AWS Summit Washington DC 2026 - Keynote — Amazon Web Services

[00:00:00] *Dramatic Music* [00:00:30] *Dramatic Music* [00:01:00] *Dramatic Music* [00:01:04] *Dramatic Music* [00:01:10] *Dramatic Music* [00:01:14] *Dramatic Music* [00:01:28] Please welcome Vice President, Worldwide Public Sector, Amazon Web Services, Dave Levy. [00:01:34] *Dramatic Music* [00:01:40] Good morning, DC. Wow, I love seeing so many people here today. [00:01:46] I want to take a real quick moment and thank all of our sponsors. [00:01:52] And before we get started, I want you to know we may run [00:01:56] just a few minutes over today, but we've got a very exciting keynote [00:02:02] and we have three very special guests [00:02:06] that I would highly recommend you stick around for. [00:02:10] This year, America turns 250. [00:02:16] That's a quarter millennium. [00:02:20] Think about what's been built in that time. [00:02:22] Not just the institutions, the ideas. [00:02:26] The notion that ordinary people, given the right tools, can do extraordinary things. [00:02:34] You can't mandate that. [00:02:36] That's a culture. [00:02:38] And it was already on full display when America turned 100. [00:02:42] 1876. [00:02:48] The country was barely holding together. [00:02:50] The Civil War had just ended 11 years earlier. [00:02:54] Reconstruction was stalling. [00:02:56] Europe looked at this young country and thought, it's probably a fluke. [00:03:02] And America's response? [00:03:06] Come to Philadelphia. [00:03:08] We'll show you what we've built. [00:03:10] And what they saw wasn't just impressive. [00:03:14] It was contagious. [00:03:16] They left thinking, if that's possible, what else is? [00:03:20] In one building, a Scottish immigrant named Alexander Graham Bell transmitted a human voice over a wire. [00:03:30] Live in front of the world. [00:03:32] In another, a Wisconsin newspaper editor named Christopher Latham Scholes unveiled a Remington machine that put words on paper faster than any hand. [00:03:45] The typewriter. [00:03:46] And in the center of it all, towering over every exhibit, was the Corliss steam engine. [00:03:56] Taller than this stage, it powered every single machine in the hall from a single source. [00:04:02] The newspapers called it the greatest exhibition of human ingenuity the world had ever seen. [00:04:10] But here's what I find remarkable. [00:04:14] Anyone can be an inventor. [00:04:16] Bell was an immigrant. [00:04:18] Scholes was a self-taught small town editor. [00:04:22] The engineers behind the Corliss learned on the factory floor. [00:04:28] What made them succeed wasn't money or institutions. [00:04:33] It was a culture. [00:04:35] It was a culture. [00:04:36] Give people a platform and tools and then get out of the way. [00:04:41] That's what America showed the world at 100. [00:04:45] And still does today. [00:04:47] We build AWS on the same way. [00:04:51] On culture. [00:04:53] 20 years ago, we started with a bet. [00:04:55] If you give builders the right infrastructure, secure, scalable, available everywhere, they do things no one predicted. [00:05:05] Researchers sequencing genomes, analysts finding signals in oceans of classified data, teachers bringing world-class computing to rural classrooms, and start-ups becoming the most consequential companies in the world. [00:05:23] The cloud didn't just change things. [00:05:25] It changed who got to invent. [00:05:30] But every customer we talk to is wrestling with the same gap today between what their mission demands and what their systems can actually deliver. [00:05:41] So at AWS, we remove the barriers between your goals and what's possible. [00:05:48] Because we believe that people doing the most consequential work deserve the most capable tools. [00:05:56] You can't run AI at the edge on infrastructure designed for email. [00:06:02] You can't process sensitive data through a cloud that wasn't built for it. [00:06:08] You can't modernize an organization on a foundation that buckles under the weight of your requirements. [00:06:15] What you need is infrastructure purpose-built for the mission from day one. [00:06:22] In 2011, when the cloud was barely a few years old, we did something many didn't understand. [00:06:28] We built GovCloud. [00:06:30] The first purpose-built cloud infrastructure for U.S. government-regulated workloads. [00:06:37] We listened to our customers and worked backwards from their needs. [00:06:42] And after 15 years of continuous investment, we're able to deliver purpose-built infrastructure for governments all over the world. [00:06:53] But infrastructure is only as valuable as your ability to choose the right tools and services and deploy them at scale anywhere. [00:07:03] That's why we give you options at every layer. [00:07:06] On frontier models, you pick the best one for each workload. [00:07:10] No lock into a single vendor. [00:07:12] On chips, you can optimize for performance, cost, or availability. [00:07:18] It's your call. [00:07:19] And on capacity, AWS operates the largest infrastructure footprint built for governments, including sensitive, secret, and top-secret classifications. [00:07:30] So when you need scale, the compute is there. [00:07:34] So everywhere we look, the ambition to do more, build faster, and put ideas to work is accelerating. [00:07:41] Organizations know where they want to go. [00:07:44] Our goal is to remove the barriers that stand between your mission and goals and reality. [00:07:52] So here's how we make that real. [00:07:55] Last November, a presidential executive order launched the Genesis mission, the most ambitious scientific research initiative in a generation. [00:08:06] The goal is to connect America's supercomputing resources, 17 national labs, and the private sector into one AI-powered ecosystem. [00:08:17] But with nearly 700 consortium partners, thousands of research teams, compute access, and cloud expertise remain the biggest barriers. [00:08:28] Earlier this year, we committed up to $100 million in AWS credits across two accelerator programs designed to remove the financial barriers that slow down our most critical missions. [00:08:42] The AWS Genesis accelerator provides up to $50 million in credits to help the Department of Energy national labs, research organizations, and their private sector partners bring AI-powered scientific breakthroughs to life on AWS. [00:09:00] And the AWS warfighter capability provides up to $50 million in credits to the Department of War and the Defense Industrial Base to accelerate AI and advance manufacturing at the speed the mission demands. [00:09:16] Both support workloads at every classification level, up to top secret SCI, with direct access to AWS architects and AI specialists. [00:09:27] We're also investing up to $50 billion to expand AI and supercomputing infrastructure across all gov cloud, secret, and top secret regions. [00:09:38] The goal is to provide 1.3 gigawatts of AI capacity, making a generational commitment to the future of government, that government runs on secure, intelligent cloud. [00:09:53] One organization that is investing heavily on AWS is His Majesty's Revenue and Customs, the UK's equivalent of the IRS. [00:10:03] 50 million taxpayers, three legacy data centers. [00:10:07] HMRC is investing 473 million pounds in migrating from those legacy data centers to AWS and rebuild their digital foundation for the future. [00:10:20] By moving their legacy estate to AWS, HMRC will then modernize their infrastructure, improve resilience, and lay the groundwork for AI and data-driven services that improve the experience for millions of citizens. [00:10:37] We know cost has long been one of the biggest barriers to cloud adoption. [00:10:44] The GSA OneGov agreement is another initiative that unblocks federal modernization by providing agencies up to $1 billion in savings for cloud services, modernization credits, and training support. [00:10:57] And the work doesn't stop with civilian agencies. [00:11:00] And the work doesn't stop with civilian agencies. [00:11:01] For the intelligence community, operating at a different level, the bar is even higher. [00:11:06] So today, I'm excited to announce the Intelligence Community Accelerated Modernization Framework. [00:11:14] This is a $1 billion AWS investment to get IC workloads off on-prem and on to a cloud that can run everything. [00:11:25] When you migrate, we share the cost. [00:11:33] When you modernize, we cover the engineering. [00:11:36] We invest alongside you until your most critical workloads are in production and on modern infrastructure. [00:11:45] We also understand many critical projects depend on industry partners who need to collaborate at the secret level. [00:11:52] Same cloud, same services, and same speed. [00:11:57] Following a multi-year collaboration with the Defense Information Systems Agency and the Defense Counter Intelligence and Security Agency, I'm very excited to announce the AWS Secret Cloud for Industry. [00:12:15] For the first time, Defense Industrial-based partners can access the same AWS Secret Cloud regions used by the Department of War with full classified workload capabilities and the services they