AI has moved on from being a buzzword to a force multiplier technology that is proactively being deployed across industries. Indian enterprises and boardrooms, specifically, are done exploring pilots and proofs of concept. Now, they want AI in production. Now, they are deploying AI to increase efficiency, streamline workflows and automate every process that can be improved. This shift is the backdrop to the partnership between IBM and Yotta Data Services, which brings IBM’s agentic AI platform, watsonx Orchestrate, onto Yotta’s GPU-powered Shakti Cloud platform, with the control layer secured through Sovereign Core.
It’s also an important part of the conversation that we had with Sriram Raghavan, General Manager, IBM Software India, and Sunil Gupta, CEO and Co-founder, Yotta Data Services, at IBM Think 2026, the India version of which happened recently. Beyond productionising AI, we also touched upon key concepts including sovereign AI, and India’s AI ambitions, during our conversation with the two leaders. Here are some of the most interesting snippets of our conversation with them.
IBM and Yotta Data Services: From pilots to production
IBM, earlier this year, announced a partnership with Yotta Data Services. At the time, the two companies had that said that they would enter into a partnership to deliver an Agentic AI platform, which would be built using IBM watsonx Orchestrate and deployed on Yotta’s Shakti Cloud, for enterprises and government organisations in India. More recently, the two companies announced the general availability of their Sovereign Agentic AI Platform for Indian organisations. While talking about this partnership Raghavan described it as a marriage of complementary capabilities. He said that scaling agentic AI needs an agentic control plane that “lets companies manage, deploy and run agents without worrying about what runs where”.
“It drives a lot of automation so that you can focus on the AI innovation and the things you actually want to accomplish,” Raghavan said adding that pairing that with infrastructure located in India means IBM and Yotta can “guarantee the residency of the data, the control of the data and offer basically a full stack solution where from the infrastructure all the way to the services, Indian clients are able to experience something that is completely sovereign”.
Gupta framed the moment more forthrightly. “Even the stage of pilots and POCs is over. Everybody has now mandated to take it into production,” he said. Once AI moves to production, enterprises ask three questions: how easy is it to adopt, how cost-effective is it, and how safe is it.
“In a way, the two of us are just combining these two value propositions. With Orchestrate, you are actually making it very easy through a single control plane, through a single window,” Gupta added.
On cost, Gupta pointed to the cloud playbook. Building GPU infrastructure in-house demands heavy investment and deep expertise. Shakti Cloud with Orchestrate delivers that “complex technology and that costly technology” on a usage model, which lets companies scale up, scale down or adjust the usage and costing as needed.
More than chatbots
One of the most common issues that many enterprise agents face today is that they break down while dealing with messy, unstructured, legacy workflows. This not only disrupts the progress but also existing processes in many cases. While talking about this issue, the IBM leader said that the IBM-Yotta Data Services’ partnership tackles this issue at two levels.
The first is offering pre-packaged workflows. “…we are bringing pre-canned services to market where you don’t have to deconstruct the problem. You have a HR workflow that you want to implement, between Orchestrate, the models that run on Shakti Cloud, you are able to just use the workflow and customize it without having to worry about all of the plumbing,” Raghavan said.
The second is building blocks for teams that prefer to build. “So now if you are working with messy unstructured data, you can bring that document processing workflow onto this platform, use the document processing capabilities, and pipe it back to your agent if you are going to be building it,” he added.
Together the two companies are piloting this solution with use cases in HR. They plan to expand it to other domains later.
Data residency is not sovereignty
AWS, Azure and Google Cloud are all investing heavily in Indian data centres. So what can Yotta Data Services offer that the global giants cannot? To this, Gupta conceded that, technically, any AI capability can run on a hyperscaler, and that localisation alone is not the issue.
“The point is where and who has the control over the control plane,” he said. If that layer sits outside the country, someone could, in a hypothetical emergency, access or remotely control it.
Gupta also argued the geopolitical mood has changed. “Fifteen years back, everybody used to say the world is flat,” he said. Today, every country is asking whether it controls not just its citizens’ data but also the layer that sits on top of it.
He also draws a sharp line between residency and sovereignty. “Data residency means just that your data happens to be in India,” he said. You can hold data abroad and keep full control, or hold it at home and have none.
So what is Yotta Data Services’ solution? Its pitch is a fully integrated stack under one operator: its own land, data centres, network, CPU and GPU compute, cybersecurity and people.
