# Research Note: India Datacentre Theme — AI & DC Infrastructure Build *Generated by Curiosity Stack | March 2026* *For research and educational purposes only. Not investment advice.* --- ## Signal Twelve of the top thirty performing stocks in portfolio over the past month were concentrated in the datacentre theme — STLTECH (+69%), POWERINDIA (+57%), GVPIL (+55%), APARINDS (+50%), GVT&D (+46%), SCHNEIDER (+46%), DEEDEV (+42%), MTARTECH (+42%), KRN (+41%), PRECWIRE (+39%), TDPOWERSYS (+35%), AEROFLEX (+34%). A performance cluster of this density is not random. It indicates a structural capex cycle in early-to-mid acceleration. The market is paying for the infrastructure build, not yet for operator occupancy or revenue. Three external signals reinforce this internally: India's Union Budget granting datacentre infrastructure status, Microsoft committing $3B and Google committing $15B to India datacentre builds, and a live global fibre optic supply shortage with G.652D prices rising 275–340% in 14 months as of February 2026. --- ## Mechanics A datacentre is a stack of interdependent layers — not a single product. The investor-relevant taxonomy: **Physical shell:** Land, structural steel, pre-engineered buildings (INTERARCH, WELCORP). **Power infrastructure:** Grid connection, transformers, UPS systems, DG sets, and increasingly fuel cells for clean backup (POWERINDIA, GVT&D, SCHNEIDER, ABB, CUMMINSIND, TDPOWERSYS, MTARTECH via Bloom Energy). **Cooling:** Precision air cooling at lower densities; liquid cooling (direct-to-chip or immersion) above 30–40 kW/rack (KRN, DEEDEV, AEROFLEX, VOLTAS, AMBER, BLUESTARCO). **Connectivity:** Fibre optic cables for backhaul and internal structured cabling (STLTECH, FINCABLES, HFCL, APARINDS). **Compute:** Servers, GPU clusters, storage, networking hardware (NETWEB, REDINGTON, SYRMA, E2E). **Operations and software:** Monitoring, management, managed services (DSSL, BBOX). The critical distinction for the current cycle: AI-grade datacentres draw 40–120 kW per GPU rack versus 5–10 kW for a standard server rack. This multiplies the value of every transformer, cooling unit, and cable per square foot of datacentre by a factor of 10–30x. The build is not just bigger — it is structurally denser per unit. --- ## Cause Analysis Five root causes are driving India's datacentre build. They are not equally durable. **Cause 1 — Digital consumption scale (Structural):** India has 900M+ internet users. UPI processes 15B+ transactions per month. OTT streaming, gaming, and e-commerce are compounding. Data volumes requiring local storage and processing grow at 25–30% annually. This cause does not reverse. **Cause 2 — Data localisation mandate (Regulatory floor):** The Digital Personal Data Protection Act 2023, RBI payment data localisation, SEBI and IRDAI requirements collectively mandate that financial and personal data must be stored onshore. Every Indian bank, NBFC, and insurer is a forced buyer of local datacentre capacity. This creates a demand floor that is independent of economic cycles. **Cause 3 — Policy and infrastructure status (Cyclical accelerant):** Datacentres received infrastructure status in Budget 2022, enabling cheaper long-term financing. State-level incentives (Maharashtra, Telangana, UP) front-load capex that would otherwise be phased over 7 years. This accelerates procurement timelines for every vendor in the supply chain. **Cause 4 — Hyperscaler India build (Near-term catalyst, semi-cyclical):** Microsoft ($3B), Google ($15B, including AdaniConneX Visakhapatnam campus), and AWS (multiple Availability Zones) are running active procurement programmes. This is the most near-term revenue-visible cause. It is real and funded — but hyperscaler capex cycles do peak and trough. **Cause 5 — AI inference density (Structural, newest driver):** The shift from cloud-era to AI-era datacentres is a step-change in the economics of every component. AI inference workloads require GPU racks drawing 40–120 kW. This drives transformer capacity, cooling density, and cable spec requirements to levels India's datacentre market has never previously seen. This cause compounds with Cause 4 — hyperscaler spend buys more equipment per dollar than any prior cycle. --- ## Solution Landscape Each cause maps to a solution category, and each solution category is an industry. Cause 1 and 2 together demand: **colocation and sovereign cloud infrastructure** — physical space and compliance-certified facilities with India-resident data. Cause 3 demands: **accelerated private capex** — the