already know. [00:12:33] The AWS Secret Cloud for Industry gives Defense Industrial-based partners their own tenancy within the AWS Secret region, the same services and security without the need to build or maintain classified infrastructure. [00:12:47] And to accelerate adoption, we're committing up to $20 million in credits over three years so Defense partners can get on the program, validate their classified workloads, and move to production. [00:13:00] We've already started onboarding our partners, helping them go from zero to operational in months, not years. [00:13:08] In fact, Northrop Grumman will be the first Defense contractor to deploy classified workloads on AWS, completing the authorization and onboarding process. [00:13:19] Without AWS Secret Cloud for Industry, their initial workload in the AWS Secret region would have required months of hardware procurement to stand up on-premise. [00:13:28] So we're excited to see how this initiative will give the partners behind our most critical defense work the same speed and scale they already count on for everything else. [00:13:39] Looking back at everything we just announced, and you'll see one thread. [00:13:46] Each one removes the barrier between an idea and the moment it goes to work in the real world. [00:13:54] It's like access for researchers, capacity for institutions, lower cost and risk for agencies, and a seat at the table for the mission. [00:14:04] Different customers, different missions, but the same idea every time. [00:14:09] Since day one, we've obsessed over infrastructure so you don't have to. [00:14:15] We've worked to give you the broadest, deepest set and choice of tools and freedom to pick the right tool for the job. [00:14:24] And it's all about choice. [00:14:26] Or as many of our customers here would say, optionality. [00:14:30] AWS continues to be the broadest selection of instances. [00:14:35] Whether you're running web servers, supercomputing workloads, or cyber security analysis, AWS offers the right instances for the task at hand. [00:14:45] Few jobs put that to the test, like sending humans around the moon. [00:14:51] Just ask NASA's flight sciences team that used Amazon EC2 and AWS GovCloud for the near real-time trajectory analysis. [00:15:00] These are precision calculations that ensure the spacecraft stays on an exact path around the moon and back, especially critical in the first 48 hours after launch. [00:15:12] And the incredible 4K images and video from Orion were transmitted by NASA's Orion Artemis II optical communication system over the AWS global network. [00:15:26] That's what our investment in silicon makes possible. [00:15:31] And we're not slowing down. [00:15:33] We recently announced Graviton 5, our next generation general purpose processor. [00:15:38] Graviton 5 gives you 25% more performance while maintaining the energy and cost efficiency that Graviton is known for. [00:15:48] For AI-intensive workloads like training foundation models, running large-scale inference, powering real-time decision support, [00:15:56] Trainium 3 offers the best price for performance for AI training in the cloud. [00:16:02] When you combine with our long-standing partnership with NVIDIA, you get the broadest selection of compute for every AI workload from fine-tuning to inference. [00:16:13] And the foundation of all of this design, of this is a design decision we made years ago. [00:16:20] You take the hypervisor and the storage controller, the network stack, and move them off the main processor entirely onto purpose-built Nitro hardware. [00:16:32] The result? [00:16:34] All communication channels within the Nitro system are encrypted. [00:16:38] All hardware and software components can be updated without any downtime. [00:16:44] No AWS operator has access to the Nitro system or your underlying server, eliminating the possibility of human error and tampering. [00:16:57] Now, I could talk about purpose-built infrastructure all day, but I'd rather you hear it from someone who's putting it to work at a scale and security level that few organizations in the world can match. [00:17:13] Please help me welcome Secretary of Energy, Chris Wright. [00:17:18] Well, we've really been looking forward to this conversation. [00:17:33] Thank you so much for joining us today. [00:17:35] Thanks for having me, Dave. [00:17:36] Okay, so before I dive in, I want to give the room a little bit of context on why this conversation matters so much to AWS and all of the folks out there who are really driving towards mission impact. [00:17:47] We've proudly supported American science at the Department of Energy for more than 10 years. [00:17:53] My entire journey here, we've obsessed over the Department of Energy, just fascinated by all the things that you all do. [00:17:59] And the partnership has only deepened. [00:18:02] When DOE launched the Genesis mission, AWS was standing side-by-side with the department from day one, [00:18:09] because we believe connecting those 17 labs with real-time AI and frontier compute is going to change how science gets done. [00:18:18] And so today we're providing access to Amazon Silicon and frontier AI models, putting the most advanced tools available in the hands of our nation's top engineers. [00:18:28] It's such a proud moment for us as a team and as an organization. [00:18:34] And I'll add one more point. [00:18:35] We're the only cloud service provider to date with the Department of Energy, with the DOE accreditation to process and store the department's most critical data pertaining to DOE's national security mission. [00:18:48] So that's a commitment. [00:18:49] We understand the commitment to security, and we're really proud of that and the trust that you put in us and take it very seriously. [00:18:57] So with that, let me start. [00:19:01] Year two. [00:19:02] Year two. [00:19:03] What excites you in year two, Secretary? [00:19:05] Oh, my gosh. [00:19:06] First of all, you said a proud moment for Amazon. [00:19:09] For me, it's a moment of gratitude. [00:19:11] It is a moment of gratitude for what you and the Amazon organization have done for us. [00:19:16] The massive investments you are making to make it possible to rapidly increase the speed of scientific discovery. [00:19:24] Undersecretary for Science, Dario Gill, you probably heard from him yesterday. [00:19:30] He and I, we grew up just passionate, tingling about science and innovation and engineering. [00:19:36] So we're thrilled to be in our roles, to be in the Trump administration driving science forward. [00:19:41] We're thrilled to have fantastic partners for you. [00:19:44] Real quick, of what we're going to get out of AI, I think a number of cancers today that are death sentences. [00:19:51] You've got a few months. [00:19:52] You've got a few years. [00:19:53] In a few years, we know exactly what that is. [00:19:56] We sequenced your thing. [00:19:57] We have a treatment thing. [00:19:58] They're going to be manageable conditions. [00:20:00] We are going to save hundreds of thousands, millions of lives. [00:20:05] I can't think of something I'm more excited about than that. [00:20:08] Truly remarkable. [00:20:09] Truly remarkable. [00:20:10] You launched the Genesis mission and called it the Manhattan Project of our time. [00:20:18] For technologists in this room who build on AWS every day, what does it look like when you connect 17 national labs, decades of federal research data and the world's most powerful supercomputers into a single cloud enabled AI driven platform? [00:20:35] In your vision, what becomes possible in the next six months or the next year or the next 10 years? [00:20:42] Well, I may give a couple ideas, but maybe the most exciting thing is so many things that I can't even think about today are going to be possible. [00:20:51] Like 12 months from now, 24 months ago, I'm going to be talking about, oh, my God, I hadn't even thought about that. [00:20:58] But this is what we can realize now. [00:21:00] But yet to take the 17 national laboratories, the greatest Nobel Prize winning factories on the planet, take all of that data. [00:21:08] Think nuclear physics, think the pursuit of fusion energy, think understanding materials, understanding biological systems at the molecular level. [00:21:17] All of that data, all of that data and all of those scientists now supercharged with artificial intelligence to learn from those data sets. [00:21:27] We've seen the incredible things AI can do with, you know, computer programming or scanning, grabbing things from language and all that. [00:21:35] Imagine if we unleashed that on these incredible scientific data sets. [00:21:40] I mentioned a little bit about the medical technology already. [00:21:43] Alpha fold is sort of an early, early use of AI. [00:21:47] And we're way beyond that today. [00:21:49] Imagine that at the molecular level, understanding of how biology works, how disease works, how cures work. [00:21:57] Look, I'm a career nuclear energy guy. [00:22:00] I went to college 40 plus years ago to work on fusion energy. [00:22:04] It's an awesomely complicated problem. [00:22:07] I always said humans are ingenious. [00:22:09] We'll figure