India Can Build Five GPT-4 Models Simultaneously on Yotta Infrastructure

Hiranandani Group's data centre firm Yotta Data Services’ chief Sunil Gupta, in a recent interview with Forbes, said that India will be able to build five GPT-4 models simultaneously using their existing infrastructure. 

 "I have ordered 16,000 [GPUs], so if there is a customer or there are like five customers, each one of them who wants to make a GPT-4, I can simultaneously handle their load," said Gupta. 

Yotta Data Services has received the first tranche of 4,000 GPUs from NVIDIA, with plans to scale up to 32,768 units by the end of 2025. 

This expansion is part of their partnership announced in December last year, where Yotta ordered 4,096 NVIDIA H100 Tensor Core GPUs initially, set to increase to 16,384 GPUs by June 2024, marking a significant investment of close to $1 billion.

The state-of-the-art chips procured from NVIDIA will power Yotta’s upcoming Shakti Cloud platform, positioning it as the 10th quickest supercomputer globally.

In addition to enhancing their GPU capabilities, Yotta Data Services partnered with Deloitte India to offer clients access to NVIDIA GPU computing infrastructure for developing innovative Generative AI applications efficiently.

Moreover, Yotta Data Services recently announced a partnership with Nepal's BLC Holdings to construct Nepal's first supercloud data center, named 'K1', in Ramkot near Kathmandu. This supercloud data center will provide a comprehensive suite of cloud, managed IT, and cybersecurity services, catering to various use cases, including AI models and enterprise applications.

The multi-million K1 facility will offer up to 4MW critical IT load capacity, spread across 3 acres and 60,000 sq ft area. Located within 20 km of Kathmandu airport, it will provide cloud, managed IT and cybersecurity services to store and process data, AI models and enterprise applications.