Adi Kumar

Opinionated view on AI buildout and automation & controls. Shaped by deep domain expertise, backed by rigorously validated evidence, and infused with inputs from other leading voices in the Industry.

Long-form essays on data centre power architecture, industrial technology, commercial due diligence, and capital flows. Written for PE sponsors, hyperscaler strategy teams, industrial acquirers, management consulting firms, and equity research teams by an operator with twenty years at Eaton, Honeywell, Cisco, and BCG.

How I work with clients → Where to start ↓

20 years Eaton · Honeywell · Cisco · BCG

Recently

FEATURED ESSAY ·Physical Infrastructure

Will Inference Change the Data Centre Architecture?

Training and inference are increasingly discussed as if they require different data centres. Sometimes they do; often they don't. A GB300 rack running reasoning inference looks close to a training rack. A metro colo under a tight SLA is a different category.

25 September 2026 · 62 min read

ESSAY ·ai-power-chain

The modular data centre: how factory-built power and IT halls compressed construction time

Grey-space power skids, white-space IT-hall pods, the four interfaces between them, and the 2024-2026 M&A window that reshaped the vendor field.

22 September 2026 · 77 min read

ESSAY ·Thermal Management

Data Centre Leak Detection

OCP standardisation is making the rope substitutable. NVIDIA's GB300, Ecolab's $4.75bn CoolIT deal and CDU/BMS bundling are moving the economics one layer up. Deep-dive with a 100 MW topology calculator, 24 named vendors, and the qualified-path map.

16 September 2026 · 61 min read

The framework stack

Three frameworks, all published

Three canonical analytical frameworks compose across every essay, case study, and reference asset on the site. Physical infrastructure (AI Power Chain) sits at the bottom of the stack, capital allocation (Investment Layer) sits above it, operational underwriting (Due Diligence) sits alongside as the layer that applies both to live transactions.

Physical · 6 engineering layers

The AI Power Chain

Six engineering layers from grid interconnection to the accelerator die. Vendor concentration, standards state, lead-time exposure per layer.

Capital allocation

The Investment Layer

How the AI buildout gets financed. Where the capital comes from, who deploys it, and what breaks the underwriting when the thesis fails.

Operational · 14 workstreams

Due Diligence

Fourteen workstreams across four stages, with three companion reference assets (Question Banks, Seller-side workbook, DD Evidence Matrix). Operational underwriting for AI infra deals.

For the canonical index of all three, see /frameworks/. For the full site architecture including series, hubs and reference surfaces, see /knowledge/.

See the frameworks calibrated on live 2024-2026 deals: AirTrunk (UNDERWRITE benchmark), CoreWeave DDTL 3.0 (AMBER worked example), Crusoe / Stargate Abilene (RESTRUCTURE, realized).

Start here

Three entry paths by buyer type. Each opens the framework most relevant to your decisions, plus one calibration or series showing it applied.

If you are a PE sponsor or corp dev: start with The Investment Layer, then see it calibrated on the AirTrunk UNDERWRITE benchmark.
If you are on a hyperscaler strategy team: start with The AI Power Chain, then follow the DC-DC Transition series on the 800V DC shift.
If you are an industrial acquirer or strategic: start with Due Diligence for the AI Buildout, then see the framework applied to the Empower / ADI acquisition.