The Indian government has lowered the minimum artificial intelligence (AI) compute capacity requirement for companies seeking to participate in the next IndiaAI Mission procurement round, reducing the threshold from 1,000 units to 512 units at a single data centre location in India. The change is intended to widen participation among infrastructure providers and accelerate the addition of up to 10,000 AI compute units to the country’s shared computing pool.
The revised requirement covers GPUs, tensor processing units (TPUs) and other specialised hardware used for AI workloads. The move comes as demand for computing infrastructure increases and prices of advanced chips remain under pressure. However, the lower capacity threshold does not remove other financial, technical and operational eligibility requirements, meaning smaller companies may still face significant barriers to entry.
IndiaAI Mission Lowers Minimum Capacity Requirement by 50%
Under the earlier procurement framework, companies were required to demonstrate at least 1,000 AI compute units at a single Indian data centre location. The revised threshold has been reduced to 512 units, effectively cutting the minimum requirement by almost half.
The change is part of a new procurement exercise being conducted by the Centre for Development of Advanced Computing (C-DAC), which implements the compute pillar of the IndiaAI Mission under the Ministry of Electronics and Information Technology (MeitY).
The latest tender seeks to empanel providers capable of supplying up to 10,000 additional AI compute units. These resources are intended to support AI model training, fine-tuning, inference, high-performance computing and generative AI applications.
| Procurement detail | Revised requirement |
|---|---|
| Minimum compute capacity | 512 units |
| Previous minimum capacity | 1,000 units |
| Capacity procurement target | Up to 10,000 units |
| Eligible hardware | GPUs, TPUs and other AI accelerators |
| Location requirement | Single data centre in India |
| Implementing agency | C-DAC |
| Service period | Five years |
| Bid submission deadline | November 13, 2026 |
The revised threshold is expected to make the tender accessible to a wider group of infrastructure operators, although the actual number of new participants will depend on their ability to satisfy the remaining eligibility conditions.
Why the Government Is Expanding Participation
Access to affordable computing power has become one of the biggest requirements for India’s growing AI ecosystem. Training large language models and operating advanced AI applications require specialised processors, high-speed networking, storage infrastructure and reliable data centre operations.
For startups and research institutions, purchasing and maintaining large GPU clusters independently can be expensive. The IndiaAI Mission aims to address this challenge by making computing resources available through empanelled service providers at subsidised rates.
The government has already built a shared compute ecosystem through earlier procurement rounds. In an August 2026 parliamentary update, MeitY said 15 compute service providers had been empanelled across four rounds. It also reported that 237 projects had been approved for subsidised compute support, with 93.18 lakh GPU hours sanctioned.
Reducing the minimum capacity requirement could bring additional providers into the ecosystem, potentially increasing competition and the range of infrastructure available to users.
New Tender Seeks Up to 10,000 AI Compute Units
The latest procurement exercise is designed to add up to 10,000 AI compute units to India’s shared AI infrastructure. The tender allows providers to offer capacity in clusters of 4,000, 1,000 or 512 units, with each cluster located at a single data centre in India.
The tender identifies high-performance accelerators such as NVIDIA’s H100 and H200, along with later-generation or equivalent hardware. The final hardware configuration will be determined during the subsequent procurement stage.
The infrastructure is expected to support several categories of users, including:
- Central government ministries and departments.
- Universities and academic institutions.
- Scientific research organisations.
- Indian AI startups and technology companies.
- Businesses developing generative AI applications.
- Organisations training and fine-tuning large AI models.
The service requirement also extends beyond supplying processors. Providers must deliver an integrated computing service covering networking, storage, orchestration, monitoring and round-the-clock operations.
Lower GPU Requirement Does Not Mean Easy Entry
Although the government has reduced the minimum capacity requirement, companies must still meet substantial financial and technical conditions.
