Inside India's Data Center Build-Out
Inside India’s Data Center Build-Out
ELI5/TLDR
India generates 20% of the world’s digital data but stores only a tenth of it domestically — a gap that regulation and AI demand are now forcing shut. The country’s data center capacity is expected to grow 5x to 8 GW by 2030, requiring roughly $30 billion in facility capex alone. Hyperscalers (Microsoft, Amazon, Google) have collectively pledged over $65 billion toward Indian AI infrastructure. Singularity AMC frames this as a massive EPC opportunity and, not coincidentally, holds a position in Sterling & Wilson’s data center business.
The Full Story
The global backdrop: AI is eating capex
Since ChatGPT launched in late 2022, AI-dedicated data centers have gone from 10% to 53% of global capacity. McKinsey projects 156 GW of AI-related data center demand by 2030 — 124 GW of that added between 2025 and 2030 — requiring $5.2 trillion across the compute value chain. In the US, data center and high-tech spending drove 80% of growth in private domestic demand in H1 2025. Information processing equipment and software contributed more to GDP growth than consumer spending. Hardware is the new macro.
India’s gap: lots of data, not enough racks
India ranks seventh globally in cumulative private AI investment at $11 billion (2013–2024). The US sits at $471 billion, China at $119 billion. The structural oddity: 84% of India’s AI investment goes into application development, leaving the infrastructure layer — the actual physical pipes — comparatively starved.
Data traffic has grown 30x since FY17, but domestic storage hasn’t kept pace. One-fifth of the world’s digital data is generated in India; barely a tenth is stored there.
What’s forcing the build-out
Two forces. First, regulation. The RBI’s 2018 data localization rules, SEBI’s 2023 mandates, and the 2023 Digital Personal Data Protection (DPDP) Act collectively require firms to keep sensitive data onshore. Compliance isn’t optional.
Second, AI demand. The hyperscalers need inference capacity closer to Indian users, and they’re writing checks to prove it:
- Microsoft: $17.5 billion committed to Indian AI infrastructure
- Amazon: $35 billion pledged for AI-driven digitization by 2030
- Google: $15 billion for an AI data hub
That’s $67.5 billion in announced commitments from three companies alone.
The 5x expansion and who builds it
India’s data center capacity is projected to hit 8 GW by 2030, up from roughly 1.6 GW in 2025. The facility capex required — excluding servers and GPUs — runs to about $30 billion.
Singularity breaks the $5.2 trillion global capex stack into three layers: technology developers (60%), energizers (25%), and builders (15%). The “builders” are EPC contractors — the firms that pour concrete, run cable, install cooling, and commission facilities. Execution tolerances are razor-thin. A data center that’s late or built wrong is a write-off. Singularity’s thesis is that this segment is underappreciated relative to its criticality.
Their named pick: Sterling & Wilson’s data center vertical, which they hold in their portfolio.
Claude’s Take
The macro framing is solid. India’s data localization push plus hyperscaler AI spending equals a real infrastructure cycle — that part holds up. The numbers (30x traffic growth, 5x capacity expansion, $30 billion in facility capex) are sourced from McKinsey and industry estimates, not invented.
Where this gets slippery: the article is written by an asset manager talking its own book. Singularity AMC is invested in Sterling & Wilson’s data center business, and the entire piece builds toward justifying that position. The EPC “builders” layer getting 15% of a $5.2 trillion pie sounds attractive until you remember EPC is a low-margin, execution-risk-heavy business where cost overruns eat profits. The article doesn’t discuss margins, competitive dynamics among Indian EPC players, or what happens if hyperscaler commitments get delayed or downsized (as they often do).
The hyperscaler commitment numbers ($67.5 billion combined) are announcements, not disbursements. These figures routinely get revised, stretched over longer timelines, or quietly shelved. The article treats them as locked in.
Still, as a primer on why India’s data center sector is inflecting, it’s competent. Just read it as a pitch deck with footnotes, not independent analysis.
claude_score: 6 — Useful overview of a real trend with good numbers, but the conflict of interest is material and the analysis conveniently stops before asking hard questions about the investment it’s promoting.