Artificial Intelligence has swiftly transitioned from an experimental technology into an essential business force multiplier. Indian enterprises and boardrooms are no longer satisfied with exploratory pilots and proofs-of-concept—the mandate today is clear: get AI into production.
This paradigm shift formed the backdrop of a conversation at IBM Think 2026 with Sriram Raghavan (General Manager, IBM Software, India and Software Innovation Lab) and Sunil Gupta (Co-Founder, Managing Director & CEO, Yotta Data Services). The two leaders discussed their strategic partnership, which integrates IBM’s agentic AI platform, watsonx Orchestrate, onto Yotta’s GPU-powered Shakti Cloud, secured via Yotta’s Sovereign Core layer.
Below are the key insights on why data residency is not true sovereignty, how to manage “rogue” AI agents, and what India must do to transform from an AI consumer into a global exporter.
1. Moving from Pilots to Production
Enterprise deployment of AI hinges on three fundamental questions: How easy is it to adopt? How cost-effective is it? And how safe is it?
The partnership between IBM and Yotta directly addresses these questions:
-
Single Control Plane: IBM’s
watsonx Orchestrateoffers an agentic control plane that allows enterprises to manage, run, and deploy autonomous agents seamlessly across multi-cloud environments. -
Usage-Based Infrastructure: Building local GPU hardware requires substantial capital. Yotta’s Shakti Cloud provides high-performance compute on a flexible usage model, allowing businesses to scale up or down based on demand.
-
Beyond Chatbots: Enterprise AI often breaks down when confronted with messy, legacy workflows. To solve this, the collaboration provides:
-
Pre-packaged Workflows: Turnkey solutions (e.g., HR workflows) that allow businesses to implement automation without rebuilding underlying processes.
-
Custom Building Blocks: Modular document-processing tools designed to digest unstructured data and feed clean inputs back to active AI agents.
-
2. Data Residency vs. Digital Sovereignty
Hyperscalers such as AWS, Azure, and Google Cloud are making heavy investments in Indian data centers. However, Yotta’s leadership argues that hosting data within national borders does not equate to complete sovereignty.
“Data residency just means your data happens to be in India. Sovereign AI is about who controls the switch—the control plane—not where the data sits.”
— Sunil Gupta, CEO, Yotta Data Services
If the control plane resides outside the country, external entities could theoretically restrict access, deploy kill switches, or override governance during an emergency. Yotta’s counter-approach relies on a fully locally-operated stack spanning land, data centers, networks, compute (CPU/GPU), cybersecurity, and open-source control software developed entirely in-house.
3. Guardrails: Keeping AI Agents from Going Rogue
As enterprise AI agents gain access to operational and financial software, the risk of unvetted actions increases. Raghavan outlined three necessary guardrails for enterprise deployments:
| Guardrail | Function | Implementation |
| Agent Identity | Extends human IAM (Identity & Access Management) systems to autonomous agents. | Ensures agents never act on their own in sensitive or regulated workflows without explicit authorization protocols. |
| Least Privilege Delegation | Restricts permissions strictly to the task at hand. | Grants temporary access tokens valid only for the exact duration (e.g., 30 seconds) required to complete a single action. |
| Continuous Governance | Ensures real-time monitoring and safety verification. | Automates guardrail testing both prior to deployment and continuously while running live in production. |
4. Transforming India: From AI Consumer to AI Exporter
To transition from consuming foreign AI technologies to exporting homegrown capabilities, both leaders emphasized that India must build out every layer of the AI ecosystem:
The Integrated AI Stack
-
Hardware & Infrastructure: Accelerate local data center component manufacturing (over 80% is now made locally) alongside semiconductor initiatives like the IndiaAI Mission.
-
Contextual Models: Train LLMs on India’s localized cultural contexts (e.g., IBM’s active collaborations with Sarvam and BharatGen on Indic models).
-
Agentic Workflows & Skills: Train a skilled workforce capable of running and maintaining autonomous systems.
The Three Strategic Pillars for Sovereign AI
-
Capacity Building: Expanding data center footprints and specialized hardware pipelines.
-
Scale Deployment: Capitalizing on India’s 1.4-billion-person domestic market to test, adopt, and mature AI solutions in parallel with model development.
-
AI Digital Public Infrastructure (DPI): Similar to how UPI unbundled digital payments, creating a national AI DPI will allow public and private data to be served to autonomous agents via secure, regulated, Indian-hosted infrastructure.
Key Takeaway
For Indian enterprises and policymakers alike, true technological independence requires full-stack ownership. By coupling open-technology management platforms with locally owned infrastructure, organizations can scale production-grade AI safely while maintaining total control over their data, models, and operations.
