Inside Adani’s $100 billion bet on sovereign AI infrastructure

Generative AI is transforming not just software but the physical foundations of computing itself. Training frontier models and running large-scale inference require unprecedented amounts of power, ultra-dense hardware configurations and infrastructure that can operate continuously at extreme loads. As a result, companies worldwide are rushing to build a new class of facilities designed specifically for AI workloads.

India’s data centre market is now undergoing the same structural shift. Over the past year, investment commitments toward AI-focused infrastructure have accelerated, drawing in large domestic conglomerates alongside global technology operators. Among the most ambitious is the Adani Group’s plan to invest $100 billion by 2035 in AI-ready digital infrastructure, one of the largest long-term bets yet on India’s emerging role as an AI compute hub.

Inside the shift to AI-ready data centres

AI-ready facilities differ fundamentally from traditional hyperscale data centres built around CPUs and enterprise workloads. They are designed for GPU-heavy environments optimised for parallel processing, the backbone of modern generative AI systems.

“Unlike legacy environments designed for web services and enterprise workloads, AI-ready data centres prioritise massively parallel processing using GPUs and AI accelerators, supported by dense rack configurations,” says Manish Rawat, semiconductor analyst at TechInsights. “They demand ultra-high intra-cluster bandwidth and low-latency networking, often requiring customised fabric architectures beyond conventional east-west traffic models.”

Cooling represents an equally profound redesign challenge. Conventional air cooling struggles to handle the sustained heat generated by high-density GPU clusters, making liquid cooling systems essential rather than optional.

“Air cooling at scale is not viable at sustained high density. Liquid cooling systems must be designed into the facility, not retrofitted,” says Sanchit Vir Gogia, CEO and chief analyst at Greyhound Research. “That means coolant distribution units, leak detection, water management protocols, and revised maintenance models. Once you move to liquid systems, operational discipline must tighten. A cooling failure in a 6 kW rack is an inconvenience. A cooling failure in a 120 kW rack is a systemic risk.”

Beyond hardware, software orchestration layers are also evolving. AI-ready centres increasingly integrate Kubernetes with GPU operators, distributed training frameworks and MLOps stacks, enabling dynamic scaling across clusters rather than traditional virtual-machine provisioning.

Most existing facilities worldwide were not designed for this level of sustained density. Retrofitting is possible but costly and complex, reinforcing that AI-ready data centres represent a generational shift rather than incremental upgrades.
 
Adani scaling capacity and ambition

Adani’s $100 billion investment focuses on building green-energy-powered AI data centre capacity through AdaniConnex, expanding its current 2-gigawatt footprint toward a target of 5 GW by 2035.

Alongside this, the group has outlined a further $150 billion investment across manufacturing, servers and sovereign cloud services, though detailed deployment plans remain undisclosed.

Biswajeet Mahapatra, principal analyst at Forrester, says the two figures reflect different layers of the emerging AI ecosystem. The initial investment is largely infrastructure-focused, land, power systems, cooling and physical facilities, while the additional spending targets the broader AI value chain, including domestic hardware manufacturing, cloud platforms and services.

Gogia argues the larger figure also signals deep integration with energy systems. “Energy generation tied directly to compute infrastructure reduces exposure to power price volatility and grid instability. Transmission alignment shortens energisation timelines,” he says. However, he cautions that such integration cannot solve all dependencies. “Accelerator availability remains influenced by export controls and geopolitical currents.”

Global partnerships shape early deployments

Despite being framed as a sovereign infrastructure initiative, the rollout relies heavily on collaborations with global technology companies.

Adani has announced plans to work with Google on a gigawatt-scale AI data centre campus in Visakhapatnam, as well as additional campuses in Noida. Partnerships with Microsoft will span Hyderabad and Pune, while existing arrangements with Walmart-owned Flipkart include a second AI-focused facility to support high-performance computing and large-scale digital commerce workloads.

The company says a portion of GPU capacity will be reserved for Indian startups, research institutions and deep-tech firms, aiming to strengthen domestic innovation capabilities.

“Sovereign AI infrastructure does not necessarily mean excluding global players, but ensuring that data, compute, and control reside within national jurisdiction,” Mahapatra says. Even when serving multinational clients, facilities can qualify as sovereign if they comply with local data laws and support strategic domestic use cases.

Closing the compute gap

India’s data centre capacity remains modest compared with global leaders. Ratings agency ICRA estimates worldwide capacity at roughly 42 GW, with the United States accounting for nearly half. India’s operational capacity stood at about 1.2 GW in FY2025 and is projected to reach around 2.5 GW by FY2028, concentrated in major metro hubs including Mumbai, Chennai, Hyderabad, Delhi-NCR and Bengaluru.

Large-scale investments could help narrow the gap and improve data residency compliance while reducing latency for domestic workloads. Yet analysts caution that infrastructure alone does not guarantee technological independence.

“This investment meaningfully strengthens India’s position as a regional hosting and infrastructure hub,” Rawat says. “It localises compute capacity, enhances data residency compliance, and reduces latency for domestic workloads.” However, he notes that if demand continues to be dominated by foreign hyperscalers and AI model providers, India will remain deeply embedded in global cloud ecosystems.

True autonomy would require advances far beyond physical infrastructure, including domestic foundation models, indigenous semiconductor capabilities and homegrown cloud platforms operating at a global scale.

For now, Adani’s massive investment reshapes India’s standing in the digital value chain, but it stops short of delivering full-stack technological sovereignty.