Vultr Plans 20MW AMD MI355X Cluster in Chennai

Chennai is getting a 20-megawatt cluster of AMD’s current flagship accelerators. Vultr said on Wednesday that it will deploy AMD Instinct MI355X GPUs at a new supercluster in the city, according to a release on Business Wire. The cloud provider pitches it at Indian organisations that want to train and run AI models without sending data abroad.

Vultr says Chennai will be India’s biggest AMD Instinct site

Vultr, which calls itself the world’s largest privately held cloud infrastructure company, says the site will be India’s largest AMD Instinct deployment. It also says the site will give it two of the world’s largest AMD AI deployments, alongside its flagship supercluster in Springfield, Ohio. Each MI355X carries 288GB of HBM3E memory and 8TB/s of memory bandwidth, the release says.

The intended customers are AI labs and foundation model builders, AI-native start-ups, enterprises, and research and academic institutions. Vultr lists large language model and foundation model training, fine-tuning and high-throughput inference as the target workloads. The cluster will run on AMD’s ROCm open software stack.

Both companies frame it as in-country compute

The pitch is about where the data stays. “Enterprises deploying AI at scale need infrastructure they can access locally to meet data sovereignty, data residency, and latency requirements,” said J.J. Kardwell, Vultr’s chief executive. He said the Chennai site lets organisations in India “train, fine-tune, and run inference on in-country infrastructure”.

AMD made a similar point. “India has the talent and ideas to build the next generation of AI. Access to compute at this scale will help turn more of those ideas into reality,” said Jaya Jagadish, AMD India’s country head and senior vice-president of engineering.

Vultr says it runs 34 cloud data centre regions in 17 countries across six continents. It says it has hundreds of thousands of active customers in 185 countries. It announced an equity round at a $3.5bn valuation in December 2024.

The release leaves out the numbers that set scale

Several basics are missing. The release gives no GPU count, no investment figure and no date for the cluster to start running. It also doesn’t say whether 20MW means the power available to the IT equipment or the whole facility’s draw, and it names no customers. Until those details appear, the “largest” claim is hard to measure against other Indian AI capacity, though it is plausibly the biggest AMD-only site.

The deal fits a longer partnership. Vultr has used AMD Instinct GPUs from the MI300X through the MI325X to the MI355X. Its VX1 Cloud Compute service runs on AMD EPYC processors.

The MI455X and Helios are pencilled in for 2027

Vultr also says it plans “significant additional capacity expansion in 2027”, including MI455X GPUs and AMD’s Helios rack-scale architecture. It doesn’t say where that capacity will go or whether any of it will be in Chennai. TSN covered an earlier MI355X build in Europe when Scaleway announced a cluster in Stockholm, and sovereign-cloud demand is also behind Microsoft’s latest sovereign cloud update.

The first useful number from Chennai will be a GPU count. Until Vultr gives one, 20MW is all there is to judge the site by.