U.S.-based cloud infrastructure provider Vultr launched a 20-megawatt (MW) artificial intelligence supercluster in Chennai on 2 November 2026. The facility, powered by AMD Instinct MI355X GPUs, is expected to begin with around 10,000 GPUs and is Vultr’s largest AMD Instinct deployment in India.
Computing capacity and uses
The supercluster is designed for large language model and foundation model training, fine-tuning and high-throughput inference. Its intended users include AI labs, startups, enterprises, research institutions and universities. The facility will use AMD’s ROCm open software ecosystem, which gives developers access to AI frameworks, libraries and models.
The Tamil Nadu government has enabled 150 MW of connected power at the location, with 20 MW available for the initial deployment. A substantial portion of the Chennai facility’s capacity has already been pre-reserved by global enterprises.
Expansion and employment plans
Vultr plans to add AMD’s MI455X GPUs and Helios rack-scale AI architecture in 2027. It is targeting 600 MW of AI infrastructure capacity in India by 2028 and gigawatt scale by 2029, but has not committed to firm timelines for these targets.
The company has hired more than 100 people in India over the past three months and anticipates thousands of additional hires in network engineering and operations by March. Its Chennai workforce could eventually reach 5,000. Under a memorandum of understanding with StartupTN, eligible startups across Tamil Nadu can receive up to $45,000 in cloud-computing credits and preferential pricing.
Important Facts for Exams
- ROCm programming models: ROCm offers HIP for GPU-kernel programming, OpenMP for directive-based programming and OpenCL as alternative programming models for GPU computing.
- GPT pretraining sequence: GPT-style large language models are pretrained to predict the next word on extensive data before being fine-tuned.
- MI355X architecture: AMD's MI350-series accelerators, including MI355X, use its CDNA 4 architecture, a GPU compute design built for data-centre AI and high-performance computing.
- Instinct interconnect: AMD's Infinity Fabric links connect GPUs in some multi-accelerator Instinct platforms, enabling high-bandwidth peer-to-peer communication for distributed workloads.
