Decoding India's AI Infrastructure: Top Investment Opportunities & Listed Stocks

Global AI infrastructure spending has crossed a trillion dollars in committed capex, and India until recently a footnote on the global data centre map is now attracting a meaningful slice of it. AWS has committed $12.7 billion to India by 2030; Microsoft, Google, Meta, and Oracle have all announced multi-billion-dollar builds; Adani alone has earmarked $100 billion (~ ₹8,00,000 Crores) an integrated renewables-plus-data-centre platform. For investors, one should now think how the AI computing works? And what actually data centre companies making value addition in AI computing?

Let’s first understand, What is AI computing, in plain English?

An AI model is a very large mathematical function trained on enormous datasets. Two things happen inside a data centre: training (building the model which needs massive parallel compute for weeks or months) and inference (running the model to answer queries lighter, but billions of times a day).

Ordinary CPUs do one thing at a time, very fast. AI workloads need thousands of small calculations run in parallel which is what GPUs do best. NVIDIA's H100 and B200 chips are the industry standard because they combine parallel compute with the CUDA (Compute Unified Device Architecture) software ecosystem developers depend on. A single training cluster today uses tens of thousands of these GPUs, wired together, cooled aggressively, and fed by megawatt-scale power. Every one of those requirements creates a listed-company opportunity.

NVIDIA captures more gross profit per dollar of AI capex than any other company in the stack over 70% gross margins, because there is effectively no substitute as their chips are CUDA compatible. AMD is the credible 2^nd company. Indian investors cannot own this layer directly through domestic listings, but the ripple effect of NVIDIA's roadmap dictates capex cycles for everyone else below.

Chips arrive from NVIDIA; someone has to design the actual server motherboard, power delivery, thermal management, rack interconnects and manufacture it. Globally, Supermicro is the poster child. In India, Netweb Technologies occupies this niche as an official NVIDIA MGX partner, licensing the modular reference architecture and shipping servers under its "Tyrone" brand. In Q1FY27, AI Systems became 62% of Netweb's revenue, up from 7% two years ago. A genuine deep-tech OEM story, but with two caveats: gross margin is thin because GPUs are a pass-through cost, and revenue is capped by NVIDIA's chip allocation.

Servers need buildings with reliable power, cooling, and low-latency connectivity a real estate-cum-utility business with long contracts and stable annuity cash flows. STT GDC (Tata Communications owns 26%) is India's largest colocation player. AdaniConneX  the 50:50 Adani-EdgeConneX JV has anchored Google's Vizag and Microsoft's Hyderabad builds. Nxtra (Airtel 74%, Carlyle 26%) is a strong number 2. Yotta (Hiranandani Group) is heading to Nasdaq via a SPAC merger at a $4.2 billion valuation. Pure-play Indian listed exposure is limited investors access this layer indirectly through Adani Enterprises, Airtel, or Tata Communications.

Not every enterprise wants to own AI infrastructure. GPU cloud players rent capacity by the hour, arbitraging the gap between chip capex and rental income. Globally, AWS, Azure, and Google Cloud dominate; specialist CoreWeave has become a $30-billion-plus US-listed company doing exactly this. In India, E2E Networks now majority-owned by NSE-parent NKCL is the closest listed pure-play. Yotta's Shakti Cloud is the private-market equivalent. Opex-heavy and margin-sensitive, but scales beautifully at high utilisation.

This is where many Indian investors under-appreciate the opportunity. Every megawatt of AI compute needs roughly 1.2–1.5 MW of standby power, plus cooling that consumes another 30–40% on top of IT load. That translates to demand across:

• Diesel gensets (backup): Cummins India data centres are already ~25% of its power generation segment and rising
• Power transformers: CG Power, Transformers & Rectifiers, Hitachi Energy — CG's ₹900 crore Tallgrass order in early 2026 was the first major benchmark
• Alternators (for gas-turbine gensets abroad): TD Power Systems — an export play on US behind-the-meter gas turbine deployments, riding GE Vernova and Siemens Energy's capex cycles
• Cooling systems: Blue Star and Voltas the domestic proxies for a business globally dominated by Vertiv
• Fuel cell components: MTAR Technologies a critical Tier-2 supplier to Bloom Energy, whose solid-oxide fuel cells are being deployed by US hyperscalers for fast behind-the-meter power. This is an indirect but real AI-DC connection.
• Precision electronics manufacturing: Kaynes Technology riding both semiconductor OSAT (via its Sanand facility) and broader EMS demand from data centre and telecom customers

Why India? Five structural pulls at once

1. Data localisation — The DPDP Act (2023) requires many categories of personal data to be stored in India; hyperscalers can no longer serve Indian users purely from Singapore or Frankfurt.
2. Sovereign AI push — The IndiaAI Mission has committed ₹10,372 crore for a sovereign compute stack, with 18,000+ GPUs already empanelled. Netweb, Yotta, and CtrlS are direct beneficiaries.
3. Hyperscaler capex — AWS ($12.7 bn by 2030), Microsoft ($3 bn), Google ($10 bn via Adani JV), Meta and Oracle all have India commitments in flight.
4. Renewable power availability — India offers what the US cannot: cheap, abundant, buildable renewable capacity with fast interconnection. Adani's Khavda 30 GW and NTPC Green's pipeline solve the industry's biggest emerging bottleneck power.
5. Cost arbitrage — Land, labour, and construction remain a fraction of US or European equivalents, even after normalising for reliability differences.

Silicon captures the most value per dollar but is not accessible domestically. Support infrastructure is the least glamorous but offers the cleanest, longest-duration exposure through established listed names. Data centre real estate and GPU cloud sit in the middle high visibility but capital-intensive. The Indian AI compute theme is best played not as a single stock but as a thematic basket across these layers, weighted toward whichever cycle you believe is turning next. The capex is landing, and the value chain is now clearly mapped. Position accordingly.

Thank you for joining us in this special edition of the Financial Chronicle! We hope you're as excited about these changes as we are. Until next time, Happy investing!