Indian Government to Source AI GPUs for Startups, Plans Deal with NVIDIA for AI-Capable GPUs

The Indian government is reportedly in talks with chipmaker giant NVIDIA to secure a deal for AI-capable graphics processing units (GPUs) aimed explicitly at empowering Indian startups. This initiative aligns with India's growing ambitions in Artificial Intelligence. To boost India's AI infrastructure, the country is looking to partner with chipmaker NVIDIA to purchase its graphics processing units (GPUs) and Neural Processing Units (NPUs) and offer them at reduced rates to local startups, researchers, academic institutions, and others. This effort will cost India around ₹10,000 crore and is still in its early phases. 

NVIDIA owns a substantial chunk of the GPU industry

The government aims to provide these high-performance GPUs to local startups, researchers, academic institutions, and other users at subsidised rates. This would significantly reduce the financial barrier for these entities to acquire cutting-edge technology crucial for AI development. NVIDIA owns a substantial chunk of the GPU industry and has become the obvious choice for India's AI computing infrastructure requirements. Globally, nations and enterprises investing in AI computer infrastructure are lined up to buy NVIDIA GPUs, particularly H100 processors.

Providing AI compute-capable NPUs and GPUs

Specifically, there are two models: "rent-and-sublet" and marketplace. As a leading manufacturer of AI-specific GPUs, NVIDIA's involvement would ensure access to advanced hardware like the NVIDIA H100 and Blackwell series. These GPUs are specifically designed to handle complex AI workloads. In the rent-and-sublet method, the Ministry of Electronics and Information Technology would purchase GPUs from Nvidia. The government would encourage businesses to negotiate leasing or sublease agreements directly with Nvidia using the marketplace approach. Incentives will be offered depending on the increased productivity obtained while utilising the GPU, similar to the methodology used in India's PLI plan.

GPUs are relatively inexpensive, making acquisition a big barrier for businesses, particularly startups. NVIDIA's H100 GPUs often sell for $50,000 each, while its current Blackwell cards cost around $40,000.

Even the most basic AI data centre, which can run existing AI models, requires at least 100 to 300 GPUs. We need 5000-10,000 GPUs to construct, develop, and train huge language models.

OpenAI, for example, appears to have employed 720,000 H100 GPUs valued at $36 billion to construct and train Sora, their AI video creator. Similarly, Meta is working with around 350,000 H100 units valued at $17.5 billion. Meta intends to acquire more of these GPUs.

Equipping startups with powerful GPUs

This could lead to a surge in innovation and accelerate India's startup ecosystem in the AI domain. To put things into perspective, India's supercomputer 'AIRAWAT,' based at the Centre for Development of Advanced Computing (C-DAC) in Pune, is the 75th fastest supercomputer in the world and has 640 GPUs. However, compared to the world's fastest supercomputers, which have over 30,000 GPUs, India has to close a significant gap to remain competitive in AI research and development.

In March, the Union Cabinet allocated Rs 10,372 crore for India's AI Mission to deploy 10,000 GPUs through public-private partnerships. Under the rent-and-sublet concept, the government intends to supply GPUs at a discount to qualified entrepreneurs, researchers, and others.

Alternatively, a marketplace approach is being studied, in which performance may be objectively monitored and rewards provided correspondingly, similar to a production-linked incentive (PLI) program.

Globally, both businesses and governments purchase GPUs

However, allocating such limited resources is unlikely to prioritise any country based on its federal system. As a result, India is investigating alternative solutions and consulting with stakeholders to optimise its strategy.

Demand for GPUs has driven Nvidia's market value to $2.16 trillion, making it the world's third most valuable corporation, after only Microsoft and Apple. Indian enterprises such as the Reliance Group, Tata Group, and Yotta Infrastructure have already signed agreements with Nvidia to purchase GPUs for domestic usage. AI companies and renowned CEOs in India have advocated for government investment in computing power to boost the country's competitiveness in the global AI space.

India's AI mission aspires to promote innovation and protect the sovereignty of Indian data. Because GPUs are in short supply domestically, several businesses have turned to cloud resources based elsewhere. The final decision on the deal is yet to be made and is likely to happen after the 2024 general elections.