it out. [00:22:10] Turned out to be harder than I thought. [00:22:12] We've continued to make progress. [00:22:14] But AI and using those data sets and these tools from AI, we are actually, I think in the very near future, going to figure out both how to confine a plasma to get to this. [00:22:27] This lawrence criterion to 10 to the 14th. [00:22:30] We're going to be able to get conditions to harness controlled fusion in a magnetic confined plasma. [00:22:36] And I believe with target design and everything else, we're going to figure out how to harness fusion energy in inertial confinement systems. [00:22:45] Those are things we've been creeping up on, but we're not there. [00:22:48] I think the progress will massively accelerate. [00:22:51] And in the next couple of years, we're going to understand the physics and the control required to realize these things. [00:22:57] Then there's the engineering to implement it. [00:22:59] But that science has held us back. [00:23:01] This is going to supercharge it. [00:23:02] This is going to make it a reality. [00:23:04] This is fantastic. [00:23:05] My dad, who's a retired nuclear physicist, just sent me his resume. [00:23:09] He's begging to come back. [00:23:10] Bring him in. [00:23:11] Bring him in. [00:23:12] We need him. [00:23:13] There are thousands of people in this audience deploying AI workloads and watching their energy demands grow. [00:23:21] You said if done right, this wave of data center investment will actually drive electricity prices down over time, not up. [00:23:31] That's a counterintuitive idea to what most would imagine. [00:23:36] Can you unpack that for this room and discuss the role nuclear energy will play? [00:23:41] I will. [00:23:42] And it's so key. [00:23:43] Of course, people's gasoline bill and their utility bills, not only their meaningful expenditures, [00:23:50] particularly for lower income people. [00:23:52] There are quality of life issues for lower income people. [00:23:55] And they're very politically salient because everyone pays a utility bill. [00:24:01] And look, in that previous administration, we went four years. [00:24:04] We had almost 30% growth in electricity prices. [00:24:07] And we produced no more electricity at the end of that administration than at the beginning. [00:24:13] I think we sort of lost our way in energy over the last 15 years. [00:24:16] I won't go into that too much. [00:24:18] But historically, the goal of an electricity grid was to deliver affordable, reliable, secure electricity. [00:24:26] And for 100 years, we increased the reliability of the grid and lowered the inflation-adjusted cost of electricity. [00:24:33] That's just what would happen. [00:24:35] And then we lost our way. [00:24:37] But we're finding our way back now. [00:24:39] And the biggest thing that's going to help us drive down electricity prices is two things that you're going to bring. [00:24:45] One is a load that's a more uniform load. [00:24:50] It's not flat, but it's throughout all the day. [00:24:53] The biggest problem with the electricity grid compared to, like, your car. [00:24:57] You could drive around your car and you see a cheap gas station. [00:24:59] Well, fill it up there and wait until you find another cheap one, you know, the next week to fill it up again. [00:25:05] Electricity, at every moment of every day, we have to match supply and demand. [00:25:11] So it isn't just people tell me we just want more electrons on the grid. [00:25:15] I'm like, no, we don't. [00:25:16] No, we don't need more electrons on the grid. [00:25:18] We need to increase our capacity to deliver electricity at peak demand. [00:25:24] That's what matters. [00:25:26] And today we have, like, as we sit here right now, we have hundreds of gigawatts of excess capacity on the grid. [00:25:33] But you can't put a giant data center on that grid right now because we could power it right now. [00:25:38] But when we get to peak demand time, we won't be able to. [00:25:41] So we have to look intelligently. [00:25:43] And so a lot of the times the load is only 50 or 60 percent of the grid capacity. [00:25:49] All that spare capacity is just sitting there. [00:25:52] When you bring a more uniform load, not a flat load, but a more uniform load spread throughout the day, [00:25:58] that means you add 10 percent more load. [00:26:01] We're going to put 10 percent more kit on the grid to deliver that. [00:26:05] But the average utilization of everything goes up, the average utilization. [00:26:10] And of course, Amazon and the other hyperscalers have said, we're going to work with you. [00:26:15] We're going to bring money up front for the for the transmission infrastructures that are needed. [00:26:19] We're going to fund or build the capacity we need. [00:26:23] In fact, we're going to build a little bit more than the capacity we need. [00:26:26] We're going to sign long term purchase agreements so people can build long term reliable electric generating capacity. [00:26:33] You know, by by subsidizing wind and solar, we just destroyed. [00:26:37] We had three nuclear plants close. [00:26:39] Why do we have three nuclear plants close? [00:26:41] Because we so distorted the electricity market that a lot of the times during the day when there's surplus capacity, the bid price is negative. [00:26:50] Because if you're a wind power producer, you get three cents no matter what anyone wants your electricity or not just in a federal government subsidy. [00:26:58] Plus, you've got a contract from the utility and you've got subsidized transmission to connect up to your remote things. [00:27:05] And what you do when the wind blows you as existing natural gas plant just gets turned down and per kilowatt hour, you save two cents in natural gas. [00:27:15] The federal government pays 50% more than that just in a subsidy. [00:27:19] Plus the utility contract, plus all the new transmission that's now in the ratepayers base. [00:27:24] So we just did silly things that drove up electricity prices. [00:27:28] Now we're going to do intelligent things. [00:27:30] Amazon thinking thing, data center thinking things. [00:27:33] We're going to put on reliable dispatchable capacity that's there 24/7. [00:27:38] And you've already proven it. [00:27:40] When data centers have been built in communities, three, four, five year freezing of electricity contracts have been signed. [00:27:48] In Indiana, we're going to have the first application for rate reduction. [00:27:52] So abs because I hear this all the time. [00:27:54] Oh, I don't like data centers going to drive up my electricity prices. [00:27:57] The answer is the opposite. [00:27:59] And last thing, if you look at the last five years, the states that have had the best performance in electricity prices are the states with the fastest demand growth. [00:28:09] North Dakota leads that 35% growth in demand. [00:28:12] Their price of electricity prices in real terms have gone down every year. [00:28:17] If you look at California, New York and the places with expensive and fast rising, [00:28:21] they produce less electricity today than they did five years ago. [00:28:25] If you look at England and Germany, three times higher electricity prices than the United States. [00:28:31] They produce over 20% in the case of Germany, over 30% in the case of the United Kingdom. [00:28:37] Less electricity today than they did 10 or 20 years ago. [00:28:40] Demand growth is your friend. [00:28:42] That's how you can stop price rises and ultimately drive price down. [00:28:45] Well, thank you. [00:28:46] Thank you. [00:28:46] Secretary Wright. [00:28:47] And we're as a, as a, as a company and as a team, we're committed to leaving these communities better off for us having been there and doing our part. [00:28:55] So thank you so much. [00:28:56] All right. [00:28:57] Last question. [00:28:58] Just appreciate all the time you spent with us today. [00:29:00] We know you're working on a lot of really important things. [00:29:03] You've said that government's job is to get out of the way and let private enterprise bring hundreds of billions in capital needed to lead in AI. [00:29:12] For the builders and the technologists sitting here today, what do you need specifically from them, from us? [00:29:19] What's the call to action? [00:29:21] Well, I'm going to say two things. [00:29:24] First is keep running and driving hard. [00:29:27] I know you're already doing that. [00:29:29] The capex and the development of the technology and the building of the infrastructure. [00:29:34] This is the fastest in American history. [00:29:37] As a percentage, GDP, the investment rate now is greater than the peak of building the railroads in our country. [00:29:43] As a percent of GDP, you are transforming what's possible in our country. [00:29:48] You see it from business processes, innovation, learning. [00:29:52] It's going to drive scientific progress. [00:29:54] Like you are changing the game. [00:29:56] And it is critical that the United States be the leader in this effort. [00:30:01] Not that I don't want other nations to grow. [00:30:03] They're all going to grow. [00:30:04] You're going to build data centers