Industry Outlook
Capacity Trajectory:
In the past decade, the capacity of data centers has grown significantly. In 2007, it stood at 75 MW, escalating to 597 MW by 2020. Projections indicate a continued upward trend, with expectations of 1,650 MW by 2025 and reaching 5,000 MW by 2030, suggesting a compound annual growth rate (CAGR) of 26% over the next five years. There is an active pipeline of approximately 3 GW, requiring an estimated capital expenditure of around $25 billion.
The development of the industry can be categorized into distinct phases: the 2G/3G era (2007–2014), the rise of smartphones and Jio 4G (2016–2020), the COVID-induced digital shift, and the upcoming period characterized by 5G, cloud advancements (2022–2024), and significant AI developments expected from 2025 onwards.
Trends
The industry is witnessing a dual demand engine characterized by two main aspects:
- Foundational Drivers: These include cloud adoption, data localization, 5G, and business continuity planning (BCP).
- AI Workloads: Initiatives such as the IndiaAI Mission, enterprise-level AI integration, and sovereign computing are pushing demand.
Several structural shifts are underway:
- Transition from megawatt (MW) to gigawatt (GW) scale competition, primarily driven by AI factories.
- A shift from training to inference dominance, with global GPU expenditures showing a projected rise from 34% inference application in 2023 to 36% by 2027. The inference segment is expected to grow five times faster than training.
- A transition from air-cooled to liquid-cooled systems is becoming the standard.
- A movement from colocation towards vertical integration that combines data centers (DC) with GPU cloud methodologies and software.
- The landscape is evolving from domestic-only operations to the entry of global capital and operators.
AI Adoption Insights
As of now, India holds the position of being the second-largest user base for ChatGPT globally, accounting for 9% of the market, trailing behind the United States, which comprises 18%. The AI market in India is projected to grow from $13 billion in 2025 to $130 billion by 2032, at a staggering CAGR of 39%. Insights indicate that 45% of enterprises are currently integrating AI into their operations, while only 6% have yet to initiate any form of AI deployment.
Furthermore, 64% of enterprises express a desire for in-house AI solutions utilizing cloud-based GPUs, underscoring the demand for domestic Neoclouds. The landscape boasts over 1,800 global capability centers (GCCs), with over 500 focused specifically on AI, and 89% of new startups being AI-native.
The segment-wise breakdown of the AI market projected for 2025, valued at $13 billion, is as follows: $2.5 billion in Banking, Financial Services, and Insurance (BFSI); $1.8 billion for startups; $1.6 billion in media; $1.4 billion in manufacturing; $1.3 billion in tech services; $1.2 billion in public services; and $3.3 billion categorized as others.
The IndiaAI Mission, with a budget of $125 billion over five years, has already committed over 38,000 GPUs, with approximately 22,000—58% of the total—allocated. The mission includes 3,000 datasets and 243 AI models designed for various sectors. Selected developers of large language models (LLM) include Sarvam, Gnani, Soket, and Gan.AI.
Key Building Blocks (Value Chain)
The burgeoning industry operates on a comprehensive bottom-up stack:
- Data Centers: Key players include Sify, NTT, and Equinix, which provide the necessary physical infrastructure.
- GPU Hardware: Nvidia dominates with a market share of 90-95%, while AMD holds about 5%, and Intel has less than 1%.
- AI Cloud Service Providers: Companies like AWS, CoreWeave, and Yotta serve as gateways into the market.
- Compute Software: Solutions from companies such as GPT, Gemini, Sarvam, and DeepMind drive this layer.
- End-user Applications: Prominent applications include Copilot, ChatGPT, and Gemini.
Nvidia’s roadmap for GPU advancements includes the Hopper in 2024, Blackwell in 2025, GB300 in 2026, Rubin in 2027, and Feynman in 2028. The utility cycle of GPUs is categorized as follows: the latest generation is suited for training models, lasting 12-15 months per large language model, while previous generations can be used indefinitely for inference. Financial projections indicate a six-year financial life for GPUs, with a longer physical lifespan.
GPU Cloud Unit Economics
In terms of revenue generation metrics associated with an Nvidia H200 8-GPU server:
- Capital Expenditure (Capex): ₹27.3 million
- Annual Revenue: ₹12.5 million
- EBITDA: ₹9.8 million/year
- EBITDA Margin: 78.6%
- Payback Period: Approximately 2.8 years
- Pricing Model: ₹195 per GPU per hour (blended)
- Utilization Rate: 88% (blended)
- Power Usage Effectiveness (PUE): 1.4 (for liquid-cooled systems)
For a scaled setup comprising 3,000 GPUs distributed across 375 servers, the projected internal rate of return (IRR) stands at 20.3%, with an equity IRR of 28.4% under a hold-to-maturity (HTM) basis. The financial structure entails 60% leverage, with a debt cost of 10% over a five-year tenor and a debt service coverage ratio (DSCR) of 1.5x. Contractual arrangements forecast 75% as take-or-pay, 20% as merchant, and 5% as spot pricing.
Price tiers are structured such that take-or-pay is valued at ₹300/month, merchant at ₹225, and spot pricing at ₹300 with reduced utilization projections.
DC Hub Map (1,650 MW Total)
The data center hub map indicates the following capacities and developments:
- Mumbai: 801 MW capacity, 2.9% vacancy, 448 MW under construction, and an additional 893 MW in the pipeline.
- Chennai: 268 MW capacity, 12.4% vacancy, with 153 MW under construction and 273 MW in the pipeline.
- Delhi-NCR: 161 MW capacity, 10.2% vacancy, with 47 MW under construction and a pipeline of 270 MW.
- Hyderabad: 138 MW capacity, 9.7% vacancy, with 106 MW under construction and 200 MW planned.
- Bengaluru: 119 MW capacity, 6.9% vacancy, with 19 MW under construction and 107 MW in the pipeline.
- Pune: 111 MW capacity, 2.0% vacancy, with 30 MW under construction and 160 MW planned.
- Kolkata: 17 MW capacity, 3.5% vacancy, with 15 MW under construction and a pipeline of 84 MW.
Key takeaways include that Mumbai accounts for 50% of total capacity and 47% of new supply, bolstered by a substantial cable landing infrastructure of 12 stations. Chennai is poised for 15% of incremental growth, particularly with the arrival of three new subsea cables set to land between 2026 and 2027. Meanwhile, Hyderabad is emerging as a self-build hub for hyperscalers, with Mumbai currently ranked sixth globally regarding under-construction capacity.
