart, Kairos SMR, Vogtle 3/4 — three paths to firm baseload for AI compute. Timeline, cost, risk.

H100, B200, GB300, Vera Rubin, Rubin Ultra. Per-rack power draw and the architectural implications at each tier.

Interactive scoring across the six-layer power chain for AI infrastructure deals.

162-point physical + 105-point thesis scoring on the AI Power Chain framework.

Scenario tree + FY28 revenue sensitivity for Rubin design-win capture.

Paired-exit valuation across the Rubin Ultra 2027 CPO scale-up window.

VFD vs soft starter TCO across industrial application profiles.

Monte Carlo across macro scenarios for Emerald AI, Enchanted Rock, Verrus.

How grid and AI compute interact through the flexibility right, with an opinionated view on Emerald AI's Series A valuation.

The 2026 state of play across the wide-bandgap semiconductor cohort. Supply, pricing, design wins.

NVIDIA OpenAI backstop, OCP Mt. Diablo spec release, July hyperscaler capex guidance.