Sunil Gupta, Chief Executive of Yotta Data Services, which has placed a purchase order of 16,000 H100 GPUs from Nvidia and went live with 4,000 H100 GPUs a few months ago, concurred. The Government has halved the revenue threshold for companies seeking to bid for the Rs 10,000 crores graphics processing units (GPUs) tender, conceding a key demand of the industry that will allow even small enterprises to participate. The threshold has been cut to Rs 50 crores from Rs 100 crores, the Ministry of Electronics and Information Technology (MeitY) said in a clarification issued after heeding suggestions from several small cloud companies. The earlier threshold would have required smaller companies to partner with others, some in the industry had complained. “The relaxations on revenue qualifications have allowed us to qualify directly, and we deeply appreciate this flexibility,” Tarun Dua, Chief Executive of Delhi-based E2E Networks, an artificial intelligence (AI) cloud service provider, told ET. However, he said, there are several new clauses that necessitate additional clarifications. As per the corrigendum, the bidder or primary partner must have had annual turnover of more than Rs 50 crores on an average for the past three financial years, or since registration in case less than three years have elapsed since registration.
“By lowering turnover thresholds and introducing greater flexibility in AI compute infrastructure requirements, the Government has made it easier for a diverse range of players including and especially the startups in the AI services space to contribute to this mission.” During the pre-bid meeting, companies had suggested that producing the confirmed purchase order of 1,000 GPUs should be waived and instead a letter of intent should be considered. As per the corrigendum, the bidder can now produce the purchase order not at the time of submitting the bid but rather three months after becoming the lowest bidder. However, the MeitY is now asking for a bank guarantee of Rs 50-100 crores (Rs 1 crore for 10 GPUs) in lieu of the purchase order, and if the purchase order is not submitted within three months of the empanelment, then the bank guarantee amount would be forfeited. This may be a big loss to the bidder given the fact that the GPU supply chain is still very constrained and GPU supplies are mostly controlled by a single vendor,” said Sumant Parimal, founding partner, Innogress Ventures, a tech advisory firm. “Such type of risk can only be taken by very large bidders, maybe by those having Rs 200-500 crores of turnover.” “Essentially with this lowering of the turnover criterion, a few more bidders may be eligible to participate in the AI mission bid, but not many startups will get qualified,” Parimal said. This bid is not for SMEs (small and medium) and tech startups. It is essentially suited for large corporations and Big Tech.” Gupta, however, said that by giving extra scores to those who took early business risks and created an existing large base of GPUs, the government is rewarding entrepreneurship in AI infrastructure, which is critical for achieving India’s AI goals. He was referring to a new clause which said one mark will be awarded for every 20 AI compute units above 50, up to a maximum of 15 marks. If half of the installed capacity constitutes AI compute units with double precision (FP64) capabilities, five bonus marks will be awarded over and above the 15 marks allocated under this criterion.
“Last year, we took early and bold bets by placing large orders for Nvidia chips and built an advanced computing infrastructure with a full stack of AI platform software capabilities,” Gupta said. “The focus on Make in India underscores the Government’s commitment to self-reliance and sovereignty in the overall AI value-chain, and not just data localisation, which will undoubtedly drive growth in India’s tech manufacturing sector.” Top chip companies, cloud makers and cybersecurity firms have shown interest in the Government’s Rs 10,000 crores GPUs procurement tender, which is central to India’s AI Mission. Nvidia, Intel, Advanced Micro Devices, Qualcomm, Microsoft Azure, Amazon Web Services, Google Cloud and Palo Alto Networks were among the scores of companies that attended a pre-bid meeting on August 29. Also present at the meeting were representatives from the likes of Yotta Data Services, L&T, E2E Networks, Tata Communications, CtrlS, ST Telemedia Global Data Centres, NTT Data, Dell and Vigyanlabs. The massive interest in the tender is because commercial benefits from the domestic GPU cloud market makes it a lucrative business. There will be tremendous demand eventually for these GPUs and companies today are trying to leverage this, they said. ET had reported on September 11 that small and medium companies were scrambling to strike partnerships among themselves as well as with cloud service providers and large enterprises to build consortiums so that they could participate in the tender. As per another clause introduced on Friday, bidders will have to adhere to the Make in India initiative guidelines of the Department for Promotion of Industry and Trade, Ministry of Commerce and Industry and MeitY. Yet another new clause pertains to power usage effectiveness (PUE), a metric that measures how efficiently a data center uses energy. It is calculated by dividing the total amount of power used by a data center by the amount of power used by its computing equipment. A lower PUE ratio indicates better energy efficiency. As per the new clause, a bidder must confirm that the PUE for the data centers from which AI services on cloud would be delivered is not more than 1.35. This could be a criterion to eliminate several players, said a cloud company executive, who did not wish to be identified. Another executive from a data center said on the condition of anonymity, “This time they have given importance to PUE which is a great step when it comes to sustainability.” While Nvidia was not immediately available for a comment, an executive from Tata Communications confirmed that the company had received the corrigendum from MeitY and was studying it. The last date for the submission of bids has been extended to October 16 from September 25.
