Dubai – Ashish (Ash) Puri has spent two decades investing in technology, so he has seen it all. And only a few innovations excite him as much as AI’s potential to reshape the world. In this interview, he explains why Deepseek is a game-changer, how emerging economies can overcome AI’s biggest barriers, and why AI could soon become as commonplace as the smartphone. He also explains how reducing infrastructure costs and embracing open-source models can accelerate global AI adoption.

QUESTION: Tell us a bit about yourself and why you got into impact investing.

Ashish Puri: My name is Ashish Puri, a partner at Lightrock – a global investment platform that backs purpose-driven companies – and I’ve been investing in tech for almost 20 years. I am an engineer by education, with a double master’s in electronics and computer science. So, I have a big passion for technology.

I’ve been investing in tech for a long time, especially where technology would meet business interests and thus you can make a profit…which is great. But at the same time technology has accelerated quite a bit. We have the best technologies in the world, yet there are fundamental issues on the planet which are still not being resolved.

One of the main reasons I was drawn to Lightrock and impact investing was the birth of my daughter. Her arrival made me acutely aware of the significant challenges our planet faces for the future. When we can address these immense problems through technology, we’re not only contributing positively to the world, but we’re also fostering sustainable outcomes. This dual achievement is crucial and is currently lacking in many areas.

For me, capitalism with a purpose was kind of what I wanted to do because I want to leave a meaningful legacy behind, a legacy of doing something good for the world as opposed to just making money for myself or our investors or whomever. That is one of the reasons I got into impact investing. Profit with purpose!

QUESTION: Focusing on the summit’s topic, viz. AI – assuming the claims about Deepseek are true, what kind of impact do you think it will have on the AI landscape globally? Can it help spur the development of LLMs and similar technologies in emerging economies currently lagging in the AI race?

Ashish Puri: I think the remarkable thing that Deepseek has done is the complete reduction in the barriers to AI adoption. This is very important.

The world can progress equally if it adopts a more democratized approach to any technology, and that’s what Deepseek is pushing if you think about it. As I mentioned during my talk on stage, developing any LLM requires resources. Now, what are these resources?

There are GPUs (graphics processing units): what do GPUs require? They require power and they require memory. And so the cost of infrastructure and the power requirements are so high, that not just any company can do it. And then they dominate and control the way access to this infrastructure.

Deepseek has challenged that philosophy, completely! Let’s not forget – the framework that Deepseek was built on was the Llama model. So it’s taken a US model that has around 64 billion parameters, but only uses a fraction of that because what they invented was this compartmentalization of data. You don’t need all the information all the time. You only need a certain number of these parameters at any one point in time, and limiting it to this reduced number leads to efficiency gains.

So, OpenAI became a major player only in the last three to four years, but at the same time, everyone has constantly talked about how expensive it is to answer these queries with AI, what’s called a cost per token. Reductions in token costs can only happen if you’re able to leverage the same infrastructure to do more. This increase in yield is what we tried to achieve with another company I had invested in a few years ago called Liqid. What we wanted to do was use the same GPU to be able to deliver more output using some clever software.

Deepseek, in its own way, has done that. They’ve compressed the model and they’ve increased their efficiency. As a result, less power and infrastructure are used, and, at the same time, it’s open source! Anybody can use it! So it doesn’t matter that it’s sitting in China or wherever – it can be used by anybody.

As AI continues to evolve at this pace, it will fundamentally change how it is adopted, much like the shift from perpetual licensing to Software as a Service (SaaS). AI will become an integral part of our daily lives. For example, when using Microsoft Word, you could simply command it, “Write me a letter,” and the customer could be charged for that specific instance of usage.

But to make that future a reality, costs must come down. I think Deepseek has made a lot of companies realize this. So yes, it is quite groundbreaking and this is probably one of the reasons why AI will start to get adopted everywhere. It’s like what Toyota did to the automobile industry – they made cars smaller and cheaper, and suddenly, everyone had a car. Deepseek is like the Toyota of the AI industry.

QUESTION: In your opinion, what has been the biggest stumbling block for emerging economies trying to get into AI? Is it cost, skill, or political will?

Ashish Puri: All of the above! Regarding political will, I think most governments are realizing that they now have to build smart models tailored to their requirements to make themselves more efficient amidst growing pressures on state finances and state-backed services. If there’s one thing Trump has done right, it is setting up an initiative to improve government efficiency! I think it’s important that governments think along those lines: how do can you become more efficient in solving citizens’ problems?

