AirTrunk Bets $30B on 5GW of AI Data Centers in India
Australian data center operator AirTrunk has committed $30 billion to build 5 gigawatts of AI data center capacity across India, one of the largest infrastructure bets on AI compute in the Asia-Pacific region.
Original sourceAirTrunk, the Australian data center operator acquired by Blackstone in 2024 for roughly $16 billion, has announced a $30 billion commitment to develop 5 gigawatts of AI-focused data center capacity in India. The buildout represents one of the single largest infrastructure investments targeted at AI compute in the region, and signals serious confidence in India's trajectory as both a consumer and producer of AI workloads.
India has been aggressively courting hyperscaler and data center investment, with the government pushing digital infrastructure as a national priority. The country's combination of a large developer population, growing enterprise AI adoption, and relatively lower land and labor costs makes it an increasingly attractive alternative to saturated Western markets where power availability and permitting have become genuine bottlenecks. AirTrunk's bet is that India won't just be a cost-optimized offshoring destination but a primary AI compute hub in its own right.
The 5GW figure is notable for its scale — for context, the entire installed data center capacity of India today is estimated at roughly 1-2GW. This investment would more than triple that, though deployment will occur over multiple years and locations across the country. AirTrunk has existing campuses in Australia, Japan, Singapore, Hong Kong, and Malaysia, and India fills a significant geographic gap in its Asia-Pacific footprint.
The announcement comes amid a broader wave of hyperscaler and data center operator commitments to India, including investments from Microsoft, Google, and AWS. What distinguishes AirTrunk's play is the sheer magnitude of the commitment and its explicit framing around AI workloads rather than general-purpose cloud capacity, suggesting a specific thesis about where GPU-dense compute demand is heading in South and Southeast Asia.
Panel Takes
The Futurist
Big Picture
“The thesis here is falsifiable and specific: AI inference demand in South Asia will grow fast enough to fill 5GW of capacity before cheaper alternatives materialize or geopolitical friction disrupts foreign infrastructure ownership. That's a bet on three things going right simultaneously — Indian enterprise AI adoption accelerating, power grid buildout keeping pace, and regulatory environment staying favorable to foreign-owned critical infrastructure. The second-order effect that matters most isn't the compute itself — it's that 5GW of dedicated AI infrastructure creates a gravitational pull for AI startups, model labs, and talent that could reposition India from an outsourcing economy to an AI-native one. AirTrunk is riding the trend line of compute geography decentralization, and at $30B they're not early anymore — they're making a conviction bet that they're still on time.”
The Founder
Business & Market
“The buyer here is hyperscalers, sovereign AI programs, and large Indian enterprises — all of whom have budget and urgency, which is the right answer. Blackstone's ownership matters because this isn't a startup burn-rate story; it's an infrastructure yield play where the moat is land, power contracts, and permitting — none of which can be replicated quickly by a competitor who shows up late. The real business risk isn't demand — India's AI compute appetite is real — it's execution: power procurement in India is notoriously complicated, and 5GW announced is a long way from 5GW energized and leased.”
The Skeptic
Reality Check
“A $30B commitment announcement is not $30B spent, and the distance between those two things in Indian infrastructure projects has historically been vast — permitting delays, state-level power negotiation, and grid reliability are not small variables. I'd want to see phased capacity milestones with hard dates before treating this as infrastructure that exists rather than infrastructure that is intended to exist. What kills this in 12 months isn't competition — it's the Indian power grid failing to deliver reliable capacity at the scale required for GPU-dense AI workloads, which forces a slower ramp and rewrites the unit economics entirely.”
The PM
Product Strategy
“The job-to-be-done for AirTrunk's customers is 'access GPU-dense compute in Asia-Pacific without routing through Singapore or Tokyo at premium latency and cost' — and India is a credible answer to that problem given the developer density and enterprise demand there. The product is incomplete until power and connectivity SLAs are public, because a data center without uptime guarantees isn't a product, it's a building. The strategic opinion embedded in this move — that AI workloads will be regional, not globally centralized — is the right call, and the company that owns the physical layer in India when that demand crystallizes wins something that's very hard to reverse.”