India AI Impact Summit: Mistral AI's Arthur Mensch on Decentralizing AI Power

The story of this trial is really the story of three co-founders, me, Guio, Mpt, we are researchers by training and I'm a researcher by training. That's Arthur Mench, co-founder and CEO of Mistral AI, talking to my colleague, Sourbi Agarwal about its origin story. And spent most of my time in academia, say five years, and then spent three years at Google when I started to work on large language models. So in 2023, Mench and his fellow researchers and friends, Guillaume Lampel and Timothy Lecroix left Big Tech companies to launch Mistral in Paris. We knew that it was a very important technology before, but we realized that the market concentration was accruing and we were, I think, maybe a little bored in the Big Tech companies we were working with and so we saw the opportunity, we saw the needs, we saw that we could actually concentrate talent, which is the first thing we need to do as a startup is just to write funding team. And they started like many AI entrepreneurs with a research lab. To be honest, in 2023, we had no idea how to build a business. We built a research lab first. Of course we raised money and that means we had to build a business. We focused very early on on the enterprise sector, which meant that we then built a good market function, we built a product function, which means my role as the part of pretty significantly from being a researcher to being operating multiple things would be lucky to hire people that are experts in their field and able to help grow the business. Mistral AI builds high performance, large language models designed to give developers and enterprises more control in deploying advanced AI. The word control is important because in the last few years it has become Europe's boldest answer to American AI giants. Backing open-weight models and digital sovereignty, Mistral is challenging who controls artificial intelligence. In this episode, Arthur Mensch answers the following questions. His USAI dominance creating dangerous leverage can open models and sovereign infrastructure reduce dependency and is the computer race financially sustainable. Listen on, it's Saturday, the 21st of Feb. I'm Anirban Jodhary, you're listening to the Morning Brief. Good morning, Arthur. Thank you so much for speaking to the economic times. It's a real pleasure. Let me begin by asking you, you're here for the INDI AI Summit, right? We've seen a lot of action over the last few days, today is the last concluding day of the summit. What have been the key takeaways you were there at the PM's plenary session yesterday? So it's great pleasure for me to be here. In the Stormwood, he made a very interesting point on artificial intelligence being a technology that needs to be brought to everyone, which is something we truly believe in. And he made a very interesting point on self-reliance. He mentioned that this was too important to technology for them to depend on others. And so what that meant is that local infrastructure, there should be Indian local infrastructure, there should be Indian models, there should be Indian applications. And so on the entire stack, there needs to be a technology that is controlled by local businesses. And so that's the point, which truly resonates with what we have been building. It's a company that we started three years ago. And we've built the entire full stack offering on artificial intelligence in a way where we can enable self-reliance in every country and for every businesses. So that's what I can say about the summit. We were pretty much the only European being represented. And we're happy to be there because as you could tell, the US was heavily represented as well. In some way do you feel that there's too much of concentration when it comes to technology creation in the United States and that as a European company, do you feel that puts you a little bit of a disadvantage in some ways? That puts our customers at a disadvantage. The problem of market concentration is that it gives accessible levels to your suppliers if you're a business that is buying digital services. So our customers are at the disadvantage because some of their suppliers are coming to them and telling them, look, your running, your data is hosted by us. It's going to be more expensive for you. And tomorrow the problem is that it's not only the data, which is going to be hosted by us. It's going to be the processes and the digital workers that are going to be operated by them. So it gives them immense leverage, immense control. It's a control that can be used for geopolitical reasons as a way of pressurizing the states. It's also a control that can be used to just increase the economic imbalance. So at Mistra, I guess, does it put us at a disadvantage? I don't think so. It actually puts us at that advantage because certainly our customers and all of the enterprise market and all of the states are realizing that the market concentration is currently too high and it needs to be reduced. I think when I talk to some of the companies in India, they have a similar experience, right? Because it's the same issue of the imbalance that you spoke about, right? What is it that India can learn from the approach that you are promoting? I don't think we have anything to teach India. A lot of our researchers are actually from India and we have much more to learn from India and then we have to teach it. But first thing is I think our better open source has proven to be the right bet. We were the first to actually release open source models and we've seen a lot of other labs in the world actually follow that. We've seen Indian labs actually use our models to make them better at certain Indian languages. And so that bet on open source is the way to reduce the leverage that a handful of providers have today. And so the second thing is local infrastructure. So betting on local infrastructure, building infrastructure oneself is something that we are doing in Europe. Big customers ask for it. They ask for more decoupling from their current digital service providers. We see India doing it. We see that a lot of investment are being made on the infrastructure, not only to serve US providers but also to serve and to be operated by Indian providers. And there we can be a partner to this construction because we can bring some bricks of the software that can be then deployed on such a kind of infrastructure. And the third thing is also going all the way to the creation of value for it because we can help them automate their core businesses. We go in depth