Vultr has launched a 20-megawatt AI supercluster in Chennai, built around AMD’s MI355X GPUs and expected to house about 10,000 GPUs initially. It gives AI developers a substantial new pool of computing capacity in India, while Vultr’s much larger expansion targets remain ambitions, not firm delivery dates.
Watch Desk analysis
What happened
The Hindu reports that the facility is designed for training and fine-tuning large language and foundation models, as well as high-throughput inference. It will use AMD’s ROCm software ecosystem, which gives developers access to AI frameworks, libraries and models. Vultr plans to expand in 2027 using AMD’s MI455X GPUs and Helios rack-scale architecture.
The company is targeting 600 MW of AI infrastructure capacity in India by 2028 and gigawatt scale by 2029, but has not committed to those timelines. Global enterprises have pre-reserved a substantial portion of the Chennai facility’s capacity, Vultr told The Hindu, without naming them or saying how much they booked.
Why it matters
A 20 MW deployment with an initial 10,000-GPU configuration is a concrete addition to AI infrastructure, not simply another campus on a planning slide. ROCm support also makes this relevant beyond the hardware: developers will be able to use AMD’s software ecosystem for workloads that include training and inference.
The details that remain undisclosed matter too. Vultr did not provide the project’s investment cost, cooling technology, projected water use or quantified energy-efficiency estimates. It said 150 MW of connected power had been enabled at the site, with 20 MW available for the initial deployment. Data centres may be built to run AI, but they still need power, cooling and a fairly large tolerance for paperwork.
Our read
This is a meaningful infrastructure launch, with real capacity and a clear initial hardware configuration. The sensible distinction is between what is operating now and what Vultr hopes to build next: the Chennai supercluster has launched, while the 600 MW and gigawatt targets are not guaranteed milestones. The next useful proof will be operational detail, not another ambitious round number.
What to watch
- How much of the announced GPU capacity is installed and available to customers.
- Whether the planned 2027 expansion with MI455X GPUs and Helios architecture arrives as described.
- Whether Vultr publishes firm timelines and meaningful power, cooling and water figures for its India expansion.
Discussion spark: Should AI infrastructure providers have to publish power and water figures alongside capacity targets, or is that an unreasonable burden at this stage?
Sources and evidence
- Vultr launches 20-MW AI supercluster in Chennai; aims 1 GW capacity by 2029 – The Hindu (2 November 2026, 12:45 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.