India Eases Entry Norms for GPU Bidders, but High Costs and Pricing Risks Remain

The Indian government has relaxed eligibility requirements for companies seeking to supply 10,000 additional graphics processing units (GPUs) under the IndiaAI Mission. The revised framework allows providers to qualify with a minimum capacity of 512 GPUs, tensor processing units (TPUs) or other AI accelerators at a single data-centre location in India, compared with the earlier requirement of 1,000 units.

The changes are intended to attract more companies into the procurement process and expand the country’s artificial intelligence computing infrastructure. However, lower entry barriers do not resolve the industry’s biggest concerns. Rising memory and chip prices, uncertain demand for computing capacity and pressure to quote competitive prices continue to make investments difficult for potential suppliers.

Government Lowers Minimum GPU Requirement to 512

Under the revised procurement framework, companies can qualify with at least 512 AI computing units at a single data-centre location in India.

The earlier framework required providers to demonstrate a minimum capacity of 1,000 GPUs.

The government has also broadened the eligible hardware definition to include TPUs and other processors designed for AI workloads. This gives companies greater flexibility in the type of computing infrastructure they can offer.

The revised requirements are part of the government’s effort to procure another 10,000 AI computing units under the IndiaAI Mission.

The move could allow smaller infrastructure providers and companies with specialised AI hardware to participate in a programme that requires substantial investment.

Technical Qualification Will Become Easier

Another significant change concerns the bidding process itself.

Under the revised framework, the first stage will focus on technical qualification. Companies will not be required to submit financial bids or commit to service-level terms at this stage.

The minimum technical qualifying score has also been reduced to 60 out of 100, compared with 75 under the earlier IndiaAI framework.

These changes are intended to separate the technical evaluation from commercial negotiations.

The revised process can be understood through two stages:

Stage	Main requirement
Stage 1: Technical qualification	Evaluation of infrastructure and technical capabilities, with a minimum score of 60
Stage 2: Commercial bidding	Financial bids and commercial terms from qualified providers

This approach could make it easier for companies to demonstrate their ability to provide AI computing services without immediately committing to pricing and service conditions.

It may also give the government a broader pool of technically qualified providers before comparing commercial offers.

India Wants to Procure 10,000 Additional AI GPUs

The latest procurement exercise is part of India’s broader effort to expand domestic AI computing capacity.

The government has been working to make high-performance computing resources available to startups, researchers, universities, enterprises and public institutions through the IndiaAI Mission.

The new procurement round seeks access to up to 10,000 additional AI computing units, with a five-year supply horizon.

The government is not necessarily purchasing every GPU upfront. Instead, the framework is designed to secure access to computing capacity over an extended period, allowing users to obtain processing resources as required.

This is important because advanced GPUs are expensive and difficult for smaller companies to purchase independently.

By making shared computing infrastructure available, the government aims to reduce the cost of developing and training AI models in India.

The programme is particularly relevant for startups developing large language models, generative AI applications and other computing-intensive products.

Why GPU Prices Remain a Major Concern

Despite the relaxed eligibility requirements, the cost of acquiring and operating advanced AI hardware remains a significant challenge.

The global AI infrastructure boom has increased demand for GPUs, high-bandwidth memory, networking equipment and data-centre components.

Large technology companies are making substantial investments in AI infrastructure, competing for limited supplies of advanced chips and memory components.

This has increased the financial burden on companies looking to build GPU clusters.

Industry executives have previously highlighted sharp increases in memory prices. In April 2026, executives from Indian data-centre companies told The Economic Times that DRAM module prices had risen dramatically, creating difficulties for infrastructure providers trying to secure equipment at commercially viable rates.

Newer GPU systems also require larger upfront investments than earlier generations.

For example, industry estimates cited in the same report placed advanced Nvidia Blackwell systems at prices as high as $500,000, compared with around $200,000 for earlier H100 systems.

These figures illustrate why reducing the minimum qualifying capacity does not necessarily make the overall investment inexpensive.

