Mistral Launches Platform for Improved Enterprise AI

Bloomberg

Watch Mistral Launches Platform for Improved Enterprise AI - Bloomberg on Dailymotion

Transcript

00:00This is a path to production, right? And the enterprise category between you, Anthropic,

00:07OpenAI, it's a fierce battle. I think let's just start by you explaining why you feel

00:12this is such a significant milestone for Mistrel.

00:17Hello, and happy to be here. For us, we're a global company really focused on creating value

00:23for the enterprises. And we have had three years now of experience in doing that. And so AI Studio

00:29is basically the one-stop shop platform where you can build your AI application as a business.

00:35And so it brings everything that we need to create value in the enterprise, to create applications

00:41that belong to the enterprises, that contains their IP, that is connected to their data,

00:46and that improves over time by leveraging the knowledge that is contained within enterprises,

00:50that is contained within employees' mind.

00:52You are a global company. You're producing foundational models, Arthur. I wonder, though,

00:57how you compete against those that are already serving the enterprise. How is your solution

01:02different from that of Anthropics, that of OpenAI, that of Xs and many others?

01:06Well, I think our solution is much more integrated in that we have thought about all of the bricks

01:12that needed to be there to go from a model, which is how we started, to an application with the right

01:17front end, with the right back end, and with the right learning mechanisms. So that integration is

01:22really the reason why we've been succeeding in delivering value with enterprises. So that's one

01:26very strong area of differentiation. The other thing is that, in contrast with some of our competitors,

01:33we really believe that enterprises should build their own AI, that they should own the system,

01:38that the IP should remain their own, that the data should remain where it is and should not

01:42necessarily flow back to us. And so that approach of building a portable platform that can be deployed

01:48on-prem, that can be deployed on private cloud, combined with the vertically integrated approach

01:55of having just one platform allowing you to go from prototype to production, I think is very,

02:01very different from the approach of our competitors. And that has allowed us to make a lot of progress,

02:05in particular in the US.

02:08Arthur, can you talk about how quickly you're able to onboard scaled enterprise customers to this,

02:15and in turn, therefore, what the capacity and compute constraints are for you in launching a new

02:21product, and then trying to run workloads with clients?

02:26Well, a good example that I can give you is what we've been doing with one of our big customers,

02:33which is a logistic company called CMA-CGM, and one of the biggest shipping companies in the world.

02:38And so as many other companies, they were a bit struggling with the adoption of AI. If you look at the

02:43the MIT reports, they state that 95% of use cases just don't go into production. And the way we have

02:49done the work with them is we use the AI studio, and we deployed our AI applied engineers, our AI

02:55applied scientists, we made them work with the business unit owners, and they realized what kind

03:00of processes could be automated end to end. And from ideating what we could automate to going into

03:05production, so from March to July, in four months, we were able to understand a full function to be

03:11automated, so cargo release and dispatching of containers to multiple customers. We're able

03:18to go from that to production that is now going at a global scale. So the way we think about it is

03:26that it takes a quarter to go from a prototype to production. And for it to happen, you need to have

03:32the right tools. You need to have the right tools. You need to have the right experts. You need to deeply

03:36understand what AI agents can do and cannot do. You need to connect the agents to the data sources.

03:42You need to connect them to software, which is sometimes legacy. So you need to have the right

03:46interfaces, the right plumbing. That's what AI studio can bring. You also need to have the experts. So

03:52that's why we deploy people in the companies that we work with.

03:55Arthur, there is intense interest in Mistral as the European champion among AI labs, frontier labs.

04:02The thing we're most interesting to learn about is the progress of the follow-on fundraising round

04:08that's reported you're undertaking. How much money are you seeking to raise? But beyond the money,

04:14why do you need to do a quick follow-on? What is the most pressing need for capital at the moment?

04:20So I'm not sure we'll talk to you about the follow-on, but we are not raising money at the moment.

04:25We've just actually announced a fundraise with ASML, which is the biggest European company,

04:31and with whom we work, with whom we have a commercial agreement, where we help them make

04:36better products, automate their processes. We help connect their vertical expertise to horizontal

04:42expertise. We make them create their own AI and deploy this AI into their systems. So that was a

04:501.7 billion euro raise, which makes us, we've raised around 3 billion euro in total. And that

04:59in turn has enabled us to grow very fast and to create unique technology and to release it

05:03in particular on the open source world. That was really interesting, this strategic investment

05:07coming from ASML. You just referenced CMA, CGM, another French-based giant. It's a global company,

05:14but it's French-based. You got work with Stellantis. You got work with BNP Paribas. What about US

05:19companies? How are you managing to penetrate here? So we are working with multiple US companies. We're

05:25working with Cisco, for instance, which has chosen us because we allow them to deploy on private cloud

05:31and we deploy people to work with them, in particular on the customer experience side. So that's one

05:37important customer of ours. We also work with Snowflake, deploy our technology, in particular

05:42on document processing and all of our technology to structure unstructured knowledge and to take

05:50unstructured knowledge and turn it into things that can then be crawled by AI systems. We also work,

05:56as you know, with Microsoft, which is a very important partner of ours and exposes our model

06:00on Foundry, our models being very competitive, in particular on their cost-to-performance ratio.

06:08We work with AWS and GCP, and we work with multiple digital natives companies as well.

06:14Arthur, your peers, Dario Amadei from Anthropics, Sam Altman, OpenAI, are looking increasingly to the

06:21Middle East, both for capital but compute capacity, right? This week, we broke the news of Anthropics

06:28securing 1 million TPUs with Google and GCP. That's the scale that people are talking about.

06:35Are you making those future plans for scale? Are you looking at debt markets to ensure the future

06:42of Mistral and what you want to do? The future of artificial intelligence is very linked to

06:48infrastructure, as you know. And as a consequence of that, we have created a new business unit called

06:55Mistral Compute, which is creating digital infrastructure, GPUs in particular, in Europe,

07:01which is lacking infrastructure today. And so we've been growing that very fast, and we are

07:07acquiring multiple hundreds of megawatts capacity in the coming year to actually be able to serve

07:14our customers, to serve AI startups, to serve all of our existing customers and customers that come to

07:21us because they are looking for sovereign capacity. We also use that capacity to train our own models and

07:26to maintain our leadership on the open source fund. But with that comes the need to continue to invest.

07:32You've raised money. And is that what's being used to finance the exploration and data center build out?

07:39Because I'm looking at a headline at the moment that the Bank of England itself is really worried

07:43about a so-called AI bubble and some of the debt and the financing that's going on of data centers

07:48at the moment. Is that something that gives you pause, Arthur?

07:50Well, we've raised equity, and we are deploying that equity to actually train models. So R&D is financed by our equity.

07:59We also have long-term contracts for our compute facilities. And with those long-term contracts,

08:04we are able to actually finance them through debt. This is bank debt and gives us confidence that we are not

08:13overly exposed to debt. There is a lot of investment happening on the infrastructure side today. We're

08:21really focused on creating the long-term value that will justify these investments. Because we operate

08:26with venture prices, because we go all the way to delivering value for them, to identifying their use

08:30cases, to transforming what looks like magic to something that looks like money, we are not exposed to

08:38whatever may happen on overly investing on the infrastructure side. We invest on the infrastructure

08:43side in a wise way.

Category

Transcript

Be the first to comment

Recommended