Knowledge base · canonical index
Knowledge base
The canonical index of frameworks, named theses, topic hubs, and comparison references on adikumar.co. Written for practitioners, sponsors and analysts working on AI infrastructure. All content is by Aditya (Adi) Kumar, an independent analyst covering AI data centre power, industrial technology, and commercial + technical due diligence.
Everything on adikumar.co maps onto three canonical frameworks that compose with each other, the series that instantiate them, and a reference layer of topic hubs, comparison hubs, living indexes, glossary and companion reference assets. The diagram below is the single-page map.
Figure 1. Site architecture: frameworks, series, reference surfaces, companion assets.
The end-to-end AI power pathway from the utility interconnection to the accelerator die. Five physical layers plus a software envelope. Operational naming for the AI Power Chain scope.
Six-layer market + engineering map from grid interconnect to on-package delivery. essays: 6 stacks + 6 technical companions + 3 cross-cutting.
13 core essays + supplements on the 800V DC data centre transition. Architecture, standards, retrofit vs greenfield, vendor economics.
essays on AI infrastructure capital allocation, sponsor economics, M&A + debt underwriting.
essay DD framework covering commercial, technical, financial, organisational, legal, regulatory + closing workstreams.
essays on the rules shaping AI infrastructure buildout: FERC + ISO reform, state PUCs, EPA, CFIUS, export controls, EU AI Act, standards, labour.
essays on capital structures financing AI infrastructure: DDTL, PE equity, ABS, nuclear PPAs, sale-leaseback, green bonds, sovereign wealth, REITs.
vendor + technology deep-dives across the AI Power Chain: Vicor, Menlo Micro, Skeleton, Navitas, Wolfspeed, CoolIT, Submer, Kairos, Fervo, Bloom, Cerebras, Groq, Ayar, Lightmatter, Compass.
8 landmark AI infrastructure deal retrospectives: Blackstone/QTS, Microsoft/TMI, Amazon/Talen, CoreWeave, Iron Mountain, Digital Realty/Teraco, FTC/Nvidia-ARM, CMA cloud.
What thermal management architectures work above 100 kW per rack. DLC + immersion + CDU + chemistry supply chain.
The 800V DC transition: architecture, standards, retrofit vs greenfield, vendor ecosystem.
SiC + GaN vendor map. Wolfspeed, Infineon, Navitas, EPC, ST, onsemi. Design-ins, capex, supply chain.
Every material AI infrastructure deal 2023-2026, four through-lines, fifteen forward hypotheses.
The capital-structure map: DDTL, PE equity, ABS, nuclear PPAs, sale-leaseback, sovereign wealth.
FERC + state PUCs, EPA permitting, CFIUS, export controls, EU AI Act, DORA.
Nuclear restart PPAs, SMR project readiness, traditional PPA structures, grid interconnection.
FERC queue, hyperscaler substation lead times, ISO reform, grid connection economics.
Where each wins, vendor maps for both, and the 800V DC transition that reshapes volume distribution.
Direct-to-chip vs single-phase vs two-phase immersion. Vendor concentration by tier.
Economics, permitting timeline, technology risk, grid interconnection implications.
Comparison of the two 2026 hyperscaler DC power specifications and what standardises when.
Architecture, efficiency, capex, standards, vendors. Where each fits the AI GPU envelope through 2030.
Living index of vendors across DC voltage classes for AI data centres. Updated monthly.
Living index of AI rack power across 11 accelerator generations, H100 to Rubin Ultra.
57 defined terms across AI infrastructure, power engineering, and due diligence. Bookmark reference for reading the essays.
Author bio, background, professional experience at Eaton, Honeywell, Cisco, BCG. Advisory relationships + disclosures.
Tag-based topic index across the site.