elsewhere. [00:30:06] But this is like the Manhattan Project. [00:30:08] The leader in this gets the huge benefits of the technology. [00:30:12] But equally importantly, if China was the leader, the clear runaway leader in AI, they would be the global superpower for the rest of this century. [00:30:23] Look, I've done a lot of business with China. [00:30:25] There's a lot of wonderful people there. [00:30:26] They've lifted people out of poverty. [00:30:28] And there's good things about China. [00:30:29] But their government does not respect human freedom, does not respect human rights the way we do. [00:30:36] We don't want China as the dominant power in the world. [00:30:39] The U.S. isn't perfect, but we've been an awesome global superpower as far as bringing prosperity and freedom around the world. [00:30:47] We want to keep doing that. [00:30:48] So you running hard and driving hard and driving this technology forward quickly, building it as much as we can in the United States. [00:30:56] That's hugely important. [00:30:57] The second thing I want to ask all of you to do, as you all well know, there's a movement that's against data centers. [00:31:05] Every demographic today says they're against artificial intelligence. [00:31:08] They're against data centers. [00:31:10] It is the same movement I faced 15 years ago in the anti-fracking campaign. [00:31:15] Even the claims are the same. [00:31:17] It uses too much water. [00:31:18] It pollutes the water. [00:31:19] It's not good for our environment. [00:31:21] It's going to drive up prices. [00:31:23] All of the same. [00:31:24] And those talking points are for a reason. [00:31:26] They resonate with people. [00:31:28] You go to someone's community and say they're going to take your water. [00:31:31] They're going to hurt the quality of your water. [00:31:32] Man, people are very rightfully very passionate about water. [00:31:36] And so these are effective arguments they're deploying against you. [00:31:41] The facts are quite the contrary. [00:31:43] As we just said, this is the road to lower cost electricity. [00:31:46] There's probably no higher value use of water full stop than there is water for these data centers. [00:31:53] The water consumption is so tiny. [00:31:55] The benefits are so huge. [00:31:57] It's going to take away your jobs and harm your communities. [00:32:00] These are the things everyone is saying. [00:32:02] And right now in the polls, they're winning. [00:32:05] They cannot win and they will not win. [00:32:07] I don't want to be alarmist. [00:32:08] We will win this argument just as we did with fracking. [00:32:11] But you're critical. [00:32:13] Everyone in this room engages with multiple people every day. [00:32:17] And those people, most everything they've heard is negative about what you're doing. [00:32:22] Respect their concerns. [00:32:23] Respect them as humans. [00:32:25] There's nothing wrong with them. [00:32:27] They've just heard one side of the story. [00:32:29] If you can honestly and candidly and sincerely engage with people. [00:32:33] Just short conversations. [00:32:35] They look in your eyes. [00:32:36] They can tell you're an honest person. [00:32:38] You've got a family at home. [00:32:39] That's how you change hearts and minds. [00:32:41] And we need to change a lot of hearts and minds. [00:32:44] Your engagement with your neighbor on the fence at the grocery store. [00:32:48] Brief engagements. [00:32:50] Believe me, that's how people change their mind. [00:32:53] Not from something they see on the internet. [00:32:55] It's from looking in someone's eyes and saying, I trust that gal. [00:32:58] I think she's telling me what's true. [00:33:00] I didn't realize that. [00:33:01] So please, just these little engagements. [00:33:04] They will multiply because they'll tell someone else. [00:33:07] So engage in this. [00:33:08] What you're doing is phenomenal. [00:33:11] You are massively improving our country. [00:33:14] You're massively improving the opportunity for quality of life of all of us going forward. [00:33:19] Of our children and grandchildren. [00:33:21] You should be over the top proud of what you're doing. [00:33:25] So please, share a little bit of that humbly and nicely. [00:33:28] But share a little bit of that pride and those facts with everyone around you. [00:33:32] It will roll over the opposition to data centers faster than we otherwise should. [00:33:37] And acknowledge there's real things. [00:33:38] Of course, when you create a powerful new tool, we're going to do a lot of awesome things with it. [00:33:42] It's going to enable some bad things too. [00:33:44] A lot of these concerns, they're not real. [00:33:47] They're just overblown. [00:33:49] The pluses are way bigger than the minuses. [00:33:52] So please, keep driving hard. [00:33:54] Keep driving these improvements in our country and the possibilities that all of us have in our future lives. [00:34:00] And don't be too shy about it when you run into people and start talking. [00:34:04] Thank you so much. [00:34:05] God bless you all. [00:34:07] Thank you so much, Secretary Wright. [00:34:09] Thank you very much. [00:34:10] Yes, thank you. [00:34:11] It's awesome. [00:34:12] All right, great to see you. [00:34:16] Okay, that was an amazing privilege. [00:34:19] Look, if we weren't ready to go break ground and build some new data centers before, which we were, I think we're all going to leave here and get our shovels out and start building some data centers. [00:34:30] So we're honored to support the Genesis mission and the Department of Energy's strategic initiatives. [00:34:36] So thank you for trusting AWS. [00:34:39] Now let's talk about AI and specifically about the gap between promise and production. [00:34:46] 95% of enterprise AI fails to reach production. [00:34:52] Only 5% of custom AI tools make it past proof of concept. [00:34:59] So how do we build AI that actually works and operates at scale reliably and securely? [00:35:08] Here's the pattern we see. [00:35:10] When electricity was first invented, factory owners took their steam engines out and put electric motors in their place. [00:35:20] It was the same layout and the same workflow, just a new power source. [00:35:25] They got maybe 5% efficiency gains and wondered what was all the fuss about. [00:35:31] It took a generation of builders to realize the opportunity wasn't replacing the engine. [00:35:38] It was redesigning the factory. [00:35:41] New floor plans, new workflows, new ways of thinking about production entirely. [00:35:48] We are at that moment with AI. [00:35:52] Too many organizations are bolting chat bots onto legacy systems. [00:35:57] That's the electric motor in the old factory. [00:36:01] It's a point solution, not a transformation. [00:36:04] True transformation means rethinking entire workflows around AI from first principles. [00:36:11] It all starts with inference, which is a key building block for applications. [00:36:17] So we thought, why not make it as simple to deploy and run a model as it is to imagine what it could do? [00:36:26] This is exactly why we built Amazon Bedrock. [00:36:30] Bell didn't transmit a voice because he had the best lab. [00:36:35] He did it because the centennial gave him a stage. [00:36:40] Bedrock is that stage, providing frontier AI for every builder without building the infrastructure themselves. [00:36:49] It handles the undifferentiated heavy lifting, model hosting, scaling, security and integration. [00:36:56] So your teams can focus on the hard problems. [00:36:59] And now it's available in GovCloud, secret and top secret regions. [00:37:05] And because no single model fits every mission, we continue to expand the model selection on Bedrock. [00:37:13] We've nearly doubled the models available on Bedrock in the last year. [00:37:17] And that includes OpenAI's chat GPT 5.4 and 5.5, which are now available. [00:37:24] These capabilities are already changing in fields like health care, particularly in a place like Singapore. [00:37:33] In Tang Fong General Hospital had a common problem. [00:37:37] Chronic disease management that didn't scale. [00:37:41] No Asian specific longitudinal data. [00:37:44] Communication gaps between appointments and care that were that was entirely clinician led. [00:37:52] So the team built Project Antenna, Asia's first population allergy database to shift from clinician led to patient empowered care. [00:38:04] AWS provided the database foundation and the AI tools with LLMs on Bedrock, driving patient communication and helped clear regulatory approval. [00:38:14] So thanks to Antenna, medication adherence went up 50%, and more patients moved from acute hospitals to community care, bringing real cost savings for patients and for Singapore's system. [00:38:28] Concept to production in weeks, and Antenna is now the blueprint for managing diabetes, hypertension, and even dementia. [00:38:38] So, so far I've talked about the infrastructure that powers your mission and the AI models that are ready for production at every classification level. [00:38:50] Now I'd like to invite Kate Zimmerman to talk about what happens when you put those two things together