IndiaAI Portal Phase II: 25,000 GPUs Allocated Across Yotta, E2E & AIRAWAT

Scaling National Compute from Pilot to Industrial Scale

When the Union Cabinet approved the ₹10,372 crore IndiaAI Mission in early 2024, the government’s initial target of establishing a 10,000-GPU public computing capacity was viewed as ambitious. However, as hundreds of domestic artificial intelligence startups surged into foundational model training, vernacular voice agents, semiconductor digital twins, and autonomous robotics, domestic compute demand rapidly outpaced initial projections.

During Phase I, the IndiaAI Compute Portal provisioned over 10,000 GPUs across empaneled providers Yotta Infrastructure and E2E Networks, alongside C-DAC’s PARAM supercomputers. Over 280 DPIIT-recognized startups and 45 academic research labs consumed their allotted vouchers within months, demonstrating immense domestic absorption capacity.

On Sunday, October 18, 2026, the Ministry of Electronics and Information Technology (MeitY) officially inaugurated Phase II of the IndiaAI Compute Portal. The expansion scales total national subsidized compute capacity to 25,000 enterprise GPUs, incorporating next-generation NVIDIA Blackwell B200, Hopper H200, and AMD MI325X accelerators to power India’s sovereign technological resurgence.

Phase II Infrastructure Allocation: The Three Core Computing Pillars

The 25,000-GPU Phase II capacity is strategically distributed across three tier-one domestic infrastructure pillars:

1. Yotta Infrastructure (14,000 GPUs): Anchored at Yotta’s NM1 datacenter in Navi Mumbai and D1 in Greater Noida, this tier provides high-density training pods of up to 4,096 interconnected NVIDIA B200 and H200 GPUs linked via 3.2 Tbps non-blocking InfiniBand fabrics. This infrastructure is reserved for large-scale pre-training of indigenous multimodal foundation models (such as BharatGen Phase II) and sovereign defense applications.

2. E2E Networks (7,500 GPUs): Focused on high-efficiency, cost-effective inference serving and fine-tuning. E2E hosts clusters across its Noida and Mohali Tier-IV datacenters, optimizing AMD EPYC and NVIDIA architectures to deliver low-latency containerized endpoints for high-growth consumer fintech and agritech applications.

3. C-DAC AIRAWAT & PARAM Supercomputing Grid (3,500 GPUs): Dedicated strictly to academic researchers, national laboratories (including CSIR, ICAR, and ICMR), and doctoral fellows conducting advanced scientific research in computational genomics, monsoon weather forecasting, and material sciences.

The Streamlined Voucher Disbursement Engine & Automated SLA Scheduling

A common bottleneck in government grant programs is bureaucratic delay in approvals and resource allocation. Phase II of the IndiaAI Compute Portal completely automates the voucher lifecycle through a digital-first self-service platform.

DPIIT-recognized startups apply via the portal by submitting their project architecture, model parameter sizing, and training duration estimates. The application is reviewed by an automated Technical Evaluation Engine that benchmarks requested resources against industry standards.

Upon approval, digital compute vouchers are credited directly to the startup’s portal wallet within 48 hours. Startups select their preferred cloud provider (Yotta, E2E, or C-DAC) and provision GPU instances instantly through standard Kubernetes CLI tools (kubectl) or Terraform providers, with MeitY settling the 50% subsidy directly with the infrastructure vendor.

Priority Strategic Verticals: Agriculture, Vernacular Healthcare, and Defense

Phase II compute allocations are targeted toward four high-impact national strategic verticals:

In agriculture, subsidized compute powers regional AI models that ingest real-time ISRO satellite imagery, soil moisture telemetry, and local weather radar data to deliver predictive hyper-local pest warnings directly to farmers’ mobile phones.

In vernacular healthcare, medical research labs utilize H200 clusters to train multimodal diagnostic models capable of analyzing mobile X-rays, ultrasound scans, and doctor clinical notes in 14 regional languages, expanding specialist healthcare access across tier-2 and tier-3 rural hospitals.

In defense and national cybersecurity, air-gapped confidential computing allocations power sovereign cyber-defense models trained to monitor domestic internet exchange gateways and thwart advanced persistent threat (APT) attacks against critical power grids and banking switches.

