DeepSeek AI vs. ChatGPT: How India’s IT Sector Will Be Disrupted!

Transcript Disclaimer: This transcript has been generated using automated tools and reviewed by a human. However, some errors may still be present. For complete accuracy, please refer to the original audio.

00:00:00 GOVING ETHIRAJ
What is the impact of DeepSeek on India? It is perhaps propitious that Sam Altman, founder of ChatGPT launched in November 22, is visiting India and was, or rather is in India right now as we speak, possibly drumming up support for his AI model and also to assuage concerns given the increasing number of lawsuits against him and his company. And Indian newswire has joined the flurry of lawsuits in many countries that allege Open AI has illegally used copy content to train the algorithms that powered it's popular chat bots. And then came DeepSeek. The Chinese frugal answer to ChatGPT launched apparently at a cost of just $6 million and in just a few weeks has kicked up a storm in the global technology community, leading to, among other things, stocks of major tech companies, including of course chip maker NVIDIA taking a battering. The big question for India is that if an AI model like DeepSeek, which is open source is available at much lower cost, that's the software and also needs much less computing power and therefore capital investment in chips and is accessible. How does that change the dynamics for India's IT industry and the potential users of AI technology? I reached out to Rajesh Nambiar, now president of IT industry body Nasscom. Rajesh was earlier managing direction and CEO of Cognizant Technology Solutions in India and I began by asking him how he was seeing the impact of DeepSeek now that we've had a few days or couple of weeks to digest it's arrival.

00:01:27 RAJESH NAMBIAR
While it's certainly, it's very early, the emergence of DeepSeek, especially the R1, you know, there are version three and then there's also R1. It's really a strong reminder that AI innovation is no longer monopolized by a handful of tech companies, right? And that's what we saw over the period of time, despite all the chip bans and export controls etcetera. And also being a private Chinese company, there is no question that this is certainly a remarkable achievement from the DeepSeek side. And I believe that the rapid strides in AI research and the ability to build, you know, highly competitive LLMS without massive GPU dependencies, that's what we saw. This totally disrupts the assumption that high compute costs are defining the IT will continue to be defining facts on the AI leadership, right? This challenge is, of course, the existence of AI hierarchy as we saw and the signals the rise of new power centers in AI development. And I believe that if at all, it teaches us 2 broad things. One of course is open systems will certainly outperform closed systems in AI. In specifically second one, I believe Govind is the constraints can be very, very useful, right? So engineers tend to actually operate best under constraints. So in my mind, these two are the useful lessons for us to take in some sense.

00:02:42 GOVING ETHIRAJ
As an India strategy, what do you think we should be? I mean, we've been talking about now, or rather there has been a debate about, you know, should we create our own large language models or just build upon other people's large language models? Where do you feel we should be or which direction do you feel we should be headed in?

00:02:58 RAJESH NAMBIAR
Often times it's not necessarily one or the other. I think it's kind of bored a little bit and I can I can talk about it a little bit, but let's look at what really was the take away, if you may, from the DeepSeek experience, right? They really used a different approach to train. It's especially the our own models, which is used to sort of by open AI. For example, contrast from the open AI method, the training actually involves less time, obviously fewer AI accelerators as we saw, and certainly less cost to develop. So they used this model called reinforcement learning, which basically means that deep seek, you know, they used a large scale reinforcement learning approach focused on sort of reasoning tasks and so on and so forth. And also they use very efficient knowledge transfer techniques. And their research has successfully compressed capabilities into surprisingly models as small as like a 1.5 billion parameters. Others are all way, way high in terms of how they're used. And then of course, this notion of what they call mixture of experts or something that they call that architecture also certainly made a big difference. 