and leverage the full potential of AI. [00:39:00] Wow. All right. This is super cool. There's so many of you. This is amazing. Thank you, Dave. [00:39:18] So I want to start a little bit of audience participation. If you guys will indulge me. [00:39:22] So who in the room, if you can raise your hand, has used an AI assistant in the past three years? [00:39:31] Okay. I see quite a few hands going up. Good. Good. Okay. [00:39:35] So we can level set here. I'm sure you guys know how it goes then. [00:39:38] You ask a question. You get an answer. You upload a document. You get a summary. [00:39:43] And clearly this has been very useful, but it's also limited. [00:39:47] You're the one deciding what to ask, what data to provide, what system you need to interact with. [00:39:53] Agents are what changes that. [00:39:56] An agent doesn't just respond. [00:39:59] It reasons about a goal, breaks it into steps, and uses tools and acts all within well-defined guardrails. [00:40:07] It works across systems, navigates complex rules, and delivers outcomes, not just answers. [00:40:15] And agentic AI is already proving to be highly transformative. [00:40:21] In fact, one of the most exciting applications of agentic AI is in building software itself. [00:40:28] And AI-powered software development tools have evolved rapidly over the past year. [00:40:34] But as these tools became more powerful, we noticed a gap. [00:40:40] They were generating code, lots of code. [00:40:43] But builders couldn't guide the process or ensure that it was aligned with their team's standards. [00:40:49] We wanted to take what is exciting about AI-powered software development and add the structure that developers really need. [00:40:59] Say hello to Kiro. [00:41:05] Kiro works with you, turning your prompts into detailed specs and those specs into working code. [00:41:12] Let's hear from some developers who experienced what Kiro can do. [00:41:16] I use Kiro in almost all the development I do. [00:41:21] I ask it questions. [00:41:22] I create specs with it. [00:41:24] With Kiro, I was able to ship more code in the last five months than in the past 10 years. [00:41:29] With Kiro, I'm able to work with a partner, so it feels like we're collaborating on the project together. [00:41:35] It operates the way my brain operates when solving a problem. [00:41:38] I can just say, "Hey, Kiro, remember that feature we added in? Can you also write a test as well?" [00:41:44] I can be hands-off once I break the problem down and just let Kiro deliver for me. [00:41:50] I feel like my world has just opened up to a completely different perspective. [00:41:55] Everything feels possible now. [00:41:57] You can go from zero to POC 10 times faster. [00:42:00] Kiro makes me want to build more. [00:42:02] Honestly, Kiro is just awesome. [00:42:09] And for government teams, this matters in ways that go beyond productivity. [00:42:14] Every spec Kiro generates, every design decision it documents, every change it records, that is an audit trail. [00:42:22] Compliance documentation, which used to be a separate, painful process, becomes a byproduct of building software. [00:42:31] Kiro is already available in GovCloud and is coming soon to the secret and top secret regions. [00:42:39] At Loyola Marymount University, a two-person cloud team faced a challenge. [00:42:45] 500 AWS Lambda functions running an outdated Python runtime across dozens of cloud accounts. [00:42:53] Fixing this by hand would have taken two months. [00:42:57] By using Kiro to encode their governance standards and security requirements into reusable specs, [00:43:03] they built a scanner remediator tool in just hours. [00:43:07] It upgraded and tested the entire state in half a day. [00:43:11] Now, LMU ships enterprise-grade deployments four to ten times faster. [00:43:19] So while Kiro is transforming how we build software, let's think about the rest of our time when we're not writing code. [00:43:25] I don't know about you, but every day I lose hours doing work that does not require my expertise. [00:43:31] Searching for files on my computer, summarizing information because my leadership had a last-minute ask for me. [00:43:37] Context switching between meetings and conversations. [00:43:41] The way we work isn't always working. [00:43:44] Amazon Quick is built to give you that time back. [00:43:49] It's the new, better way to get work done: Quick. [00:43:55] Quick is an AI-powered assistant that connects your data and tools into a single solution with agents that can search, [00:44:02] synthesize, create, and automate across all of your work systems. [00:44:07] And we're continuing to expand what Quick can do. [00:44:11] We recently launched Amazon Quick on desktop to bring together your local files, calendar, and email, and other apps. [00:44:20] By pulling knowledge from the systems that you already use, Quick can connect the dots between people, projects, decisions, and actions. [00:44:29] For example, here is Quick helping me prep for a customer meeting. [00:44:33] It used relevant files on my local machine, like a roadmap. [00:44:36] It layered in emails and contacts from Slack threads, and even a briefing that the account team had. [00:44:42] It can create a presentation, an executive briefing doc, and fire off an email to help keep everyone coordinated. [00:44:48] No matter where work happens, Quick knows what's most relevant to you. [00:44:53] So, if Quick and Kiro are helping how we write software and our individual productivity, why should we stop there? [00:45:04] Why not build a new class of agents that help push the frontier of capability forward? [00:45:11] Frontier agents are a new class that are significantly more capable. [00:45:18] They are autonomous, meaning you direct them towards a goal, and they figure out how to achieve it. [00:45:24] They're massively scalable, able to perform multiple concurrent tasks and distribute work across multiple agents. [00:45:32] And they're long running, working for hours, even days, in pursuit of ambitious and sometimes amorphous goals. [00:45:40] We created the AWS DevOps agent to help you accelerate every stage of the software development lifecycle. [00:45:49] DevOps agent is an always-on operations teammate. [00:45:53] It resolves incidents, prevents them proactively, and handles SRE tasks across AWS, multi-cloud, and on-prem. [00:46:03] And with the recent launch of release management, your agents can now go beyond writing code to shipping it. [00:46:13] DevOps agents now write code, review it, and spin up a test environment to test it before it ships to production. [00:46:20] Every release cycle becomes tighter, faster, and more autonomous. [00:46:25] Our customers are already seeing the impact of DevOps agent. [00:46:30] Take Western Governors University, for example. [00:46:34] WGU serves 200,000 students through 24/7 online learning, making system reliability essential to student success. [00:46:43] They deployed AWS DevOps agent, integrated with Dynatrace, to autonomously investigate performance issues across their entire stack. [00:46:52] When problems occur, the agent's pinpoint root causes automatically, eliminating manual troubleshooting and freeing their lean IT team for strategic work. [00:47:06] So if that's operations, we also want to help you get security right by building it in from the ground up. [00:47:13] AWS Continuum helps you build applications that are secure from the start. [00:47:19] It scans your code for vulnerabilities and reviews design docs to proactively identify issues. [00:47:25] Continuum can even help with pen testing, helping find and fix vulnerabilities before they're found by a third party. [00:47:33] Let's see it in action. [00:47:35] It's Friday, 3:01 a.m. at a federal civilian agency. [00:47:40] Every Tier 1 analyst is asleep, and an attacker just found a way into the public portal. [00:47:45] He already has a head start. [00:47:47] In a typical security operations center, the next 47 minutes are a race the humans are losing. [00:47:53] Tonight, that's not how the story ends. [00:47:56] The actor calls itself Shadow Vertex. [00:47:59] The pattern is one we know: a web exploit against a public-facing portal. [00:48:03] The application's own credentials turned against it. [00:48:06] Reconnaissance against an S3 bucket holding millions of citizen records. [00:48:10] This is what the first 30 seconds of a real incident looks like. [00:48:14] And in most SOCs, no human will see it for 47 minutes. [00:48:18] So we asked a simple question: what if the SOC didn't sleep? [00:48:24] What if we put four AI agents on Amazon Bedrock, each with one job? [00:48:28] Each with its own identity, its own policy, its own narrow scope and responsibility. [00:48:33] Detection, analysis, containment, compliance, built on Amazon Bedrock Agent Core. [00:48:40] First, the detection agent picks up the signal in seconds. [00:48:44] A blocked request at the web firewall, corroborated by network firewall and the flow logs, and routes it. [00:48:50] Next, the analysis agent pulls cloud trail logs from the last six hours, correlates the activity against MITRE attack from a bedrock knowledge base, and