SIDBI Venture Debt Co-Investment: The Capital Multiplier

Recognizing that compute access alone cannot guarantee commercial startup survival, MeitY has partnered with the Small Industries Development Bank of India (SIDBI) to launch the ₹2,000 crore IndiaAI Venture Debt Facility.

Startups that successfully receive Phase II compute vouchers become automatically eligible for low-interest, non-dilutive venture debt of up to ₹10 crore to finance technical hiring, customer acquisition, and physical hardware prototypes. This synchronized capital structure eliminates the funding gap that previously forced promising Indian deep-tech startups to seek overseas venture capital and re-incorporate in Delaware or Singapore.

Sovereign Data Boundaries: Mandatory DPDP Compliance at the Hardware Layer

In strict accordance with the Digital Personal Data Protection (DPDP) Act 2023, all 25,000 GPUs provisioned under Phase II operate within certified sovereign datacenters physically located within Indian geographical borders.

Providers are legally barred from routing model traffic through foreign intermediary proxies or offshore management consoles. Hardware-enforced AMD SEV-SNP and NVIDIA Confidential Computing modules ensure that all enterprise training datasets remain encrypted in memory, protecting domestic corporate intellectual property from foreign surveillance.

Global Economic Competitiveness: Lowering Compute Costs for Indian AI Builders

With 50% government voucher subsidies applied, Indian AI startups can rent NVIDIA H200 compute at an effective rate of approximately $1.15 per GPU hour—nearly 60% cheaper than commercial cloud rates in the United States or Western Europe.

This massive cost advantage is turning India into a global magnet for artificial intelligence engineering talent. Multinational corporations and NRI entrepreneurs are establishing R&D centers in Bengaluru, Hyderabad, and Pune to leverage world-class sovereign compute infrastructure at unmatched unit economics.

Enterprise Application Guide: Securing Phase II Compute Allocations

Founders and Chief Technology Officers preparing to apply for IndiaAI Phase II compute vouchers should follow this application guide:

1. Startup Registration Verification: Ensure your corporate entity possesses valid DPIIT startup recognition on the official Startup India portal.

2. Technical Compute Sizing: Calculate your training or inference FLOP requirements using standard Megatron-LM or PyTorch FSDP scaling calculators.

3. Submit Architecture Blueprint: Upload your application on portal.indiaai.gov.in, detailing your model architecture, dataset provenance, and DPDP compliance framework.

4. Instance Provisioning: Upon receiving digital voucher approval, configure your Terraform or Kubernetes scripts to deploy nodes on Yotta or E2E Networks.

5. Quarterly Milestones Reporting: Submit automated benchmark and training checkpoint logs through the portal dashboard to maintain ongoing quarterly voucher renewals.

Automated Multi-Tenant Quota Scheduling & InfiniBand Fabric Telemetry

Managing a national supercomputing pool of 25,000 GPUs across diverse startups and academic institutions requires advanced multi-tenant cluster scheduling. The IndiaAI Phase II portal integrates an automated Slurm and Kubernetes orchestrator that dynamically allocates GPU slices based on model parameter density and training phase SLAs.

During large-scale distributed pre-training runs, non-blocking 3.2 Tbps InfiniBand telemetry monitors inter-GPU all-reduce communication latency in real time. If an optical transceiver degrades or a network link experiences packet retransmissions, the cluster manager automatically pauses the training job, saves an incremental safetensors checkpoint, and isolates the degraded node in under 30 seconds without corrupting model weights.

Financial modeling conducted by the Ministry of Finance confirms that every rupee invested in subsidized sovereign compute under the IndiaAI Mission generates an estimated ₹7.40 in domestic economic output through patent creation, deep-tech employment, and international enterprise software exports. By expanding to 25,000 GPUs, India cements its position as the premier deep-tech innovation hub of the Global South, bridging sovereign technological capability with inclusive grassroots economic empowerment.

Industry leaders across the Confederation of Indian Industry (CII) and NASSCOM have commended MeitY’s rapid execution, highlighting that automated voucher provisioning and subsidized Tier-IV datacenter hosting have removed the single largest financial bottleneck facing Indian software engineering teams. As Phase II clusters go live, Indian startups are projected to launch dozens of sovereign foundation models and domain-specific enterprise agents over the coming two quarters.