India’s AI adoption journey is anything but uniform. “Some customers are experimenting in the public cloud. I think it’s the easiest way to start experimenting and really trying out what you can do with AI. But some customers have actually moved on to realize concerns around token cost and governance and starting to really focus on actually doing things within their private domain,” says Anindo Sengupta, VP of AI, Cloud-native and NeoCloudSPs, Nutanix.
Interestingly, Sengupta doesn’t see any particular industry dominating production deployments yet. Indian IT companies made early moves, but enterprises across industries remain at different stages of maturity.
According to Sengupta, India’s startup ecosystem, particularly the ones based in Bengaluru, could become a significant driver of AI adoption. “India is a huge startup hub. I work closely with the startup ecosystem in Bengaluru and they are really leading a lot of the adoption for all the right reasons.”
This sentiment is echoed by Faiz Shakir, Vice President, Sales, Asia Pacific and Japan (APJ), Nutanix, although he believes Indian CEOs are now treating AI as a strategic necessity, not just a fringe experiment. “Different enterprises are at different stages of adopting and monetizing AI. But I think what’s very clear is that everyone has realized that they need to have an AI strategy in place and their boards, their CEOs are clearly earmarking budgets and building plans for AI.”
I met Anindo Sengupta and Faiz Shakir at Nutanix’s .NEXT on Tour Mumbai 2026, where both tech leaders quantified India’s AI adoption curve, and how homegrown innovation could end up defining India’s AI future.
The event also brought some interesting Indian customer developments into focus. Bharat Heavy Electricals Limited (BHEL) has consolidated 20 data centres into two centralised sites using Nutanix Cloud Platform, reportedly reducing its hardware footprint by up to 97%. Meanwhile, SIS Limited has migrated its business-critical SIS Core application to Nutanix Kubernetes Platform, reducing its monthly Kubernetes operating expenditure by approximately 80% and freeing resources for future AI initiatives.
One of the more substantial parts of my conversation with Sengupta concerned the total cost of ownership of AI. Businesses consuming frontier models through APIs have relatively straightforward expenses: their token bills are the effective cost of consuming that service. But private AI introduces additional responsibilities, believes Sengupta.
“Our opinion is customers will realize that they need to control token spend. They need to have governed, sovereign AI deployments but then the hard part is, in a way, the responsibilities for the customers to manage the whole AI infrastructure and their total cost of ownership is an important aspect there,” highlights Sengupta.
Shakir points out another nuance to the tokenomics paradigm, that cheaper tokens don’t necessarily mean cheaper AI. “Cost of token is coming down, but utilization has gone up so much, it’s still a huge factor,” he says.
Also read: Deploying voice AI agents at scale: Tata Comms, TTBS target India’s SMBs
As AI becomes more accessible, organisations may consume more of it, thereby incurring high costs of AI deployments. For Nutanix, the opportunity lies in simplifying the infrastructure required to operate AI economically. “What Nutanix does is, and what our passion is, we offer our customers the ability to manage the new accelerated computing infrastructure just the way they were managing their traditional enterprise application infrastructure. And it’s the same experience for them,” explains Sengupta.
Nutanix’s pitch is simple. Enterprises shouldn’t need to reinvent their entire infrastructure management just because GPUs and inference servers have entered the chat.
Perhaps the most interesting part of our discussion concerned agentic AI, particularly on the risks associated with autonomous systems executing more complex tasks. Who takes accountability when something goes wrong? For his part, Sengupta doesn’t dismiss those concerns.
“I think the concerns around agentic AI are valid,” emphasises Sengupta. Unlike conventional chatbots, these systems can remain active, make decisions and initiate actions. Therefore, Sengupta identifies three interconnected requirements – securing the agents, controlling their access to intelligence and models, and protecting the enterprise data they have access to.
“The agent itself needs some infra. They need compute, they need data. When it comes to data, they need secure access to data, because agents are only going to be smart based on data, right?”