and we work with the lines of businesses. We work with the subject matter experts to transform a description of a process that needs to be automated or a description of how the machine is operating. To build an AI system to either automate the process or actually actually create the machine better. Talk to me a little bit about your own expectations out of the India market. How does it fit into your global strategy from a creation and adoption point of view of your models? So India is a new market for us and we are starting to hire here locally people from the commercial side at the moment. We are working with Indian providers or with global providers with a very big Indian food print like Kapshini. We think it's a huge market for in the consumer space, it's a huge market, we're not a consumer company, but we work with a lot of consumer facing companies and so we can help them scale their AI systems. We are very strong at making small models that are specifically made for a purpose and small models have a lower food print, which means that they are less expensive to serve, which means that down the line the end users can access it for a lower amount of money. And so making AI cheap is something that we can really help businesses with. The fact that we can bring a software stack on top of local infrastructure means that we are actually looking at partnering with local infrastructure provider that are building sovereign cloud, something that we do across the globe, partnering with sovereign cloud provider. And then finally we are very eager to work with two areas. One is startups, which are using our open source models. And the other sector where I believe there are some convergence is defense. Could you talk a little bit about, you know, are you in talks for partnerships during this visit? Have you met companies that could probably expand those? We've met a few. I have nothing to announce today, but we work with sovereign cloud providers that are announced in Europe, in Singapore as well. And so I think the recipe that we've followed is something that we can actually reproduce here. Talking a little bit about the data center play. There's so much of compute that is getting invested by players like OpenAI and Google, right? So that is something that Mr. also needs to significantly expand on. Or is that going to be like a big advantage or a big lever on which, you know, the success of companies in the near future will depend on. And how are you thinking about it? Now we're expanding our computer aggressively as we see a lot of demand for it. We are doing it in a way that is financially reasonable in that we are also working to while having a pastoral profitability and win very capital efficient when it comes to training on models. So that's something we want to stay. We are also very efficient when it comes to serving on models. So that's something we also invest on. So we're, we're scaring aggressively and we have more than 200 megawatt by the end of 2027 and our management. On the other hand, we just want to make sure that from a financial perspective, we can bring return to our shareholder base. So we're efficient and we are scaling. And we think that efficiency is going to be key in the world that is to come. Do you feel that there is already some frog in the US market because you know, there have been concerns of a bubble and there are conversations that you know, the kind of investments that are going into building compute are not justified by the revenues or the ROI, right? The returns on investments. What do you feel? Well, I'm not going to comment on some of our competitors, but the amount of leverage that is being used to actually build the faster is large. And so leverage is good, but leverage at some point, bikes. I think artificial intelligence is too important to technology for market concentration to accrue. It's going to reduce. And so what that means is that when you have a lower market concentration, well, then you need to be more efficient because your margins are going to be lower. And so what that means is that AI being an infrastructure business, you should run it as an infrastructure business. When you run an infrastructure business, you also focus on the return on investment on capital invested and the metrics that are boring for a tech company, but that are very well known for utility company. And as technology is becoming a utility, and AI is really much, is really a utility. Well, focus on efficiency is going to be extremely important to sustain a business. I think you are one of the poster boys of the European innovation ecosystem by having started, Mral. What can you tell us about the European ecosystem? How can we strengthen it further? And probably what lessons can India also share? Because we also have a very strong ecosystem, which is coming up. Europe is a great talent pool. And overall, I think what this company and what we've shown is that as long as you were decided enough to build a business from Europe, you could. And so I think what India can do, which we are starting to do as an ecosystem in Europe, is to actually behave as an ecosystem and make sure that the Indian entrepreneurs are doing their companies from here and not from the Silicon Valley. Because all of this innovation power, if it flows to the US, then this is lost for you. These are lost opportunities for any country that leaves the entrepreneurs leave their countries. So having a strong enough ecosystem, having access to capital, having access to markets is critical in making sure that entrepreneurs are doing what they want to do in the countries where they live in instead of going where everybody is. And so in Europe, I think we are getting there. In India, it's going to be the same. I think the inflection point is making sure that the people that are building companies are building companies here and not building companies from other places. Thank you so much for joining for this conversation. It was a real pleasure. Thank you very much. Thank you. Arthur Mench is gunning for competition over concentration. As nations race to build their AI stacks, the key question is, will companies like Mral AI actually decentralize who controls the pipes, the power and the profits of the intelligence economy? That's it for today. You are listening to this episode in the morning brief. It was produced by me sound designed by Amrit Redji. A new episode of the morning brief drops every Tuesday, Thursday and Friday, wherever you get your podcasts. But when the new cycle is too fast like now, we come to you more often. Stay tuned.