policy does not create product; it de-risks returns, which pulls forward build timelines and benefits every supply chain vendor. Cause 4 demands: **utility-scale power infrastructure and high-specification electrical equipment** — hyperscalers issue procurement tenders that flow directly to transformer and switchgear manufacturers. Cause 5 demands: **high-density power distribution, precision liquid cooling, and AI-grade fibre and cabling** — entirely new product specifications that incumbent air-cooling and standard-density vendors cannot serve. The most powerful combination is Causes 4 and 5 acting simultaneously. Hyperscaler capex funds the volume; AI density multiplies the per-rack value of equipment. --- ## Build Requirements and Bottlenecks **Power transformers — Critical bottleneck:** Large power transformers have 18–24 month lead times in India. This is not a temporary supply crunch — it is a structural manufacturing capacity constraint. New capacity cannot come online until 2026–27 at earliest. Incumbents with existing capacity (POWERINDIA, GVT&D) have order book visibility extending through FY27–28. **Liquid cooling — Critical, emerging constraint:** AI racks exceeding 30–40 kW cannot be air-cooled. Liquid cooling (direct-to-chip, rear-door heat exchangers, immersion) is the only viable solution. India's cooling equipment manufacturers are at the early stages of this transition. Air cooling held 67% market share in 2024; liquid cooling is the fastest-growing segment at 27.7% CAGR through 2030. **Fibre preforms — Global shortage, active now:** Only 10–12 companies globally manufacture optical fibre preforms. China holds 53% of global preform capacity and is running at 84% effective utilisation. The top 3 producers hold 40%+ of global capacity. Expansion takes 18–24 months. G.652D (standard fibre) prices moved from ¥16/core-km in January 2025 to ¥60–70 in February 2026 (+275–340% in 14 months). Finolex's own management stated in Q3 FY26 earnings: *"Neither fiber nor preform is available for literally any amount of money."* **DG sets and backup power:** India's grid cannot deliver 40–60 MW at hyperscaler SLA uptime requirements. Every large datacentre requires grid-independent backup. Cummins India (CUMMINSIND) is the dominant specification for hyperscale DC backup power in India. --- ## Value Chain: Key Players ### Global Context | Company | Role | Note | |---------|------|------| | Corning (US) | #1 global fibre manufacturer, ~20% share | Preform + draw | | YOFC (China) | #2 global, ~15% share | Preform + draw | | Vertiv (US) | Cooling, UPS, power distribution | DC specialist | | Eaton (US) | UPS, power management | DC specialist | | Bloom Energy (US) | Solid oxide fuel cells for DC backup | MTAR supplies components | | Prysmian (Italy) | Cables, $96.5M lawsuit vs STLTECH | Competitor | ### India — Watchlist Cross-Reference | Company | Layer | Role | Stage | Notes | |---------|-------|------|-------|-------| | POWERINDIA (Hitachi Energy) | Power | Transformers, grid automation | Mid-accel | Direct hyperscaler procurement | | GVT&D | Power | Transformers, T&D | Mid-accel | Order book full | | SCHNEIDER | Power + Cooling | UPS, energy mgmt, cooling | Mid-accel | New ₹100Cr cooling factory Bangalore | | ABB India | Power | Transformers, switchgear | Mid-accel | Competitive with POWERINDIA | | CUMMINSIND | Power/Backup | DG sets — dominant spec | Mid-accel | Previously watchlist gap | | TDPOWERSYS | Power/Backup | Generators, alternators | Mid-accel | — | | CGPOWER | Power | Electrical equipment | Mid-accel | — | | GVPIL | Power | Electrical infrastructure | Mid-accel | — | | MTARTECH | Power | Bloom Energy fuel cell components | Mid-accel | Non-obvious angle | | PRECWIRE | Power | Winding wires for transformers | Mid-accel | Indirect but direct cycle exposure | | KRN | Cooling | Heat exchangers, liquid cooling | Early-accel | Most direct India liquid cooling play | | DEEDEV | Cooling | Industrial piping for cooling circuits | Early-accel | — | | AEROFLEX | Cooling | Flexible hose assemblies for coolant loops | Early-accel | — | | VOLTAS | Cooling | HVAC, chiller plant | Mid-accel | DC is one segment | | AMBER | Cooling | Cooling components | Early-accel | — | | BLUESTARCO | Cooling | Precision cooling | Early-accel | — | | STLTECH | Fibre | India #1 OFC, ~30% share, integrated | Supercycle | US tariff risk, debt, Prysmian lawsuit | | FINCABLES | Fibre | India ~18% OFC, preform Mar 2026 | Supercycle | Cleanest 2yr risk-reward per analysis | | HFCL | Fibre | ~5% OFC, EPC/turnkey | Mid-accel | No preform capability | | APARINDS | Cables | Power cables + transformer oil | Mid-accel | Dual layer exposure | | BHARTIHEXA | Connectivity | Airtel fibre network infrastructure | Growing | — | | LT | EPC | Primary DC construction contractor | Growing | Large-cap; DC is one segment | | KPIL | EPC | Power EPC | Growing | — | | TECHNOE | EPC | 102 DC builds by FY29 target | Growing | Highest DC EPC purity | | ANANTRAJ | Operator | Real estate → DC operator, 300 MW | Early | Highest optionality, highest risk | | INTERARCH | Build | PEB structures for DC shells | Growing | — | | NETWEB | Compute | India's only listed HPC server maker | Early | Thin institutional coverage | | E2E Networks | Compute | India's only listed GPU cloud rental | Early | Pre-profitability at scale | | REDINGTON | Compute | IT distribution for Cisco, Dell, HPE | Mature | Most liquid; least levered to build cycle | | SYRMA | Compute | Electronics manufacturing services | Growing | — | --- ## Research Landscape **Power and Transformers** are in mid-acceleration. Order books are full. Revenue visibility extends 18–24 months given transformer lead times. The thesis runs through FY27–28 unless Indian manufacturing capacity expands meaningfully ahead of schedule. All names are listed, liquid, and institutionally tracked — not an undiscovered trade, but operating cycle is intact. **Cooling** is in early acceleration and is the least re-rated layer relative to structural opportunity. Liquid cooling transition is still being priced as a future event rather than a present one. KRN, DEEDEV, and AEROFLEX are mid/small-cap with thin institutional coverage — analyst discovery risk runs in both directions. **Fibre and OFC** is in active supercycle — the most time-sensitive sub-theme. STLTECH and FINCABLES are the two India-listed preform-capable players; they represent different risk profiles (see Stress Test section). The window is open; the exit signal is when new global preform capacity announcements begin accelerating, which the PDF analysis places at 2027–28 as the earliest Chinese capacity response. **EPC and Operators** are in the growing stage. TECHNOE is the most DC-pure EPC play. ANANTRAJ is highest optionality but carries real estate developer-to-operator execution risk. Watch for first MW commissioned and anchor tenant disclosed. **Compute and Hardware** names (NETWEB, E2E) are early stage, smaller-cap, thin coverage. Highest volatility; thesis depends on India GPU/AI infrastructure scaling faster than currently priced. --- ## Thesis Stress Test ### Factors Proponents Cite 1. $100B+ in committed hyperscaler capital creates multi-year procurement pipeline with high revenue visibility for transformer and cable manufacturers. 2. AI inference density is a step-change — GPU rack power draw of 40–120 kW makes the cooling and power equipment per DC 10–30x more valuable than prior cycles. 3. Data localisation mandates create a compliance-driven demand floor that does not turn off in a downturn. 4. Transformer lead times of 18–24 months mean supply cannot respond quickly even if demand softens — incumbents retain pricing power. 5. Fibre preform shortage is structural: only 10–12 global producers, 18–24 month expansion cycles, China at 84% effective utilisation with no quick response available. ### Factors Critics Cite 1. Hyperscaler capex cycles are cyclical — they peaked in 2000 and 2018 before sharp corrections. Committed capital can be delayed or phased if macro deteriorates. 2. Fibre prices historically mean-revert faster than other commodities — the 2016–18 boom was followed by a severe 2019 bust. Timing exit matters as much as entry. 3. China holds 53% of global preform capacity and 60%+ of germanium supply. An aggressive Chinese capacity response by 2027–28 could crash fibre prices. 4. STLTECH's US tariff exposure (-760bps/quarter) and Prysmian lawsuit ($96.5M) are binary risks that could reset the stock regardless of the macro theme. 5. Many DC-linked companies have limited disclosed revenue from datacentres — theme attribution may be overstated for some names in the short run. ### Core Assumption The thesis rests on one central assumption: **AI inference workloads at scale require physical GPU-dense datacentre infrastructure that must be built in India specifically** — not served from offshore or from existing low-density facilities. If AI inference shifts to edge computing, satellite delivery, or significantly more efficient architectures (e.g., models that run on 1kW instead of 40kW), the power and cooling intensity thesis weakens materially. ### Milestones Worth Tracking 1. **Finolex preform plant commissioning (March 2026)** — confirms the inflection thesis; a delay pushes back margin recovery timelines. 