Coming to AI, is there a need for India to build its own large language models (LLMs)? If you were still at the helm of an IT company, how would you have ridden this wave?⁽²⁾

It’s a complicated topic. What should IT services companies do? And what should India do? These are separate questions. For IT services companies, there are three things they should be doing. The first is that AI, in general, is a revenue opportunity, and so they need to build capabilities to take advantage of that… The second thing is that AI will change how IT services companies deliver their services… And the third, which is maybe a little less obvious, is that in getting AI to do the work, IT services companies will need to invest in R&D themselves to produce these agents that are going to do the work.

Then the question is, what should India do, which is a broader question. In my view, if you simplify LLMs in AI, there are three layers of the stack. You have infrastructure, which is largely computing. We talk about GPUs as the big constraint there, and you have power... those have the biggest constraints right now. At the next level, you have the algorithm itself. Today… algorithms are mostly being made open source, so the algorithms are largely available. And then the third layer is the data itself, that the algorithms learn and train on. My view is that the algorithms themselves are largely available… The GPU and power constraints will alleviate over time in two ways: One, they may become more available as Nvidia produces more [GPUs] and other competitors enter the space, and as more power comes online. But, also… algorithms will become more efficient… And as algorithms become more efficient, it will reduce the need for GPU use and power.

Then the question comes to data. My view is that those who control the data will win the AI [battle] overnight… Also, data becomes the real advantage. I think in the context of India, we should be focussed on building its models over time, but not general models like OpenAI is publishing, but unique models based on Indian data. And the last thing I’m quite optimistic about, because I think India is one of the most densely digitised countries for its scale.

If we are to take GenAI as an inflection point creating a level playing field, home-grown IT services firms still seem to be shying away from investing substantially in R&D or acquisitions compared to some of their global peers. What is holding them back?

In IT services, M&A is not generally an end in itself. It’s a means to an end. So, you can achieve those in different ways. When a company needs to add a particular capability, or needs to enter a new geography or enter a new industry, it’ll do an acquisition… Now, M&A is one way to get there, but there are some firms that are very good at doing that organically... The first point I would make is that there are multiple moves to the end, and M&A is not always the answer… I think across the industry, it’s probably fair to say that the volume of M&A has gone up… If you go back, say, 10 or 15 years… most enterprises in the world ran their businesses on a small number of tech platforms… the Oracle ecosystem, the SAP ecosystem, the Microsoft ecosystem, or Cisco. So, there were a small number of very large tech players, and those players sort of dominated the enterprise ecosystem. Then the cost of innovation dropped dramatically because of the cloud… [and] we saw an explosion of SaaS companies, each focussed on some small niche of the market. Over the last 10 years, the client technology stack has gone from two or three big providers to hundreds of smaller players surrounding these big players.

What does that mean for services? The big services companies historically had to have big practices in those big areas. Now they have to build practices in many of these smaller areas as well… because building practices across hundreds of these smaller technologies is difficult, so M&As become a way to fill those gaps… And my prediction is that, that will continue to be driven mostly because of this — the fragmentation of technology.

In Coversation

Francisco D’Souza’s next big bet: Backing the future giants of tech with Recognize