computes the blast radius. [00:49:00] Three principles: one controlled data bucket. [00:49:03] 94%. [00:49:06] And finally, the containment agent proposes the appropriate action: disable the credential. [00:49:11] Deny the role. [00:49:14] Quarantine the pivot. [00:49:15] At this point, a member from the analyst team can review the recommended actions and decide whether to approve the quarantine or not. [00:49:21] However, we're not done yet. [00:49:23] The analysis has surfaced something deeper. [00:49:26] The weigh-in was a flaw in the portal itself. [00:49:29] So a different AWS agent picks up the thread: AWS Continuum. [00:49:34] It first runs a targeted pen test against the application. [00:49:37] The pen test finds several vulnerabilities, including the one used by the attacker, to mount the attack on the portal. [00:49:44] It turns out, it was a SQL injection vulnerability, and the agent was able to fully replicate the attack and provide evidence. [00:49:51] And it doesn't stop there. [00:49:53] It flags the issue. [00:49:54] Identifying the root cause in the code and submits a PR to fix it. [00:49:59] The analyst can now review the code fix and approve accordingly. [00:50:02] This is how AWS Continuum augments the human potential. [00:50:06] But Continuum goes beyond just addressing security vulnerabilities in production by providing code patches. [00:50:12] One of the biggest benefits of Continuum is that it allows us to shift even further left and capture security issues long before the code is ever deployed. [00:50:21] AWS Continuum code reviews can continuously scan our entire code base and surface all security issues at design time, adding multiple layers of defense. [00:50:31] This means that we could have caught our SQL injection error at design time, long before it hit our production environment. [00:50:38] 45 seconds from exploit to containment. [00:50:41] One human-approved click. [00:50:43] Agent actions logged. [00:50:44] Decisions reconstructable. [00:50:46] AWS Continuum can surface the risk, prove it is exploitable, and drive the fix at machine speed and under your control. [00:50:54] The future of the SOC is not coming. [00:50:56] It is here. [00:50:57] And it is running on AWS. [00:51:01] To every agent I've shown you today, Kiro, Quick, DevOps Agent, Continuum, they all need the same things to work in production. [00:51:18] Persistent memory, identity and access controls, observability, and the ability to scale. [00:51:24] Building that infrastructure every time for every agent is exactly the kind of undifferentiated heavy lifting we believe you shouldn't have to do. [00:51:37] And that's why we built Amazon Bedrock Agent Core. [00:51:40] To help build, connect, and optimize production agents securely at scale. [00:51:45] It provides the core components needed to build agents, a managed runtime, built-in session memory, secure authentication to tools, APIs, and data sources. [00:51:56] And with the new web search on Agent Core, your agent can ground its response in current, accurate web information with zero integration work needed. [00:52:05] You can use Agent Core with any framework, like Strang, SlangChang, Crew AI, or even your own custom code. [00:52:15] It's available on both commercial and GovCloud, and enables you to connect to any model. [00:52:20] We do not force you into a single fixed path. [00:52:22] So whether you're developing with autonomous agents, or training foundation models, or running your most demanding workloads at scale, [00:52:32] we are focused on making AWS the best place to build. [00:52:36] Because what's most exciting for us is seeing what you build when you have the freedom to invent anything. [00:52:43] And nobody is navigating that freedom more thoughtfully right now than governments who are rethinking how they serve citizens at scale. [00:52:53] Next, I'd like to welcome Sonia Patel, Chief Technology Officer for the United Kingdom government, [00:53:00] to talk about what it takes to actually succeed with AI at national scale. [00:53:05] Well, good morning, USA. [00:53:18] Well, it's a great time to be in this fabulous country celebrating 250 years of independence, [00:53:26] but also being blessed with our beautiful game. [00:53:29] So 30 years of following football has taught me one thing. [00:53:33] Hope for the best, plan for the worst, and always have a backup plan. [00:53:39] Which, as it happens, is quite a good summary of leading technology in government. [00:53:45] So four months ago, I started the role of Government Chief Technology Officer for the United Kingdom. [00:53:51] On my first day, someone handed me a briefing pack and said, [00:53:55] "Here's everything you need to know about the technology estate." [00:53:58] It was enormous. [00:54:00] I considered asking for an executive summary, only to be told, "This is the executive summary." [00:54:08] So my organization's responsible for setting the technology direction for hundreds of departments and agencies [00:54:15] at national and local levels across the United Kingdom. [00:54:20] thousands of systems, decades of technical debt, and 67 million people who depend on it every single day. [00:54:33] But what I learned there, and what I want to share with you this morning, it's not about the technology per se. [00:54:42] It's about three shifts in how you lead when the stakes are mission critical. [00:54:48] So the first shift is from code to narrative. [00:54:53] So early in my career, much like those in yours, success meant technical expertise. [00:55:03] Can you design the architecture? [00:55:05] Could you debug the system at 2:00 a.m. in the morning? [00:55:10] That mattered. [00:55:11] It still matters. [00:55:13] Though I do sleep better now. [00:55:15] But when you move from building systems to leading transformation, the skill that changes is this. [00:55:24] Creating meaning. [00:55:25] Not writing code, but writing the story that connects people to purpose. [00:55:33] So here's an example. [00:55:36] We recently launched Gov.uk chat. [00:55:41] Experimental, AI-powered, in the Gov.uk app. [00:55:45] Now, this helps users get answers from over 8,000 pages of government guidance quickly, accurately, and in the way they would write or speak in everyday life. [00:55:58] But if I described it as RAG-enabled LLM deployment on managed infrastructure, which it is, nobody outside this room cares. [00:56:08] So what people care about is this instead. [00:56:12] Instead of waiting on to hold for a government call center, some of which handled thousands of calls a day, a new parent can now ask at midnight, [00:56:22] "I've just had a baby. [00:56:23] What support can I get?" [00:56:25] And get the answers in seconds. [00:56:30] And before its release, more than 10,000 users took part in pilots. [00:56:35] Early accuracy scores, around 76%. [00:56:38] Then fast forward 18 months, and we are now seeing 90% accuracy in production. [00:56:45] And we're building on these foundations for what comes next. [00:56:48] So 14 million people use our one login across 120 government services. [00:56:54] And we're working towards a world where AI agents can orchestrate across departments. [00:57:02] Pulling in data from one, verifying identity in another, and then completing transactions in a third. [00:57:09] All on behalf of the citizen who asks one single question. [00:57:14] That is the shift. [00:57:16] The code gets you started. [00:57:18] The strategic narrative and the focus on outcomes gets you funded, gets you supported, and wins you users. [00:57:27] Now the second shift is understanding. [00:57:31] When you're delivering at enterprise, you're delivering more than just a product. [00:57:36] You're operating within a complex system of systems. [00:57:40] And the UK government, much like most governments, it's not one system or one organization. [00:57:46] Enterprises hardly are. [00:57:48] It's hundreds of departments, agencies, local authorities with different cultures, different processes, diverse set of services, different risk appetites, systems, and priorities. [00:58:03] So change at this scale is not about choosing the best technology. [00:58:07] It's about understanding the ecosystem, building alignment, applying the right technology to resolve real problems. [00:58:17] Now we're cloud agnostic by policy. [00:58:20] We use multiple providers across our state. [00:58:24] And that is by design. [00:58:25] But being cloud agnostic does not mean every platform delivers equally against every requirement. [00:58:33] For us, it means three non-negotiables: native security and compliance, AI as a first-class capability, and the ability to operate at public sector scale. [00:58:51] But system-level thinking isn't just about infrastructure. [00:58:54] It's about unlocking real outcomes across very different parts of government. [00:59:02] Now I've been here before leading the technology for the National Health Service in England. [00:59:08] England's National Lung Cancer Screening Programme runs on AWS. [00:59:13] Over a million people have been invited for screening so far. [00:59:17] And more than 3,000 lung cancers have been detected. [00:59:21] And now 74% of