“One advantage we also have is that we are a complete integrated stack as an operator. Right from the control of our own land, our data center, our infrastructure data center, our network layer, our compute layer—GPU, CPU, everything—our cybersecurity layer, people layer, everything is in-house. The control layer which we have put on that, Shakti Cloud and Shakti Studio, is all open source, it’s all developed by our own people…when you combine all this, you are not subjected to any external control at any layer of the stack,” he said.
“Nobody from outside can sort of create a kill switch and sort of bring this business’s entire working down,” Gupta added.
Keeping AI agents from going rogue
One of the risks that enterprises today are facing is that of AI agents taking actions and accessing enterprise systems, potentially trigger financial or operational consequences. These risks and more make it important to place stringent guardrails to prevent such occurrences. Responding to a question on the kind of safeguards needed to prevent AI agents from going rogue and the importance of human intervention, the IBM GM outlined three safeguards.
The first guardrail is agent identity. Enterprise identity and access systems were built for humans. Now IBM is extending similar identities to agents with companies deciding when an agent acts on its own and when it acts on behalf of a person. “In many regulated industries and possibly sensitive workloads, you are never going to allow an agent to act on its own…So putting that, extending the the identity infrastructure in an enterprise, and with capabilities like Vault and IBM Agent Identity, that’s things that we are bringing to market to solve that problem,” he said.
The second guardrail is least privilege delegation, which means an agent should get exactly the access a task needs. “You want to give permission for the agent exactly what it needs to accomplish that task, and if that is an access token that will let it only do that task for the next 30 seconds, that’s all you want to give it,” Raghavan explained while adding that without the right platform, the easy route is to hand agents superuser access, which creates risk.
The third is continuous governance, which includes testing agents before deployment to ensure that the guardrails hold when deployed, and monitored continuously once live so teams can intervene quickly. At scale, this cannot be manual.
While explaining this Raghavan argued an integrated stack makes the job easier. If a company buys an agent from one vendor, a model from another and infrastructure from a third, “this problem becomes much, much harder to solve.”
Attacks will only grow
In addition to talking about the need for deploying guardrails, we also discussed about the scope of cyberattacks that enterprises face in the AI era. During the conversation Gupta told HT Tech that the infrastructure itself must be protected against threats such as DDoS attacks and that where Yotta Data Services-managed cybersecurity services are folded into the full stack.
When asked how Yotta Data Services anticipates future attacks, he was direct. “They are only going to increase. That’s the reality,” he said.
India: From AI consumer to AI exporter
The Indian government is slowly building its AI capabilities with the idea of democratizing AI as a technology for the public and ensuring digital sovereignty. Even with its existing efforts, India remains one of the major consumers of AI as a technology rather than an exporter. When asked what India must do over the next three to five years to build and export AI Gupta said the country will have to work on every layer of the AI stack.
Data centres are already a bright spot with more than 80% of their components being now made in India, which “was never the case even seven years back,” he said. He also listed the Semicon Mission as a step towards Indian-made chips. Above that sit building servers by Indian companies. On top of that are the datasets that are trained in India’s own cultural context, home-grown models and agents, and the skills to run them all.
For Gupta, the stakes go beyond resilience. “Are we defining our own AI destiny for the country?” he asked.
Raghavan, on the other hand, boiled India’s agenda down to three pillars:
1. Building capacity at every layer, from data centres to AI skills, accepting that each layer moves on its own timeline. The first phase of the IndiaAI Mission “was a good start, we need to do a lot more,” he said.
2. Adopting at scale in parallel. IBM is working with Sarvam and BharatGen on Indic models, and during the conversation Raghavan sees no reason India should not aim for its own “frontier-plus models.” But adoption cannot wait. “We have the opportunity to do adoption at a scale of 1.4 billion. How many countries have that scale?”
3. Building a DPI (Digital Public Infrastructure) for AI. Just as UPI unbundled payments, India needs yo create public infrastructure that serves government and citizen data to agents as a service, with the right authentication, certification and governance, on Indian infrastructure.
Talking about the role that IBM could play in this journey the IBM leader said that continuing to focus on open technologies is how the company could contribute to India’s future. “…I think as a hybrid company grounded in open technologies, I think we have a unique role to play in helping the country in this journey,” he concluded.
