Ashwini Vaishnaw Announces AIKosha: What Is It, How Will It Accelerate India's AI Innovation?

Union Minister Ashwini Vaishnaw on Thursday announced AIKosha, India’s first central database designed to train foundational models without bias. The announcement marks the first anniversary of the IndiaAI Mission, which the government launched in 2024 to kick off a nationwide hunt for partnerships to foster India’s self-reliant AI ecosystem. AIKosha is a secure platform that includes a repository of datasets, models, and use cases that will catalyse India’s AI innovation and help achieve the target of becoming the leader in the AI industry.

According to Vaishnaw, the launch of AIKosha will enable AI research and innovation in India by offering an AI sandbox with an integrated development environment, tools, and tutorials. Meanwhile, S. Krishna, secretary of the Ministry of Electronics and Information Technology (MeitY), underscored the importance of the AIKosha in building an AI ecosystem that can be leveraged to realise Prime Minister Narendra Modi’s vision of Viksit Bharat by 2047.

What is AIKosha?

AIKosha is a sort of library that houses hundreds of datasets and over 80 models that will form a base to train foundational models. Think of it like a think tank for AI. The variety of information in these datasets will help large language models think rationally with less scope of including bias in their evaluation of questions. The Device Programming Interface (DPI) for AI, the government said, will ensure “ethically sourced, consent-based datasets” to reduce dependence on “synthetic and foreign data.” However, the minister did not specify how AIKosha will ensure ethically sourced data.

Using AIKosha as a unified platform, companies can build their AI models without having to rely on foreign datasets, such as those with OpenAI, Meta, Google, and Amazon. AIKosha will work in tandem with the new AI Compute Portal to offer AI compute services at “a highly subsidised rate” to support startups, researchers, and enterprises. The platform has features such as content discoverability, AI readiness scoring of datasets, permission-based access, security mechanisms such as data encryption at rest and in motion, secure API, and firewalls for real-time filtering of suspicious traffic.

IndiaAI Data Labs are being set up in tier-II and tier-III cities across India “to impart foundational level courses” to encourage interested people to take up the roles of Data Annotator and Data Curator. These roles will help them focus their AI development in sectors, such as healthcare, education, manufacturing, and agriculture. The courses will be available through NIELIT and ITI centres.

How will it impact India’s bid to become an AI leader?

Acting as a repository with a large quantity of datasets, AIKosha will enable companies to build AI models, which can power a wide range of applications across industries. In other words, companies that want to build AI models can use AIKosha for data training without having to pay high costs to foreign players. These companies, including startups, and researchers can explore the new IndiaAI Compute Portal to use services such as AI Compute, network, storage, platform, and cloud services “at discounted rates.”

The Compute AI portal offers “easy access” to high-end and mid-range GPUs, such as Nvidia’s H100, H200, A100, L40S, and L4; AMD’s MI300x and 325x; Intel’s Graudi 2; and AWS’s Tranium and Inferentia. These GPUs, specifically the ones from Nvidia, are in high demand due to their capability to handle intense workloads when processing data for AI inferences. Vaishnaw said the AI Compute Portal will initially provide access to 10,000 GPUs, while an additional 8,693 GPUs will be added later. The new GPU assortment is a huge jump from the existing GPU bank, consisting of about 700 GPUs, that powers indigenous supercomputers.

The government will offer up to 40 per cent subsidy on AI Compute services on the cloud to allow companies to build their AI models more conveniently. The more companies and researchers build AI models, the less India will depend on foreign companies for the essentials of building an AI ecosystem. However, whether AIKosha will appeal to local AI startups, enterprises, and researchers is uncertain and will be revealed over time. That is because even with 300 datasets, AIKosha falls short of use cases and extensive and domain-specific datasets that industry leaders like OpenAI or Meta can offer. Moreover, the government has not clarified if its subsidy model will be sustained in the long run, especially after companies have set up their AI ecosystem and plan to expand their models.

Read more: Apple's New iPad Has New Chip, More Memory But Lacks Important Feature