According to the revised tender details, eligible bidders must be Indian companies, limited liability partnerships or partnerships. They must demonstrate an average annual turnover of at least ₹120 crore across FY2023-24, FY2024-25 and FY2025-26, along with positive net worth as of March 31, 2026.
Other important requirements include:
| Eligibility condition | Requirement |
|---|---|
| Average annual turnover | At least ₹120 crore |
| Net worth | Positive as of March 31, 2026 |
| Data centre | Tier-III or Rated-3 facility in India |
| Security certification | ISO/IEC 27001 |
| Minimum compute capacity | 512 units |
| Operations | India-based network and security operations centres |
| Technical qualification score | At least 60 out of 100 |
Companies can demonstrate existing installed capacity or submit an OEM-accepted purchase order for hardware that must be deployed within six months. The purchase-order route is backed by a ₹10 crore bank guarantee, which can be forfeited if the committed hardware is not delivered as required.
The tender also requires providers to maintain control over service delivery rather than simply acting as intermediaries or resellers. Data and control systems, including relevant metadata, logs, backups and disaster-recovery arrangements, must remain in India under the stated data-sovereignty conditions.
These requirements mean the lower capacity threshold primarily widens the opportunity for providers that already have substantial financial resources and operational capabilities.
Two-Stage Procurement Process
The new tender introduces a two-stage process for selecting service providers.
The first stage focuses on technical qualification rather than price competition. Applicants are evaluated against technical, financial, infrastructure and operational criteria, with a minimum qualifying score of 60 out of 100.
Financial bids and detailed service-level commitments are not required at this initial qualification stage. Companies that qualify can participate in the second stage, where pricing, quantities, locations and service conditions will be determined through procurement rounds on the Government e-Marketplace (GeM).
Importantly, empanelment does not guarantee that a company will receive a contract or business. Providers relying on a purchase order must have the required capacity operational before they can win work in the subsequent stage.
The revised tender timetable sets November 13, 2026, as the bid submission deadline, with technical bids scheduled to open on November 14.
Chip Prices and Commercial Risks Remain a Concern
The expansion of India’s AI compute capacity comes at a time when advanced accelerator hardware is expensive and global demand remains strong. Data centre operators must also account for electricity, cooling, networking, maintenance and financing costs.
Some industry executives have raised concerns about unclear usage commitments and the possibility of aggressive pricing by competing vendors. These issues matter because providers may need to make substantial infrastructure investments without certainty about how much capacity they will ultimately supply or how frequently it will be used.
A five-year service commitment also exposes providers to changes in hardware prices, technology generations and operating costs. Equipment purchased today may face competition from newer accelerators during the contract period.
For the government, the challenge will be to balance competitive pricing with sustainable economics for service providers. If bids are priced too aggressively, providers could face pressure to maintain service quality and invest in additional capacity.
For AI startups and researchers, the benefit of the procurement will ultimately depend on how quickly additional computing resources become operational and whether they are available at affordable prices.
The Bigger Picture
The revised IndiaAI tender reflects India’s broader effort to build domestic AI infrastructure rather than relying entirely on computing resources located overseas. Affordable access to advanced processors is important for developing indigenous AI models, supporting research and encouraging startups to build applications for Indian users.
Lowering the minimum capacity requirement could increase the number of infrastructure providers competing for government-backed opportunities. However, the high turnover threshold, data centre requirements and financial guarantees show that the government is still prioritising providers capable of delivering large-scale, secure and reliable services.
Looking Ahead
The immediate milestone will be the technical evaluation of bids submitted by November 13, followed by the identification of companies eligible to participate in the next procurement stage. The outcome will indicate whether the revised 512-unit threshold attracts additional providers and creates stronger competition in India’s AI compute market.
The more important test will be the actual deployment of new capacity. For India’s AI startups, universities and research institutions, the success of the programme will depend not simply on the number of GPUs announced, but on how quickly they become available, how much they cost and whether users can access the computing resources needed to develop and operate AI systems.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
