Synopsis
According to Government e-Marketplace (GeM) tender status data, the qualified bidders include Paradigmit Technology Services, Tata Communications, RackBank Datacenters, Netmagic IT Services, E2E Networks, Yotta Data Services, Cyfuture India, Sify Digital Services and UrsaCompute.According to Government e-Marketplace (GeM) tender status data, the qualified bidders include Paradigmit Technology Services, Tata Communications, RackBank Datacenters, Netmagic IT Services, E2E Networks, Yotta Data Services, Cyfuture India, Sify Digital Services and UrsaCompute.
Among these, Yotta Data Services is offering 17,000 Nvidia B300 GPUs. The company said this capacity would support both sovereign AI model training and inference requirements under the IndiaAI Mission, while also meeting the requirements of global clients.
Sunil Gupta, managing director of Yotta, told ET that the company is already building a 21,000-GPU cluster and plans to add another 40,000 GB300 capacity in the current financial year.
Cyfuture India has bid 1,024 Nvidia B200 GPUs in this round. Anuj Bairathi, its chief executive, said the outcome of the commercial bidding will be crucial in setting benchmarks for pricing and infrastructure economics, especially at a time when global DRAM and NAND prices are rising sharply.
Despite these pressures, he said, India is likely to remain among the most cost-competitive markets for AI compute.
RackBank Datacenters has committed a mix of 1,024 B300, B200 and H200 GPUs. However, its chief executive Narendra Sen highlighted mounting challenges in the supply chain.
Sen said the price of DRAM modules has surged to nearly $2,000 from around $300 late last year, driven by geopolitical disruptions and a supply crunch among the three dominant manufacturers: Samsung Electronics, SK Hynix and Micron Technology.
He said that hyperscalers buying in bulk have further squeezed smaller players, making it difficult to secure components at viable prices.
Sen warned that the current one-year contract structure under the IndiaAI Mission is misaligned with the capital-intensive nature of GPU infrastructure, where financing typically spans five to six years.
“Earlier, return on investment could be achieved in about two years; now it may take up to four years,” Sen said, calling for contract durations to be extended to at least two years. This, he argued, would improve financing access, stabilise supply planning and ensure that domestic demand is not deprioritised in favour of higher-paying global customers.
Yotta’s Gupta had similar concerns. He advocated for contract terms of three years or more, pointing out that under the current model, companies must invest in GPUs upfront before securing government contracts, unlike global markets where investments are often backed by pre-signed customer agreements. This increases risk and limits access to funding from banks and private equity firms.
Industry players also flagged pricing pressures within the bidding process.
Gupta cautioned that aggressive bids from smaller players offering limited GPU volumes could force larger providers to match unsustainably low prices, potentially undermining their business models.
IndiaAI Mission did not respond to ET's request for a comment.
The challenges come amid a broader technology transition. Newer Blackwell-series GPUs, such as the B300, are significantly more powerful but also more expensive. They cost as much as $500,000 per system compared with about $200,000 for earlier H100 models. This shift is raising capital requirements for new entrants and intensifying competition.
Despite these headwinds, executives see the IndiaAI Mission as a pivotal step in building sovereign AI capacity.
Bairathi said the initiative could position India as a global hub for affordable, large-scale AI compute, balancing cost efficiency with growing infrastructure scale.
However, industry leaders caution that without policy adjustments, particularly longer contract durations, flexible supply timelines and revised pricing frameworks, the sustainability of the ecosystem could be at risk even as demand for AI compute continues to surge.
