Hi again, it's Andy Jassy, great to be with you here today. You can see I'm out of my more casual attire when I got to Vegas and now I'm wearing the jacket. It's Tuesday here at re:Invent and I had the privilege to be able to be back with the whole crew today and have a chance to do a short segment in Matt's keynote. Matt did an awesome job, but it was just it was thrilled to be back up on stage with everybody. So you know, we announced so many things today. I can't get through all of them and you probably don't want to sit through another conversation about all the launches, but, I thought what I would do today and then for the next couple of days, just give you some of the launches I'm excited about. And I'm going to talk about 3 today. The 1st I'm going to talk about is in the database space. And so if you've looked at what's happened to databases over the last 10 years, we went from assuming that there was really just one tool to rule the world, one database, to realizing that there's lots of Relational database opportunities and options and lots of non relational database options. And we built something almost about 10 years ago that was a cloud native database. And what it has is it's got the performance of the highest grade enterprise database opportunities and options out there while at the same time being a fraction of the cost and fully sequel compatible. And so this is something we call Amazon Aurora, which has grown really, really quickly. And what's happened overtime is that because we've had so many customers enjoying Aurora and getting so much value from it, we keep building features. And among the features that we've built that have been most successful is providing a serverless option for a world. That means that the database just scales as you actually use it. And So what we found over time is as the world was, was employing more and more complicated and sophisticated applications. They, they had end users all over the globe. And So what they were and this was pushing the limits by the way of most relational databases. So what they really wanted was they wanted a multi region database that was also low latency, that was high availability, that had strong consistency, that had 0 operational burden. And oh, by the way, with Sequel compatible and just just a few of those things. But they want it all. And what you found is that there are. Couple of these multi region database options today, but they either have low latency and high availability, but they don't have strong consistency or SQL compatibility, or they have high availability and strong consistency but they don't have low latency or SQL compatibility. And so you're having to make these "or" choices. And we talk about this a lot inside Amazon, which is the tyranny of the "or." You don't want to have to make org choices because it means that you have to give up on some of the customer capabilities you want to provide. So we worked hard this and we launched today something called Aurora DSQL. And so this is a new flavor. It's a distributed database, it's multi region, it's very low latency, it's high availability, strong consistency. It has 0 operational burden because it's all serverless, like the serverless Aurora people are loving and its sequel compatible. So people are very excited about this. I think it's going to be a huge change for what people can do. People ask us about this will compare to something like Spanner, apart from just having better SQL compatibility when you benchmark it, because the hard part of these distributed databases is the latency. Aurora DSQL has four times lower latency than Spanner. So I think people are going to love it. So that's the first one I was going to talk about. The second is really around chips. And so one of the big lessons that we've learned from having about 1,000 general AI applications. that we're either in the process of building or have launch at Amazon is that the cost of compute in these generative AI applications really matters. It's often the difference maker of whether you can do it or you can't. And to date all of us have used really just one chip in the compute for generative AI and people are hungry for better price performance and so the launch that Matt shared today of trainium2, which is our new chip that we built, which are in our Trainium2 instances in EC2 or Trn2 as we call them is going to be a game changer. It has 30 to 40% better price performance than the current GPU instances in EC2. Think about that 30 to 40%. better price performance is a big deal when you're doing generative AI at scale. And so what we've done in, in each of these Trn2 instances is you've got sixteen of these Trainum2 chips in there. It allows you to run 20.8 petaflops of compute its peak performance, which is quite a lot. And then we built an additional innovation that we call UltraServers, which basically what we've done is we've taken 4 Trn2 servers and put them together through ultra fast Neuronlink connect. So that means that you effectively have 64 Trainium2 chips in this one UltraServer and that gives you, I think the math is 83.2 petaflops of compute at peak performance. It's completely different in terms of what you can do. You're going to use those UltraServers and a cluster of those to train the most sophisticated, gorgeous foundation models in the world. In fact, we just announced a collaboration with Anthropic where they're going to build their next