Also read: Design for waste with no jugaad: Dassault Systemes on Indian tech IP innovation
Because enterprise data often deals with intellectual property, giving any agentic AI system unrestricted access to it is unacceptable, according to Sengupta. This is also where he believes private AI environments become particularly relevant. “Our view is customers will lean towards controlling that because they can control risk when they control the full stack.”
That doesn’t mean eliminating frontier models. Sengupta expects organisations to continue using them where their capabilities justify the expense, while increasingly turning to privately deployed open models for other workloads.
Sengupta also remains unconvinced that every enterprise AI requirement needs the largest, most sophisticated model to do their work.
“I think that one-size-fits-all is definitely not the right choice. Obviously that is the easy thing to do and that’s a great way to kick-start your journey, but for some use cases, a small model is good enough. And a large open weight model is really good for 80% of use cases.”
Sengupta’s response reveals Nutanix’s product philosophy. Instead of exposing customers to an overwhelming selection of models, Nutanix validates a curated catalogue of approximately 40-50 models for its enterprise AI inference offering.
“Customers will very quickly find patterns where you don’t need thousands of models but tens of models will meet 80% of your use cases.”
The accompanying architectural requirement is intelligent model routing. Sengupta explains that Nutanix is investing in an agent gateway and open-source technologies for the same. “Because smart routing is important. You could bring in logic that saves you cost by routing to the right model for the right prompt,” mentions Sengupta. In terms of enterprise AI, competitive advantage will increasingly depend on choosing and managing models intelligently, rather than picking the most powerful one.
Both Nutanix executives see sovereignty becoming increasingly important, in India and globally.
For Sengupta, sovereign AI reflects a broader geopolitical and regulatory movement extending across Europe, the Middle East and Asia. Shakir, meanwhile, believes India’s emphasis on sovereignty is particularly pronounced compared with other APJ markets.
Also read: Infrastructure control is key for sovereign AI stack, says IBM
“There’s a lot of emphasis here on Sovereign Cloud. While Sovereign Cloud is important elsewhere too, but I see that this is more spoken here than it is elsewhere,” underscores Shakir.
India’s significance to Nutanix goes beyond demand. Shakir says the country has become its largest APJ market, with considerable adoption among major financial institutions. He claims eight of India’s ten largest private banks and half its public-sector banks use Nutanix technology.
But perhaps his most striking India-specific revelation concerns intellectual property. “Forty percent of all the patents come from India. So, that’s a huge statement for us that our engineering staff is doing this at scale, and a lot of the patents come out, get used in our products and stack and that gets consumed by the world.”
According to Shakir, the company’s Bengaluru and Pune engineering operations contribute innovations deployed internationally, not merely products customised for domestic requirements. “It’s not just that India is building for India, India is building for India and the world.”
He also pushes back against the assumption that Indian AI engineering talent necessarily trails international counterparts. “From what we’ve seen from a Nutanix trend, and that’s what I will restrict my answer to very honestly, we’ve not seen a skill gap between the talent we have in India and the talent we have in other centers across the world.”
India hasn’t produced very many globally dominant frontier AI models. Does that necessarily mean the country is falling behind? Sengupta isn’t particularly concerned.
“In India, when I think of what we do very well is focus on application of technology. I mean if you take UPI as an example, there is no better example of how to make digital finance easy, right?”
He expects Indian innovation to emerge across specialised models, regional languages and applications addressing domestic requirements.
“Models are becoming commodity in a way but how do you use models and apply them is the next frontier. You’ll have centralized model farms and you’ll have edge AI as well. My personal view is we’ll find novel use for consumers which are very India centric that will be powered by AI,” believes Sengupta.
That could ultimately prove more consequential than chasing leaderboard positions. And according to Nutanix, that’s precisely where the next phase of India’s AI ambitions could become genuinely interesting.
Also read: AI impact on Indian workplaces: Humans are getting more important