them at an early treatable stage. [00:59:27] That is what infrastructure looks like when it works. [00:59:30] Bringing cloud computing, AI tooling, and clinical expertise together in a meaningful way [00:59:37] so a radiologist can review a scan taken hundreds of miles away [00:59:42] and a patient gets a diagnosis early enough to do something about him. [00:59:47] So once again, the point is not about technology. [00:59:52] The point here is system-level change in a big ecosystem requires system-level thinking. [01:00:00] You have to understand the cultures, the behaviors, the incentives, the blockers, [01:00:06] and design for the ecosystem, not just for the end user. [01:00:10] And the third shift is the hardest. [01:00:15] Moving from personal performance to constructive outcome-orientated partnerships. [01:00:21] So early in my career, I was rewarded for what I delivered. [01:00:26] That bit of code, that bit of architecture. [01:00:29] Now, I'm rewarded for what I enable others to deliver. [01:00:34] And that requires a fundamentally different relationship with control [01:00:39] and with the people outside traditional boundaries. [01:00:42] So let me give you the clearest example we have. [01:00:46] Legacy technology is still the single biggest break on modern AI-enabled public services. [01:00:56] More than a quarter of UK government systems are out of date. [01:01:00] So we have built a legacy modernization community, [01:01:05] which is now opening up conversations that we were simply not having before. [01:01:09] AWS is an active member of that community. [01:01:17] And last month, we ran our first hands-on legacy hackathon with AWS in our offices, [01:01:24] bringing together various government organizations, [01:01:29] one room using AWS transform to refactor real legacy code. [01:01:35] The feedback one organization shared with us, [01:01:40] what would normally have taken days, now takes minutes. [01:01:44] That is the partnership on the debt plate. [01:01:51] And as we build in this era of artificial intelligence, [01:01:56] we are in an era of building an intelligent government. [01:02:00] The question is no longer whether to adopt the AI or technology choices we are making. [01:02:08] It is how to lead its adoption responsibly, at scale, [01:02:13] and in ways that genuinely serve the people who depend on us. [01:02:20] This stuff isn't easy. [01:02:22] I will not pretend as government we have all the answers. [01:02:25] But the new mother searching for support at midnight, [01:02:29] the person opening their cancer screening results, [01:02:32] they are counting on us to be deliberate, pragmatic, responsible, [01:02:38] for delivering what matters most to them. [01:02:42] So our role as technologists, my role, my story, [01:02:47] is to build the conditions where it works for everyone, everywhere. [01:02:53] Thank you very much. [01:02:54] Thank you very much. [01:03:04] Thanks, Sonia. [01:03:05] 67 million people, hundreds of departments, [01:03:10] and a team that decided to stop waiting for perfect and start shipping. [01:03:15] What I love about Sonia's story is that it's not about one application. [01:03:21] It's about building the system that lets every department move faster. [01:03:26] That's the multiplier. [01:03:28] We talked about purpose-built infrastructure, [01:03:32] AI that works in the real world, [01:03:34] and agents that reason, plan, and act within guardrails. [01:03:40] But none of it matters if you're spending 70 cents of every IT dollar just keeping the lights on. [01:03:49] 70% of enterprise IT budgets trapped in legacy maintenance. [01:03:55] Systems outdated before most leaders even took the job. [01:04:00] Mainframes processing millions of transactions. [01:04:05] Cobol running core banking. [01:04:07] Databases holding tens of millions of records. [01:04:12] These systems work. [01:04:14] They're not broken, but they're brittle, they're expensive, [01:04:19] and leave almost nothing for what you need to build next. [01:04:23] Modernization doesn't have to be a big bang. [01:04:27] The most valuable transformation often starts with the boring stuff. [01:04:33] The system nobody wants to touch. [01:04:35] The database that nobody wants to migrate. [01:04:38] The mainframe that runs on institutional knowledge and a prayer. [01:04:43] You know, back in 1876, [01:04:46] half of those exhibits replaced something that already existed. [01:04:52] The typewriter didn't create writing. [01:04:54] It transformed it. [01:04:56] Modernization isn't replacement. [01:04:59] It's transformation. [01:05:01] That's exactly why we built AWS Transform. [01:05:06] Transform uses AI to accelerate the modernization of legacy platforms like VMware and mainframes. [01:05:13] It's already saved over 1.6 million hours of manual migration effort and transformed over 1 billion lines of mainframe code. [01:05:24] And while we knew helping you modernize faster would be popular. [01:05:29] It turns out that there's a really long list of legacy code that you want to transform. [01:05:36] Like Python, Node.js, C to Rust migrations, and more. [01:05:41] And you need custom tools. [01:05:43] That's why we built AWS Transform custom. [01:05:48] Now you can create custom transformation agents that can modernize any code. [01:05:53] API, runtime, even programming languages and frameworks that are unique to your organization. [01:06:00] It continually improves accuracy by learning from every execution and code change and making each transformation more reliable and efficient. [01:06:11] So with Transform, you can finally tackle those migration projects and get back to building features that matter. [01:06:19] But we knew we could make it even easier. [01:06:23] Instead of manually configuring these transformations, you can now let agents tackle your technical debt and keep your code base modern. [01:06:31] The AWS Transform continuous modernization is a new feature that works in the background. [01:06:37] Finding the debt, fixing it, validating the fix, and learning from every transformation to make the next one better. [01:06:46] It plugs into your existing pipeline tools like CodePipeline, Jenkins, GitHub Actions, and GitLab. [01:06:55] We meet you where you are. [01:06:58] Now let's see how some of our customers are leveraging Transform for their modernization efforts. [01:07:03] The Defense Counterintelligence and Security Agency is moving from legacy infrastructure that constrains their mission [01:07:10] to a modern cloud-native platform on AWS GovCloud, projecting $114 million in savings over the next three years. [01:07:22] Edemia is another company that has seen significant benefits from AWS Transform. [01:07:29] Their identity platform, used by the Department of Motor Vehicles in over 45 states, was running on .NET 3.5. [01:07:37] End of life, monolithic. [01:07:40] Every update took 32 hours to deploy across 16 state and federal environments. [01:07:46] Using Transform for .NET, they modernized from .NET 3.5 to .NET 8 at four times the speed of a manual rewrite. [01:07:57] They also broke their monolithic solution into microservices on EKS and migrated a five-terabyte database to Aurora, Postgres, SQL. [01:08:08] With this modernization, they achieved a 30% reduction in total cost of ownership and improved recovery time from hours to minutes. [01:08:20] So we have the infrastructure. [01:08:21] We have the AI services and the modernization tools to accelerate innovation. [01:08:26] But through working with thousands of organizations, we've learned something that having the right technology and the right people still isn't enough. [01:08:37] Getting AI from pilot to production at enterprise scale breaks on speed, repeatability, and operational rigor. [01:08:47] The last mile isn't a skills problem. [01:08:51] It's an acceleration problem. [01:08:53] We didn't just observe that. [01:08:55] We lived it. [01:08:56] Since 2017, we've been embedding engineers alongside customers across the industry, building and moving AI solutions to production in days and weeks instead of months. [01:09:09] This experience is now the foundation for something purpose-built. [01:09:17] Today, I'm excited to announce AWS Forward Deployed Engineering. [01:09:22] It's a new business unit designed to embed AWS engineers directly with your team. [01:09:28] These aren't advisors. [01:09:38] They don't just hand you a deck and leave. [01:09:40] They are engineers from our frontier AI teams, including the ones behind the services you use daily. [01:09:47] These engagements are outcome-based, delivering production systems that actually work. [01:09:55] They run on AI-driven development lifecycle where delivery agents handle design, build, security, and deployment. [01:10:04] And more importantly, each engagement builds a knowledge base with your architecture decisions. [01:10:11] Domain rules and operational patterns all owned by you. [01:10:15] Every workflow ships faster than the previous one, compounding expertise with every release. [01:10:22] We're backing this with a billion-dollar investment, scaling our engineering teams and accelerating customer outcomes alongside our AWS