Companies still need to finance expensive hardware, data-centre facilities, electricity, cooling systems and networking infrastructure.

Uncertain Demand Creates Another Problem for Suppliers

Infrastructure providers also need greater certainty about how much computing capacity customers will actually use.

A company investing in hundreds or thousands of GPUs must recover its costs over several years.

If demand remains below expectations, expensive computing equipment could remain underutilised, affecting returns on investment.

The concern is particularly relevant to India’s AI ecosystem, where demand for advanced computing resources is growing but may not always be sufficient to absorb the capacity being developed.

The earlier IndiaAI procurement rounds have also highlighted the difference between installing GPUs and ensuring their commercial utilisation.

Industry participants have argued that longer-term commitments and clearer usage guarantees would make it easier to obtain financing and plan infrastructure investments.

The new procurement framework’s five-year horizon is intended to address some of these concerns. However, suppliers still want greater clarity about actual demand and the commercial terms under which their computing capacity will be used.

Aggressive Pricing Could Put Pressure on Profitability

Pricing competition is another concern for companies participating in government GPU tenders.

When multiple providers compete for the same procurement opportunity, some may quote extremely low prices to improve their chances of securing contracts.

While lower prices can benefit the government and users of AI computing services, they can also create problems for infrastructure providers.

Companies that quote below sustainable operating costs may struggle to recover their hardware investments, particularly when chip prices, electricity expenses and financing costs are rising.

Established providers with larger GPU clusters may also face pressure to match low bids from smaller competitors.

The earlier IndiaAI procurement rounds discovered an average lowest bid, or L1, rate of approximately ₹115 per GPU hour, according to government figures and industry reports.

However, pricing varies significantly depending on the type of GPU, its computing performance and the services included.

The challenge is to find a price that makes AI computing affordable for startups while allowing suppliers to operate profitably.

IndiaAI Mission’s Importance for Startups and Researchers

The expansion of shared computing infrastructure could help address one of the biggest barriers facing India’s AI developers.

Training advanced AI models requires substantial computing resources, which are often expensive for startups and academic institutions to access independently.

Through the IndiaAI Mission, eligible users can access subsidised computing capacity instead of investing in their own large GPU clusters.

The government has reported that the mission’s shared computing capacity had expanded beyond 45,000 GPUs by June 2026.

By August 2026, 237 projects had accessed subsidised AI computing, covering more than 93 lakh GPU hours.

Additional capacity could support:

• AI startups: Lower barriers to developing and training AI products.
• Universities: Greater access to computing resources for research and experimentation.
• Researchers: More capacity for testing advanced models and algorithms.
• Enterprises: Access to infrastructure for developing and deploying AI applications.
• Public institutions: Computing resources for government-focused AI projects.

The availability of more computing capacity could also encourage the development of AI models tailored to Indian languages and local requirements.

The Bigger Picture

India’s GPU procurement challenge reflects a larger problem facing countries trying to build domestic artificial intelligence capabilities.

Access to advanced computing hardware has become a strategic requirement, but building the infrastructure requires enormous capital investment.

The government’s decision to lower entry requirements is an attempt to widen participation and attract additional suppliers. However, the long-term success of the programme will depend on balancing affordable computing prices with commercially sustainable returns for infrastructure providers.

If suppliers cannot recover their investments, expanding GPU capacity could become difficult even when technical eligibility requirements are relaxed.

Looking Ahead

The next important step will be the technical evaluation of participating companies, followed by commercial bidding and the finalisation of supply arrangements. The government will need to ensure that the revised process attracts enough qualified providers while maintaining realistic pricing and reliable service commitments. The availability of advanced hardware and the cost of memory components will remain important factors in determining how quickly new capacity can be deployed.

For India’s AI ecosystem, the additional 10,000 computing units could improve access to advanced infrastructure for startups, researchers and businesses. However, the programme’s effectiveness will depend on actual utilisation, affordable GPU-hour pricing and long-term supplier economics. Lowering entry barriers is an important step, but making AI computing affordable and financially sustainable remains the bigger challenge.