So for India, to your question, I certainly believe that this is an opportunity and probably also a wake up call in sunsets, right? Because it reinforces the need for self-reliance and AI infrastructure, certainly talent development of course, and more importantly, the open source collaboration, which is I believe is one of the foundations. As we discussed earlier, our government has already laid out plans, you know, especially with the recent announcement, setting up a massive AI infrastructure with the part of the AI mission, including a facility of 18,000 GPUs and hardware infrastructure that is sort of the heart of creating AI, essentially to build a homegrown AI systems that understands India's very unique challenge, unique cultural nuances, and of course, the needs of our country. And I believe that we also heard that from the finance minister's budget speech announcing substantial increase of allocation towards AI emission. You know, it was about 173 crores, even though 400 crores were allocated last year to this year, it will be about 2000 crores. Of course, much of this is expected to go towards building LLMS along with sitting above centers of excellence as you saw that in the budget speech as well. 

All of that now coming to what we should really do LLMS versus application development expertise, some folks would say, I think the competition is not just about being who creates better chat bots or agents or AI systems. I think these are merely tools and probably different steps to an ultimate goal. The real objective is of course achieving the advanced AI, AGI etcetera that can really match or surpass the human performance across various domains, right. So when that level of capabilities reached, it will lead to profound and widespread disruptions across industries and and we are getting very, very close to that reality. In some sense, I believe that LLMS alone don't create value unless they translate into real world product and services. You know, you looked at, for instance, the AI mission in India, They have multiple approaches starting from compute to application challenges, safe and trusted tools, foundational model creation, you know, which is what was called Innovation Center and of course killing apart from many other objectives that they had. So the sweet spot is going to be a little bit of a dual pronged approach. 

As I said before, developing a little bit India centric, not necessarily India based LLMS, but India centric LLMS while aggressively integrating this AI into our enterprise, startup, public sector ecosystems and so on. This will of course ensure that AI does not remain as an isolated research endure, but certainly translates into tangible lack of progress. So it's got bit of both in my mind.

00:06:31 GOVING ETHIRAJ
Right. So let me ask you the enterprise question. So given now that we've seen DeepSeek and it's clearly or seemingly much lower resource call, what does that mean for enterprises who have been thinking about integrating AI for business outcomes? Assuming that business outcomes are now much clearer in terms of what they could have done with AI, would they be able to do more or will they still have to be? I mean, do they still have to get clarity on what business outcomes do they actually want to achieve?

00:07:02 RAJESH NAMBIAR
If you start from the problem statement and go backwards, I think it'll be more useful, right? Otherwise people will jump onto this as a yet another technology tool and then they may not end up where we really want these enterprises to do. Certainly the open source nature of deep seeks model will present a very, very unique chance for Indian developers certainly innovate and create very indigenous AI solutions, right? And the open source nature of deep sea truly fosters collaboration and innovation. 

So obviously building on its model creates a sort of a virtual cycle of improvement and adaption, very stark contrast to the proprietary models of many Western companies that we saw in the last whatever number of months and couple of years, if you may. But you know, in deep sea just doesn't stop at building a massive, I know. I thought they built a 671 billion parameter model. It also created smaller, what they call distilled versions. As I said before, smaller models that can run on a laptop or as some sort of a basic hardware. This makes very, very big sense for for our own startups and small businesses and individuals in India who don't have access to that expensive infrastructure. So unlike many other AI models that focuses on processing data, DTC Carbon also prioritizes accuracy and reliability in tasks like math, science and coding, etcetera, which are very relevant to our businesses. For our businesses needing high stake results, this trade off is very, very invaluable. 

I mean, I believe that for Indian startups where resources are often limited but ambitious are very sky high, you know, and this can become and certainly a game changer. Startups can integrate this R1 into their products without worrying about hefty cost. They need to pay API cost, they need to end up paying. I definitely believe that this opens doors to creating AI powered applications, you know, across all segments, whether it's healthcare, education, certainly fintech and other, even some sort of a rural development. I believe that can certainly make a difference.

00:08:56 GOVING ETHIRAJ
So what you're saying is that the transformative power of AI, which may have been visualized earlier, is more likely to be actioned now because it's clearly become more accessible and purely from a cost and resources point of view, at least at this point?