2. **Quarterly transformer order inflows at POWERINDIA and GVT&D** — sustained sequential growth validates the hyperscaler procurement cycle; any slowdown is an early warning. 3. **China preform utilisation rate crossing 90%** — would signal full capacity constraint, extending the fibre supercycle further; a drop below 70% signals incoming oversupply. 4. **ANANTRAJ first MW commissioned and anchor tenant disclosed** — moves the story from announced capex to operational reality. 5. **India-US bilateral trade deal progress** — resolves STL's -760bps quarterly margin drag; a deal would be a significant catalyst; no deal by end-2026 is a sustained overhang. --- ## How Wrong Can This Be? The biggest single risk is the fiber supercycle resolving faster than expected. The NAND memory analogy from the TheWrap research is structurally apt — but fiber prices mean-revert faster than NAND because fiber is less IP-protected (process more replicable), China has 53% of preform capacity ready to respond, and government-linked demand (BharatNet, military drones) is less durable than enterprise/hyperscaler demand. A scenario in which Chinese producers aggressively expand capacity in 2025–26 in response to the current price supercycle, combined with a Ukraine conflict resolution reducing military drone fiber demand (~10% of global supply), could compress fiber prices by 50–60% from peak by 2027–28 — exactly as happened after 2018. For STLTECH and FINCABLES specifically, the entry timing on the fibre sub-thesis matters as much as the thesis itself. For the power/transformer layer, the risk is lower in the 2-year horizon but real beyond it: India is actively planning transformer manufacturing capacity expansion, and if multiple new plants come online in 2026–27, the 18–24 month lead time advantage narrows. Watch capacity announcement news from new entrants. --- ## 5 Things Worth Reading 1. **CRU Optical Fibre and Cable Market Outlook** — CRU Group: The primary industry data source for fiber/preform pricing, utilisation rates, and capacity forecasts. Referenced in the TheWrap deep dive as the source for China preform utilisation data. [https://www.crugroup.com/market-analysis/products/optical-fibre-cable/] 2. **India Data Center Market Size and Forecast to 2031** — Research and Markets: Comprehensive market sizing, capacity trajectory (350MW 2019 → 1,825MW 2027), and operator landscape. [https://www.researchandmarkets.com/report/india-data-centers-market] 3. **India Data Center Cooling Market 2025–2030** — Mordor Intelligence: Quantifies the liquid cooling transition — 27.7% CAGR for liquid vs air at 67% share declining. Key for validating the KRN/DEEDEV/AEROFLEX thesis. [https://www.mordorintelligence.com/industry-reports/india-data-center-cooling-market] 🇮🇳 4. **Top Data Center Stocks Set to Benefit from India's $100B Investments** — Equitymaster (December 2024): India-focused analysis of listed beneficiaries across the value chain. Useful for cross-checking operator and EPC names. [https://www.equitymaster.com/detail.asp?date=12/13/2024&story=1&title=Top-5-Data-Center-Stocks-Set-to-Benefit-from-Indias-100-Billion-Dollar-Investments] 🇮🇳 5. **IEEMA (Indian Electrical Equipment Manufacturers Association) — Industry Reports**: Primary source for transformer order backlog, lead time data, and manufacturing capacity pipeline in India. Directly validates the 18–24 month transformer bottleneck thesis. [https://www.ieema.org/industry-statistics] --- ## Disclaimer ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ ⚠️ IMPORTANT DISCLAIMER This document is a research and structured thinking output only. Nothing in this document constitutes investment advice, a recommendation, or a solicitation to buy, sell, or hold any security, fund, or financial instrument. The analysis is not produced by a SEBI registered investment advisor. All content is for educational and research purposes only. All named companies are mentioned for research context only — their appearance in this document does not constitute endorsement, recommendation, or any implication of suitability for investment. Always conduct your own independent due diligence and consult a SEBI registered investment advisor before making any financial decision. Curiosity Stack | March 2026 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━