Vultr and AMD Plan a 20 MW Instinct MI355X Supercluster in Chennai, India's Largest AMD AI Deployment

The US cloud provider Vultr has announced plans to deploy AMD Instinct MI355X GPUs in a new 20-megawatt AI supercluster in Chennai. The company says it will be the largest AMD Instinct deployment in India. The announcement, made on 20 October 2026 as part of a wider Vultr–AMD collaboration, adds another large private GPU cluster to India's fast-growing AI infrastructure. It also gives Indian AI developers a high-end alternative to the Nvidia hardware behind most of the country's compute today.

What was announced

According to the companies' joint release:

• Location and scale: a 20 MW supercluster in Chennai. Vultr says this, together with its cluster in Springfield, Ohio, gives it two of the largest AMD AI deployments in the world.
• Hardware: AMD Instinct MI355X accelerators. Each GPU carries 288 GB of HBM3E memory with 8 TB/s of memory bandwidth. Large memory per GPU matters for training and serving big language models, because more of a model fits on each chip.
• Software: the cluster runs on AMD's open-source ROCm stack, which supports widely used AI frameworks, libraries and models.
• Workloads: training, fine-tuning and inference.
• Customers: AI labs and foundation-model builders, AI-native start-ups, enterprises, and research and academic institutions.
• Roadmap: Vultr says it plans further capacity in 2027 using AMD's next-generation MI455X GPUs and the Helios rack-scale architecture.

Vultr CEO J.J. Kardwell said companies scaling AI need compute that is available locally to meet data-sovereignty, data-residency and latency requirements. The Chennai cluster is meant to let organisations in India train and run models on infrastructure inside the country.

What has not been disclosed

The announcement leaves out several details. As Data Center Dynamics noted, the companies have not disclosed the number of GPUs, the investment value or an expected go-live date. They also have not named the data-centre facility that will host the cluster. Chennai already has large campuses run by operators such as Digital Realty, Equinix, Sify and CtrlS, but none has been confirmed as Vultr's host. DCD also reported that Vultr's Chennai cloud region went live earlier in the same week, adding to the Mumbai region it opened in 2022. For now, the 20 MW supercluster is a stated plan, not an operating facility.

Why Chennai, and why it matters

Chennai has become one of India's main data-centre hubs. It has several submarine-cable landing stations, relatively reliable industrial power and a state government actively courting data-centre investment. Recent announcements include CapitaLand's liquid-cooled Chennai 01 campus. A 20 MW GPU cluster is a substantial addition. At AI-cluster power densities, that is enough for thousands of accelerators, although Vultr has not given a number.

The announcement matters to India's AI ecosystem in three ways.

1. More compute choice. The IndiaAI Mission has subsidised access to tens of thousands of GPUs for start-ups and researchers, but most of the country's AI compute runs on Nvidia hardware, which has been in short supply worldwide. A large AMD deployment gives Indian model builders a second high-end option. That could ease supply constraints and put pressure on prices.

2. Sovereignty and latency. As India's data-protection regime takes effect and government and regulated industries push to keep sensitive data inside the country, the location of training and inference matters. In-country capacity lets banks, health-care companies and public agencies use frontier-class hardware without sending data abroad. For real-time uses such as voice assistants in Indian languages, local inference also cuts response times.

3. A push for ROCm skills. AMD has been working to build a developer base for ROCm in India, including a partnership with Delhi University to train students on the platform. A large local cluster gives Indian engineers access to the hardware those skills are meant for, which could widen the pool of developers who can work on more than one hardware platform.

How it compares with public compute

The Chennai plan comes on top of the government's own build-out. Under the IndiaAI Mission, MeitY has empanelled private cloud providers to offer subsidised GPU access to start-ups, academics and model builders. Electronics and IT Minister Ashwini Vaishnaw said at the India AI Impact Summit in February 2026 that around 38,000 GPUs were already available through the programme and that about 20,000 more would be added. Most of that capacity is Nvidia-based. A private 20 MW AMD cluster is not part of that pool, but if Vultr offers it on competitive terms, it could become a practical overflow option for teams that cannot get enough subsidised capacity, and a benchmark for what non-Nvidia hardware can do on Indian workloads such as multilingual speech and text models.

The MI355X is AMD's current flagship data-centre accelerator, built for the same large-model training and inference market as Nvidia's Blackwell-generation GPUs. Its large on-package memory is its main selling point: models that need several GPUs elsewhere can sometimes run on fewer MI355X chips, which lowers inference costs.

The caveats

Announcements of GPU capacity in India have often run ahead of delivery. Power connections, cooling and chip supply have all delayed projects. Until Vultr names a host facility and a start date, the Chennai cluster should be treated as a commitment rather than a capability. The Indian AI developer ecosystem has also been built mostly around Nvidia's CUDA software, and moving workloads to ROCm, though easier than it was, still takes engineering effort.

The bigger picture

India's AI policy has focused on making sure Indian researchers and companies are not short of compute, both through public subsidies under the IndiaAI Mission and by attracting private investment. A 20 MW AMD cluster from a global cloud provider, aimed at foundation-model builders and universities as well as enterprises, fits that strategy. If it is built on schedule, Chennai would host some of the most capable non-Nvidia AI hardware in the region, and Indian developers would have another option for training and serving models.

Sources

• https://www.businesswire.com/news/home/20261020586292/en/Vultr-and-AMD-Expand-Global-Collaboration-with-New-AI-Infrastructure-in-India
• https://www.datacenterdynamics.com/en/news/vultr-to-deploy-20mw-amd-mi355x-cluster-in-chennai-india/
• https://analyticsindiamag.com/ai-news/chennai-to-host-one-of-the-worlds-largest-amd-gpu-clusters
• https://cxotoday.com/media-coverage/vultr-and-amd-expand-global-collaboration-with-new-ai-infrastructure-in-india/
• https://digitalterminal.in/tech-companies/vultr-to-deploy-amd-instinct-mi355x-gpus-in-20-mw-chennai-ai-supercluster
• https://www.pib.gov.in/PressReleasePage.aspx?PRID=2229171&reg=3&lang=2