PDF ↗
Dart ConsultantsDart HomeAI Compute StackDC Physical InfrastructureUnified: Wafer to Substation
Home/Primer/Synthesis
Section 10

Synthesis

Where the opportunity actually originates, across the whole stack

OriginWho captures itTime horizon
1. Reference-design assemblyWhoever has EMS capacity and NVQual certification — increasingly contestable and margin-compressing, not lessNow, but thinning generation over generation
2. Engineered power/grid equipmentVoltamp, Hitachi Energy India and ABB India show the clearest evidence; several peers show rich valuation without matching evidenceNow through at least the current global equipment-scarcity cycle
3. Government-incentivised capability buildingWhoever wins scheme approval and executes on schedule — CG Semi and Kaynes Semicon already in commercial productionNow through the early 2030s, government-set timetable
4. Chip/IP design-services exportsLTTS, Cyient and Sasken all participate, none discloses how muchOngoing but structurally undercounted (§8)
5. Differentiated cablingSterlite Technologies most concentrated; Polycab and KEI Industries more diversifiedNow — the single most concentrated DC-revenue story in this report
6. GPU-scarcity and equipment-scarcity rationingNobody in this report's universe — both ceilings sit entirely upstream, at Nvidia/TSMC/memory makers on one side and global transformer/GIS manufacturers on the otherAt least through 2027 (chip) and on an unclear, slower timetable (equipment)
7. GPU-cloud consumption and sovereign-AI demandE2E Networks and ESDS most directly — priced at multiples (157-401x trailing) this report's own arithmetic cannot support on either name's currently disclosed earningsNow, but on unstable, early-stage financial bases

Six distilled conclusions for the combined stack

  • The same asymmetry — commodity execution beside scarce allocation or engineering — repeats independently at every layer, which is the strongest evidence this is one industry, not two. Two research efforts, ten sub-industries, thirty-five companies, one pattern every time.
  • The gatekeeper's shape changes the investment case more than its existence does. A concentrated gatekeeper (Nvidia) reprices its layer fast, in either direction; a diffuse constraint (global equipment manufacturing) moves slowly in both directions and rewards patience over positioning.
  • Scale and stack-relevance are not the same axis, anywhere in this report. A ₹12,491cr pure-play GPU-cloud operator and a ₹11.8 lakh crore telecom major can both appear in a "data-centre stocks" screen — the question is never how big, but how much of the company's revenue and margin sits on the scarce side of the line at whichever layer it occupies.
  • The industry's own reported economics argue against most richly-valued names in this report being undervalued. Real (not indicative) margin data shows compression at the assembly and civil/EPC layers — yet several of those exact names trade at rich multiples, and two GPU-cloud names trade at 157-401x on earnings this report's own research found reasons to distrust.
  • Six of thirty-five companies share one under-appreciated pattern: real data-centre exposure that is financially immaterial to the parent. Bharti Airtel, Tata Communications, Adani Enterprises, RailTel, Anant Raj and L&T all belong in this report for completeness of India's AI/data-centre ownership map — not because the data-centre angle moves any of their own investment cases today.
  • Seven of thirty-five names carry structural, governance, legal or data-integrity overhangs a reader must understand before any valuation multiple means anything — see §9's dedicated callout.

Bull case / bear case

The bull case

Global hyperscaler capex is not slowing — four companies alone guide to ~$725bn in 2026, up 77% YoY — and India's own build-out is executing, not just announcing: three semiconductor OSAT/ATMP facilities in commercial production, and data centres genuinely the largest single contributor to at least one power-equipment company's order intake this report confirmed directly with management. The IndiaAI Mission is a policy-guaranteed domestic demand stream already generating confirmed nine-figure orders. At least three companies across both halves of the stack (Voltamp, ABB India, Sterlite Technologies) show management-confirmed, quantified current demand, not analyst speculation, and several carry improving, multi-agency-endorsed credit ratings reflecting genuine balance-sheet strength.

The bear case

This report's own, real margin data shows compression, not expansion, at the two largest sub-industries by company count in its universe — server/EMS assembly and civil/EPC construction — regardless of order- book size, with management guidance at more than one construction name pushing recovery past FY28. The single best-quantified data-centre revenue story in the report (Sterlite Technologies) sits on financial disclosures this report's own research found genuinely unreliable. Two GPU-cloud names trade at valuations unsupported by any earnings base this report trusts, and seven of thirty-five companies carry a governance, legal or data-integrity overhang serious enough to dominate their own rating independent of the sector story. Both this report's scarcity mechanisms — chip allocation and global equipment-manufacturing capacity — sit entirely outside the control of every company covered here.

Numbers to track, across the whole stack

  • Nvidia's own component-customisation percentage in each new reference-design generation — the cleanest read on whether compute-layer assembly margin compression is easing or tightening
  • Whether the IndiaAI Mission reaches its 100,000-GPU end-2026 target, and which India-listed companies win the resulting server-supply orders
  • Resolution of Sterlite Technologies' three-way conflicting quarterly PAT figures against its audited FY27 annual report
  • Whether KEC International's own FY28 double-digit-margin guidance is met, as a test of whether civil/EPC margin compression across this report's construction layer is bottoming or continuing
  • Whether any India-listed company enters the precision-cooling/DC-UPS space this report found structurally absent from the public markets
  • Whether E2E Networks and ESDS post consecutive profitable quarters on a cash-earnings basis, rather than the depreciation-distorted or single-contract-concentrated figures this report's research found
  • The outcomes of Anant Raj's Enforcement Directorate matter and Adani Enterprises' OFAC/PIL overhangs — both live, unresolved as of this report's research date
Educational material only — not investment advice. Dart Consultants is not a SEBI-registered Investment Adviser or Research Analyst.