The second thing is infrastructure, which I touched on earlier. Infrastructure is the biggest barrier. The cost of GPUs is not trivial, neither is the energy required to run them. And for a developing nation, it always comes down to deciding how to allocate budgets and resources

And then the third thing is having access to data to build these models. But if you have open source, this problem is mitigated because you’ve got access to the code base. Having access to data is specifically important for developing nations to address their specific needs when it comes to education, poverty, healthcare and infrastructure.

And honestly, getting data is the easy part since it already exists. The hard part is how you get relevant insights from the data to enable better decision-making, and this is where I think Deepseek is going to be a step change in helping governments make more intelligent decisions. And since it is open source, governments don’t have to build everything from scratch, driving costs down to a point that makes widespread deployment a realistic goal.

And I also feel that this probably will drive prosperity in  a more uniform manner. Technology in the last twenty to twenty-five years has created prosperity in certain pockets of the world. There are lots of very intelligent coders around the world who have not had a chance to get access to necessary infrastructure, and with things getting democratized, you can start creating clusters of innovation across the planet.

QUESTION: Speaking of government efficiency, how can governments learn to set up these departments? Are there models they can follow and hit the ground running, not needing a lot of trial and error to suss things out?

Ashish Puri: Let’s get some facts straight. When it comes to data and being digitized, pretty much every government out there is working on this. It’s hard to believe that there are governments which are still not digitized to some appreciable extent.

So what do these AI models do? They essentially feed on this data to build insights that can tell you what operational efficiencies you can gain, or the actionable steps you can take to bring about a certain outcome. That’s what we’re trying to achieve here. It’s the same for a business as it is for governments. Why do businesses apply AI? Because they want to create operationalefficiencies,  to come out on top in the cutthroat competition between Company A and Company B! And what they’re trying to do is help to optimize costs to be able to produce the same product cheaper and yet have more mass appeal.

So why can’t governments use AI for the same purpose? They can take the data they already have and then use these models to understand where they are spending or acting inefficiently. This concept resonates a lot with me. Money is a finite resource. If you can put it to the right use, companies and countries can prosper.

The biggest challenge with a lot of economies, developed or not developed, is the fact that there is widespread stagnation right now. These technologies are excellent tools to help address this stagnation. They can give us the opportunity to look at the issues from new perspectives, help us better understand how we might drive growth, and gain a greater understanding of the barriers to growth .

Question: What can or should emerging economies do to attract investors?

Ashish Puri: Investors look at an investment from a return on capital perspective. The return on capital is based on two factors: the attractiveness of the investment because of what the business does, and the stability of the economy and context in which it operates.

At the end of the day, when you’re considering an investment, what you need to know is it possible to invest in this country. What factors govern this investment that allow it to grow? At the time of exit, how do you return the capital to your investors? The stability of the economy and the stability of legal frameworks are extremely important here.

This is why certain countries attract more investment attention than others—the simplicity of investment structuring and the robustness of legal frameworks that protect investors and other stakeholders.. I’m no politician, so all I can say is governments need to help investors and give them predictability to be able to come in and comfortably invest their clients’ capital. Because if there is anything that keeps an investor up at night, it’s thinking about what happens if things go wrong and how they can exit the investment.

Talent, however, is there: everywhere. Especially now with the internet. We have resources available that are allowing even children to learn how to code. So you have very smart people globally who have got access to the tools to be able to develop things. It’s just a matter of providing them with infrastructure, which comes with capital, which comes from investors.

The UAE is a good example of this. It has built itself as a hub to attract investors because investors see this as a stable and safe country. This, in turn, fuels innovation because capital attracts the brightest minds to follow it, who, in turn, want to build the most innovative technologies. It’s a snowball effect. This was the reason why Silicon Valley blew up in the way it did.

Now capital wants to diversify, it wants to go to different pockets around the world. We see it – Lightrock has offices globally.

To give a few examples, we see tech clusters forming in Southeast Asia, lots of things are happening in the Middle East across Saudi Arabia and the UAE. Kenya and Nigeria area also very exciting places. Similarly, there’s some brilliant companies coming out of South Africa. 

These are very ambitious nations. Many of these have articulated clear growth-centered goals and I think investors are starting to get comfortable with the way of doing things in these places.

I am excited to see what the future holds. 

Reported by Abi Nagarajan

Send us your press releases to [email protected]