Bengaluru: Data centre and AI infrastructure provider Yotta Data Services plans to invest an additional $6 billion to expand its artificial intelligence infrastructure, significantly increasing its deployment of Nvidia GPUs as demand surges from both Indian and global customers.The company, which had earlier announced plans to deploy 20,000 Nvidia Blackwell GPUs, has now raised that target to 30,000. Nvidia's Blackwell architecture is designed for accelerated AI training and inference workloads. Yotta will also bring in 8,000 Nvidia B200 GPUs over the next month and is evaluating the deployment of another 36,000-37,000 next-generation GB300 or Vera Rubin GPUs next year."The demand has been so high, not only from Indian customers but also global customers, that we are extending the 20,000 GPUs to 30,000 GPUs," Yotta managing director and CEO Sunil Gupta told TOI.According to Gupta, the initial deployment of 30,000 Blackwell GPUs will require investments of about $3 billion, while the proposed deployment of an additional 36,000 GPUs could cost another $4 billion. The 8,000 B200 GPUs represent a separate investment of around $600 million.Yotta expects the first batch of 8,000 B200 GPUs to go live within a month. The initial 20,000 Blackwell GPUs are scheduled to become operational by September, while the additional 10,000 units are expected by November. The larger GB300 or Vera Rubin deployment is targeted for May next year.The expansion reflects India's growing ambition to emerge as a global AI compute hub as countries race to secure the advanced computing infrastructure needed to train and deploy AI models."India will have to become self-sufficient across the AI value chain. You cannot be dependent on a single company or a single country that can simply pull a kill switch," Gupta said.While Yotta is not building foundation models, it aims to provide the infrastructure layer powering sovereign AI initiatives and global AI companies. The company operates a 2-gigawatt campus in Mumbai and a 250-megawatt facility in Delhi and currently supports several government workloads through National Informatics Centre (NIC) data centres.Yotta is also increasingly serving overseas customers. About 75% of its order book now comes from international clients, with several US and European AI companies already using its GPU infrastructure for model training and deployment.The economics of AI infrastructure are significantly different from those of traditional data centres, Gupta said. Conventional facilities typically require about $6 million in capital expenditure per megawatt, but adding GPUs raises that figure to nearly $40 million per megawatt.To fund the expansion, Yotta has adopted an "off-book financing" model under which global investors own the GPU assets while the company operates the infrastructure and shares revenue generated from long-term customer contracts.The model is supported by long-term take-or-pay agreements under which customers commit to fixed GPU capacity for four to five years. Customers pay for reserved capacity regardless of utilisation, giving Yotta predictable revenue streams and making it easier to secure financing."All these GPUs are already pre-contracted by customers. I may deliver them after four months, but they are already contracted," Gupta added.Yotta is also preparing for a public listing. While media reports have pegged the IPO size at around $900 million, Gupta said the overall fundraising could exceed $1.5 billion, with a substantial portion expected to come from pre-IPO institutional investors.