Countries should focus on AI stack that's needed the most: Dell's Mohindra

Dell adviser says India's phased approach to sovereign AI is right path

Vivek Mohindra, special adviser to Dell's vice chairman and chief operating officer Jeff Clarke, says that a country's sovereign AI mission should focus on the part of AI stack thats needed the most. In a conversation with Avik Das, Mohindra, who also works with Michael Dell, talks about the company's AI strategy and what it takes to be an AI- ready organization. Edited exceprts…

 

How do you view the Indian government’s take on sovereign AI?

 

I think the term sovereign AI has different interpretations for different countries. The point I make around sovereign AI is that which part of the AI stack a country needs to own and control so that it controls its destiny in terms of what it's able to do. And number two, how the country is able to use that stack to fulfil its aspirations for its citizens and its ambitions. So if you look at sovereign AI from that perspective, I think all countries are approaching it slightly differently.

 

So for countries in Africa, their anchor points around sovereign AI has to do with not only data models, which most countries ink about, but also thinking about infrastructure from a very different perspective, becaus many of them do not have, nor do they need, to the tune of $50 billion per gigawatt to set up a training cluster. What they do need is smaller scale infrastructure clusters coupled with PCs to be able to allow their citizens to be able to take advantage of that, incorporating the local data, the local languages, and the local models, if you will.

 

I think the Indian approach to sovereign AI is the right approach, which is you cannot go from zero to 100 percent sovereignty across the whole stack. It has to be done in phases. And the way India is approaching it right now is the right measured approach towards sovereign AI.

 

From a Dell perspective, what we have done so far is taken the approach of whatever the right appropriate interpretation of sovereign AI is for any given country.

 

What sort of engagements do you have with the Indian government when it comes to building sovereign AI?

 

At Dell, we don't operate at the model layer from a software perspective, but we provide the infrastructure for that. For us, from a global perspective in terms of what is required to fully unleash the power of AI, which in that case revolves around providing an affordable edge, which is really not these big data centres in the central, but pushing AI to the edges, providing turnkey AI factories, because that's what a lot of MSMEs need, And then because security and governance becomes an issue, making sure that there's enough skilling that occurs as well, which the government is implementing.

 

Since you advise Michael Dell and Jeff Clarke, can you outline what are some the key pillars in Dell’s AI strategy?

 

First, we decided that our AI strategy has to be in support of our business strategy. Second, we took that AI and deconstructed it so we could help inform our strategies appropriately. And the way we just deconstructed that was in four simple things, AI in, on, for, and with. We thought about how do we embed AI in our products so they just run much better.

 

We thought about what our roadmaps across servers, storage, networking, PCs needs to be so our customers can run their workloads much better on our products, that's AI on. AI with was who do we need to partner with to bring these capabilities to our customers? And AI for was how do we take AI and apply it for ourselves to transform ourselves? And the second or the third thing we did is we developed a set of guiding principles around AI, One, it's all about the data. Without good data, you cannot do good AI. In a world where half of the data that enterprises manage sits outside of the public clouds and data centres, it's much better to bring AI to data than data to AI. It's cheaper and more secure. Number two, there's no one-size-fits-all. While the world is consumed with large language models and training, the reality is that the world will have small language models. Next, you need an open modular architecture because different layers of the architecture are moving very differently. And lastly, you need a broad ecosystem of layers to bring these capabilities to the bear.

 

You have a career that spans across tech enterprises, consultants, as well as private equity. As you steer an organization to being an AI-led one, what are the challenges, that enterprises face in being AI ready? 

 

From a strategic lens perspective, the real question is, how do you come up with a strategy that takes advantage of that pace of change and adapt the internal operations to be able to deliver on that pace of change? Then, how do you think about the risk-reward trade-off within those?

 

Frequently, it comes down to not only figuring out what are the use cases you focus on and getting the whole company focused on those, but doing things like getting the data in shape as well.

 

Change management is another thing. Because when you're doing it at scale, when you have tens of thousands of team members, then even if a coding tool appears to be really good in terms of helping you code more, doesn't mean that everybody will suddenly adopt that tool because they're used to doing things a certain way.  When you think about it from a private equity perspective or a venture capitalist perspective, it comes down to thinking through what are the big risks that emerge and how do you accrue value with that risk framing in the case of change.