00:09:11 RAJESH NAMBIAR
Definitely. And as we saw, the Indian AI adoption is actually moving from, no. It used to be very exploratory in nature in many ways. It will now slowly move to the execution mode where companies no longer just experiment with AI, but they're actively integrating into the business operation. That's where you see the value, right? Customer experience, decision making extra, you're able to deploy some of this. I think they'll start to see the value.

00:09:33 GOVING ETHIRAJ
A question on open source. Now open source, we've seen open source over the years, decades actually, in various software counterparts or softwares. Now, what's your sense on open source in general, having used open source or not used open source? I'm sure there are reasons for both and. Therefore how would you see deep seek in the open source context or open source in the DeepSeek context?

00:09:53 RAJESH NAMBIAR
I think the open source was sweeter in many ways and I believe that the strong open source ecosystem has laid the groundwork for companies to develop what what I call home grown AI solutions tailored to both local as well as global needs. Right. But having said that, this journey cannot be undertaken in isolation and we should be building a collaborative AI ecosystem, one that integrates, of course, investment in R&D, which we should do a lot more investment in infrastructure and AI expertise with India's established technological strength. This is going to be very critical for us to truly leverage the open source ecosystem that you mentioned.

00:10:30 GOVING ETHIRAJ
And conceptually, you feel that open source is a good way to go. I mean, it works, but it also works with some boundaries, isn't it?

00:10:38 RAJESH NAMBIAR
It is true, but I think now we are talking about open source, which are not necessarily sort of the the best model of open source itself, but a little bit of a constraints around open source. So these are not kind of open source that in some parts of the ecosystem we experienced. But I believe that, you know, even in some of the models, I won't name them, but even from the Western world, it's not if The thing is true, there are one or two models which are open source as well. Quite honestly, I think even DeepSeek actually use some of the open source available model train fundamentally how they even came to where they are. But that helps. I mean, open source helps everyone and the more you use, the more it gets benefited and you're putting more and more into the ecosystem by which everybody benefits, the users as well as people who are creating these open source models.

00:11:20 GOVING ETHIRAJ
Right, Rajesh. Last question, So what was your first reaction when you heard or maybe experienced DeepSeek? What did you think?

00:11:28 RAJESH NAMBIAR
I just couldn't compliment them so.

00:11:30 GOVING ETHIRAJ
You believed it, and then you felt, oh, I mean you. But did you believe it? I mean, was there still shock and surprise? There were.

00:11:35 RAJESH NAMBIAR
A little bit of a scepticism I must say for the first 24 hours or so to say that is it real? Is it? Is it exactly what they even today for example in a lot of people question whether the cost at which they actually implemented is real or not. Is that some way they would have gotten some help from the government, to make what they did? Certainly, I mean, because it is a phenomenal achievement, as I mentioned before, with so much of constraints and so much of constraints put around them, whether it is the access to real compute power, access to certain chips, the restriction that the US government imposed on China, and so on so forth. So it is a little bit of unbelievable, but at the same time, now we know that it is for real. Certainly, I think there's enough amount of testing which is done on this shows that this is something which all of us should leverage. And I have seen including our own broader advocacy groups and we're looking at deep, very, very seriously. I think they've certainly come off all of those initial skepticism which came about the moment it was announced. I believe it's a phenomenal achievement by this company.

00:12:30 GOVING ETHIRAJ
Rajesh, thank you so much for joining me.

00:12:31 RAJESH NAMBIAR
Thank you Govind, really appreciate it.

00:12:34 GOVING ETHIRAJ
Meanwhile, India's finance ministry has asked its employees to avoid using AI tools like ChatGPT and DeepSeek for official purposes, citing risks posed to confidentiality of government documents and data, according to an internal advisory. Reuters says countries like Australia and Italy have already placed similar restrictions on the use of DeepSeek, citing data security risks.