generation of large language models on top of training them too, but on an ultra cluster of hundreds of thousands of training them to chips. So this is very exciting and I think this is going to make a big difference for you as you start to scale your compute and generative AI. And then the third one I'm going to mention is really around models. And so what's interesting in those thousand. Internal generative AI applications is that a lot of our internal builders have asked us to provide them better latency, lower cost. The ability to fine tune their models, the ability to connect knowledge bases, they can ground their answers and their own data, the ability to take agentic actions on their behalf. And so we, we have a lot of model providers that are partners and, and, they have been partners and will be partners for a while, for a long time as far as I can see. And when we mentioned this feedback, they're very receptive to it, but at the same time they're busy. And so it's one of the reasons why we have continued to work. On our own Frontier models, and we've made tremendous progress on these models the last four or five months. And we figured if we found them useful, that our customers would find them useful as well. And so today, we launched our own Frontier models that we call Amazon Nova, and we launched several flavors of it. So on the Intelligence models, we have 4 flavors. We have a text only model called Micro, which you input text, you get back text, it's laser fast, very cost effective and it's great for a lot of the simple tasks our internal builders are using. And then we have 3 flavors of multimodal models, which means you can input text, images or video and you get back text. And so each of them in ascending order of size and intelligence. And if you look at the benchmarks, they benchmark very competitively. They're they're, they're really compelling models, but there are some things about them I think you're going to look. First of all, they're very cost effective. They're 75 percent less expensive than the other leading models in Bedrock. They are laser fast. The fastest models you're going to find there. They they all know the models allow you to do fine tuning and increasingly our application builders for general AI want to fine tune the models with their own label data and examples. It allows you to do model distillation, which means take a big model and infuse that intelligence in a smaller model so that you get lower latency and lower cost. It's well integrated with your knowledge. bases and Bedrock so you can ground that data and there are your answers with your own data. And then it gives you the ability to connect with your proprietary systems and APIs so that we can do orchestration of several automated actions for you or what people often call agentic behavior. So I think these models are going to be very compelling for you. We also launched an image generation model called Nova Canvas and a video generation model called Nova Reel, which both also benchmark very well and will let you create image and video much more easily. So I'm very excited about these capabilities. These are some of my favorite launches of today. I wanted to talk about some of the things in Bedrock. I wanted to talk about some of the things in Q. I wanted to talk about some of the the, the, the things that we're launching in SageMaker. But the good news is. That we actually have a number of launches coming in each of those areas the next couple days and I'll talk about those in a future day. Thank you.
Career Management Consultant, Career Coach, candidate advocate, full review of your status, goals, personalised strategy planning. Guidance and advice on how best to reach your goals. Senior Recruitment Advisor.
Who benefits from this in the end? Why does Amazon need an A.I supercomputer? Why are we not focusing on the environmental impacts of this technology? Oh I remember because its A.I...the hypocrisy of the environmental technocrats is glaring. I must say I am optimistic about the role of lawsuits in keeping things to account.
I just left amazon after 13 years, 4 injuries, and amazon just told me good luck 🍀 how awesome is that, we work like slaves, our bodies getting in pieces.
I'm all for progress with AI and chipmaking, but...
Where is all this massive computer power being placed in America? The American electrical grid is already on the edge. If Amazon builds the power stations necessary to power their ambitions, great!
Bring on the quantum computer revolution. With classical computing power required for cutting-edge large AI models doubled every 3.4 months using 5 percent of all energy consumption in the US. A 2020 demonstration showed that a quantum computer could perform a math puzzle using 50,000 times less energy than the world’s most powerful supercomputer at that time this could be a game changer not just for the speed of commuting but also for the environment. The science needs investment such as IBM is doing Amazon as an end user could quicken this transition by investing in the next generation of computing.
AI and Decision Science Innovator In Culture Change, Leadership Development & Human Transformation | Founder | Culture Architect | Psychologist | ❇️ Author: The Nexus Newsletter (A Weekly Publication).
Former CIO with more than 25 years of IT and cyber security experience. Cal Poly University, Commissioner, Paso Robles Airport Commission
1wWelcome back!