Five pools of value or risk are structurally undercounted, mis-modelled, or simply invisible to the headline statistics used to size this opportunity — three inherited from each half's original research, plus two that only became visible once both halves were placed in the same document.
India's global capability centres and engineering-R&D firms do real chip-level design work for global fabless clients today — Nasscom projects India's ER&D services revenue will cross US$100bn by 2030 — none of which shows up in any "India AI hardware" statistic, because trade data books it as a generic IT/BPM export. Two of this report's companies (LTTS, Cyient) earn real revenue from exactly this kind of work, disclosed at less granularity than an outside analyst would like.
JLL's own research found pre-committed hyperscale capacity made up 82% of India's H1 2026 data-centre absorption (§6) — meaning most new capacity changing hands today is contractually locked in years before it is switched on and counted in any operational-MW statistic. Every published "India data-centre capacity" figure structurally understates near-term demand for the chips, servers, transformers and cooling plants this report's companies actually sell.
Reporting aggregated in this research (single-source, not independently confirmed against a primary regulatory document, and flagged as such) suggests India may be permitted to import on the order of 50,000 H100-class-equivalent GPUs through 2027 under US export-control tiering. If accurate, this functions as a hard rationing ceiling on the entire downstream opportunity — the servers, racks, cooling systems and power equipment this report's companies sell are all ultimately gated by how many chips are allowed into the country.
This is the clearest finding that only emerged from combining both halves in one document. Companies across both the compute layer (Bharti Airtel's Nxtra, Tata Communications' divested DC business) and the physical layer (L&T's ₹36.6cr FY26 DC revenue against ₹2,85,874cr consolidated group revenue — roughly one-hundredth of one percent) share the identical pattern: real, disclosed data-centre assets or stakes that do not meaningfully move the parent company's own valuation case. Six of this report's thirty-five companies fall into this pattern (§9's dedicated grouping) — a large enough share that this report treats it as a structural category in its own right, not a handful of exceptions.
An investor working from the intuitive assumption that "the compute layer is where the real AI exposure lives, and power/cabling are secondary" would miss this report's single strongest data point entirely: Sterlite Technologies, a cabling company on the physical side of the stack, saw data centres grow from roughly 1% to 21% of quarterly revenue within one year — the most concentrated, best-quantified figure anywhere in this report's thirty-five companies, ahead of every compute-layer name. That finding sits paired with an equally important caveat this report does not let the exciting number crowd out: a genuinely unreconciled, three-way conflict in the company's own disclosed quarterly profit figures (§9).
Every company report that follows should be read against this backdrop: a company's disclosed AI or data-centre revenue is very likely either understated (buried inside a larger undifferentiated line, or locked in but not yet operational) or, less often but importantly, the headline number itself is more exciting than the underlying data quality supports. This report treats both directions of that distortion with equal weight, layer by layer.
