The Union Government has asked for bids under the India artificial intelligence (AI) mission budgeted at Rs 10,372 crore for the empanelment of entities for AI services on the cloud. The empaneled entities will provide high-speed AI infrastructure like graphics processing units (GPUs), accelerators, tensor processing units (TPUs), storage, etc. to different stakeholders at minimum possible rates, which include academia, startups, researchers, and government bodies. The rates would be decided by bidding.
MeitY, Additional Secretary Abhishek Singh shared on social media, "With an objective to empanel service providers for making this possible, IndiaAI, an IBD under Digital India Corporation, MeitY has issued an RFE (request for empanelment)."
The aim of the India AI mission is to build supercomputing infrastructure, which consists of 10,000 GPUs. There is an increase in demand for GPUs after the advent of AI because the earlier CPU based servers are not as effective now. The expense of utilising GPU-based cloud solutions in India is anticipated to be twice as high in comparison to the costs in nations that have implemented a large number of GPUs.
As stated in the tender document issued by the Ministry of Electronics and IT (MeitY), selected agencies will be accountable for delivering high-performance computing, networking, and storage solutions that are vital for executing the initiatives outlined under the IndiaAI Mission, including the creation of domestic LLMs (large language models similar to ChatGPT) and AI applications.
"IndiaAI will authorise the end users permitted to access these cloud-based AI services, guaranteeing that the services are offered at rates established through a clear and competitive process. The selected AI service providers will be contracted for a duration of 36 months, with the potential for extension based on mutually accepted conditions," the document noted.
The agencies qualified to submit bids will be required to provide 1,000 AI computing units via their cloud service platform, offering AI services in the cloud within six months from the date of signing the agreement with IndiaAI.
Each computing unit must possess a minimum processing speed of 15 TFLOPS (TeraFLOPS), which indicates the capability to perform 15 trillion floating-point calculations per second. This necessary performance speed is currently being primarily attained through GPUs or Google's TPUs.
