Charlotte Kan, TelecomTV (00:07):
Thank you for joining us here in Barcelona for MWC 2025. I'm Charlotte Kan. Delighted to be joined by Vishal Sharma, vice President, artificial General Intelligence at Amazon. Thank you for joining us.

Vishal Sharma, AWS (00:19):
Thanks for having me, Charlotte.

Charlotte Kan, TelecomTV (00:20):
And you are here with me Vishal, to discuss the recent launch of Amazon Nova. So first of all, why did you build your own models when Amazon has already existing other partnerships with providers like philanthropic for instance?

Vishal Sharma, AWS (00:35):
Yeah, so as you know the A GI division, artificial general intelligence is responsible for bringing AI services and solutions to the rest of the company. And in doing that we get exposed to many different business units within Amazon and it's a very diverse set of use cases that you see, whether it's like prime video where we are doing things like building captions and building trailers and subsegments of videos, or whether it's Amazon shopping where we are doing things like helping construct seller detailed pages for products or Alexa across the board. We just see so many different use cases. One thing that became apparent to us was that it is such a large diversity that it's never going to be a one size fits all with models. So indeed what's going to happen is there'll be some situations that require a lighter weight model that's just extremely quick with latency.

(01:34):
A different situation might require something that's a little bit more capable and where you have a little bit more room. Similarly, for a generation we saw many, many use cases where image generation would be a very valuable thing and the same thing for video generation. So some people were surprised, but this is why we came out with five models instead of just one in December at the reinvent event. And it's been quite gratifying. Anthropic is a great company. We are great partners. Claude is a very good product and this is Amazon. So we believe in customer choice and ultimately everything we do is guided by the customer. And as a customer that's working with us, you have the choice to come into Pet Rock and you can use the Amazon Nova models, which we are very proud to bring to the world and share all of our internal learning. But you also have the choice of many other models in there.

Charlotte Kan, TelecomTV (02:28):
So the new family of foundation models is giving you extra flexibility to better serve customers. What makes Amazon so competitive in this space?

Vishal Sharma, AWS (02:37):
Yes, in order to create models which are extremely capable and we benchmark all of our models with public benchmarks and there again in our experience, it's very clear that our customers will expect nothing but the best in terms of benchmark performance. You need a peculiar concentration of three things, which is AI talent of which we have a lot. We've been doing AI for about 25 years, lots of experience. Alexa is more than 10 years old and we apply AI everywhere. The second is compute and that's where AWS comes in. Unprecedented levels of compute as you know. And the third is data. And we have a combination of proprietary licensed open source data, which is also fairly unique in the world. So the combination of these two things is really what enables us to create these super powerful models and that really differentiates what Amazon is able to bring to the table. And then the diversity of use cases for AI within Amazon. If you are a customer of Amazon's today, almost regardless of which significant product that Amazon use, whether it's Prime video or it's shopping or it's Alexa, you are experiencing multiple instances of generative ai. So there's a lot there.

Charlotte Kan, TelecomTV (03:49):
So how are customers responding to Amazon Nova and what about adoption rates?

Vishal Sharma, AWS (03:56):
It's been very positive, Charlotte. So we've gotten just such consistently good feedback across the board. And I'll give you some specific examples maybe just to bring this to life. So you have for example T trx, which is in threat detection. And so as you can imagine in threat detection the volumes you're dealing with are very high. So you need something that's lightweight but we're also extremely capable and has very low latency. So they're using Amazon Nova Micro and to great effect. On the other hand, you have a company like AppFolio which does services and solutions for real estate managers, property managers. And as you can imagine, property managers are very busy people. You just imagine the average facility and how many things they have to do. And so they have an assistant product for property managers called Rex, and that's been using Amazon Nova Pro.

(04:52):
It's about 10 hours of saving per property manager per week, which if you imagine a property manager's week is huge and a 20% uptake in terms of adoption of the assistance. So it's been very powerful. On the generative side, we are working with many, for instance in the area of advertising. So dsu, which is a global advertising agency, is using Amazon Nova reel to generate creatives, video creatives very, very quickly in advertising. Generally speaking, the act of creating of creative generation has historically been a very expensive process. So just visualize the way advertising has worked in the past where you can only make so many versions of a video and then pick one of very few of them. And now what you can do is create multiples very quickly and you can try out many, many different things very quickly and decide what works in what segment. So very powerful and very satisfied with the uptake. It just inspires us to do even more.

Charlotte Kan, TelecomTV (05:50):
It sounds very exciting and you are clearly off to a very good start here. What's next then? What are your in the years to come?

Vishal Sharma, AWS (05:59):
Yes, interestingly, we have announced what's to come this year. So as Andy shared at Reinvent, we are working on a speech to speech model that we are looking forward to bringing to market and which fluid speech and the ability for humans to interact in a very fluid verbal way is an extremely powerful capability. So that's coming out and then we are coming out with the Amazon Nova Premier model, which will be the most capable model yet. And we are also working on an Amazon Nova any to any model. So those that is a unique kind of model which combines image, video, text, spoken word, and in a very native way and then responds with any of those modalities in a very context specifically and context relevant fashion. So it's very powerful. We are very excited for what's to come this year.

Charlotte Kan, TelecomTV (06:51):
So Vishal, we wish you a lot of success in the years ahead. Thank you so much. Vishal Sharma, vice president, artificial general intelligence at Amazon.