At the Nasscom Agentic AI Confluence, Yasha Khandelwal, co-founder and CEO of Tech4Biz Solutions, presented Models to data centre to fabric: how sovereign AI actually gets deployed.
Sovereign AI is usually discussed as policy. The talk set out what it takes on the ground, using two builds the company has completed rather than a reference architecture.
A cloud region, designed and operated end to end
The first is Cloudtopiaa: a cloud region with a production-grade network fabric and a fully open source, auditable control plane, running on Tier-4 infrastructure in Bengaluru. Where data residency matters, every layer has to be open to inspection, which is not possible on a control plane you cannot read.
Models that go to where the data already is
The second is edge inference: specialised small language models, quantized and optimised to run on edge silicon. A large share of sovereign workloads will never reach a data centre, so the model has to travel to the data rather than the other way round.
What connects them
Ownership. Sovereignty depends on being able to see, audit and operate the whole stack yourself, from the silicon up to the fabric. A team that can only operate one layer cannot make promises about the others.