partners. [01:10:33] This is how we close the last mile gap between AI ambition and actual production reality. [01:10:41] And now, to hear more about how the world's premier intelligence organization is harnessing the power of technology [01:10:51] to protect national security and push the boundaries of what's possible with cloud and AI. [01:10:58] It is my distinct pleasure to introduce the Honorable John Ratcliffe, Director of the Central Intelligence Agency. [01:11:14] Thank you, Director Ratcliffe. [01:11:15] Appreciate it. [01:11:20] Well, good afternoon, everyone. [01:11:22] Dave, thank you for the introduction. [01:11:25] It's a pleasure to be here and an honor for me to be able to speak to such a distinguished group. [01:11:31] Now, we all know that sometimes speakers get up and, to be polite to the audience, they tell them that it's a distinguished group. [01:11:38] In my case, I really do mean it. [01:11:41] I think many of you know that I don't speak publicly very often. [01:11:47] I'm doing so today because with this audience, whether you're from the public or private sector, many of you in this room play a key role in advancing American innovation and encountering some of the most daunting technological problems of our time. [01:12:08] At CIA, we take very seriously our responsibility to do the same. [01:12:13] In fact, so much so that every few months I have the privilege of swearing in new officers at Langley, I underscore to them the grave responsibility that we all share by reminding them that their purpose at the CIA is not to be somebody, but rather to do something. [01:12:35] In the lobby of the CIA headquarters is our famous memorial wall. [01:12:41] There are 141 stars etched into that marble edifice. [01:12:44] Each one pays tribute to a CIA officer who made the ultimate sacrifice on behalf of their country. [01:12:52] The stars tell 141 unique stories, but they all share one fundamental characteristic. [01:12:58] They honor patriots who wanted to do something, not to be somebody. [01:13:05] And it remains our responsibility at the CIA to carry forward their legacies. [01:13:09] So today I want to focus on what the agency has collectively done and what we're continuing to do to drive our critical mission forward, and in particular, how we're adapting to breakthroughs in technology that are revolutionizing the way that we all live and the way we all work. [01:13:30] This will require bold action and innovation. [01:13:35] And to be clear, these are not new to the CIA. [01:13:38] These are characteristics that are woven into the very fabric of our organization. [01:13:43] They are what set us apart as the number one intelligence organization anywhere in the world. [01:13:48] From our legendary espionage of the Cold War to more recently since I've been here, our audacious support to military operations and successes in Venezuela and Iran are testimony to this. [01:14:05] In Venezuela, the window to capture President Nicolas Maduro was measured in mere minutes. [01:14:13] So the intelligence picture had to be crystal clear. [01:14:16] And thanks to CIA, it was. [01:14:19] And it enabled a remarkably successful operation by American special forces who were able to locate and apprehend the targets in less than four minutes after their feet hit the ground on the largest military complex in that country. [01:14:34] In Iran, CIA first provided flawless intelligence picture that was necessary last June in Operation Midnight Hammer to get more than 120 US aircraft in and out of Iran and precisely deliver payloads against Iranian nuclear facilities before the enemy even knew we were there and was able to respond. [01:14:55] And then again, just two months ago, CIA played another pivotal role during Operation Epic Fury in rescuing the crewmen of that F-15 strike eagle that was shot down on Good Friday. [01:15:08] We launched a search that was, as President Trump said, the equivalent of trying to find a needle in a haystack. [01:15:16] It was a search that confounded the Iranians and despite their efforts to stop us and despite their home field advantage, it was a search that rested on our innovation, creativity and our technological know-how. [01:15:29] And ultimately, it was a technology-enabled search that only the CIA could successfully and did successfully pull off. [01:15:41] But as you scan the globe right now, you can see that there's still no shortage of conflicts and crisis out there. [01:15:57] So this isn't really a time where we have afforded the luxury of being able to focus on our past successes. [01:16:04] Increasingly, all of our future successes are going to depend on technology. [01:16:08] We have to continue to push the boundaries of what's possible because the nation that best harnesses the power of technology will determine the global future. [01:16:23] That's why when I became the director, I made clear that emerging technologies right up there with China were going to be our top priority. [01:16:33] 18 months later, that focus hasn't changed, but what has changed is the rate at which technology is evolving. [01:16:42] Even by the standards of just last year, the pace of change, I think you'd agree, is astonishing. [01:16:48] Advances in AI and quantum and biotech are not just going to profoundly impact economies around the world. [01:16:57] They are, as we speak, rewriting the reality of conflict and asymmetric warfare. [01:17:06] A great illustration of this is how AI powered robots and drones, on the one hand, can help our manufacturing facilities with exponentially better speed and efficiency do their jobs. [01:17:21] But on the battlefield, their efficiency and their lethality can be equally devastating. [01:17:29] Perhaps you saw this from open source reporting from just a few days ago. [01:17:33] That estimates that the average life expectancy of a Russian recruit who reaches the battlefield in Ukraine is estimated to be between 20 and 35 minutes. [01:17:48] Minutes. [01:17:51] Much of the reason for that is technology and how drones have become super-efficient, low-cost killing machines. [01:18:00] Many of you in the private industry see this as you lead the technology developments driving the biggest societal shift in our lifetimes. [01:18:08] Our partners across the government certainly see it as we leverage these cutting-edge tools to support our national security priorities and objectives. [01:18:20] But our adversaries across the globe also see it as they work to steal and to manipulate America's advancements for their own ends and gains. [01:18:31] President Trump and this team are determined not to let that happen and we have repeatedly turned to the CIA as one of the greatest safeguards against these kinds of threats. [01:18:43] So when it comes to technology, what are the steps that we've taken to tackle these challenges and to ensure that we remain the preeminent intelligence agency, not just now but for generations to come? [01:18:56] Well, first, we're improving our communication channels between the CIA and our private industry partners. [01:19:03] We recognize that when it comes to partnering with the private industry, the CIA hasn't always been the easiest agency to work with because of both our security requirements and because we haven't always coordinated our outreach to individual companies very well. [01:19:20] I know many of you in the audience have experienced and can attest to that firsthand. [01:19:26] So to help ease the burden, we've dramatically shortened our timelines for onboarding new technology. [01:19:32] So before I was sworn in, in the average timeline that we had to acquire a new piece of technology at the enterprise level was as long as 24 months. [01:19:42] That was followed by an additional nine months to complete a security assessment. [01:19:48] So the whole process often took three years or even more. [01:19:52] By that time, the technology had become outdated. [01:19:55] I said to my team, "Guys, this is crazy. [01:19:58] Clearly, this model isn't good enough. [01:20:01] We have to do better." [01:20:02] And we have. [01:20:04] So now, with our new acquisition framework and a new procurement executive, we're moving forward with the goal of completing most of our acquisitions within six months total. [01:20:17] We're stripping away all of the cumbersome red tape. [01:20:20] We're delegating down to the lowest possible level to ensure that those who are closest to the issues are the ones who are making the decision. [01:20:29] What's the result? [01:20:31] Almost 400 acquisitions in just the last six months, something that previously would have taken only God knows how many years. [01:20:41] Another step we've taken is to provide a more structured approach to tap into the innovation that many of you in the private sector have to share. [01:20:51] We've established the Office of Corporate Partnerships to give our private industry partners a single point of access to deal with the agency. [01:21:00] So, some of you may recall, during my confirmation hearing 18 months ago, I talked about the fact that the United States was the only country in the world that could parallel park [01:21:10] a 200-foot rocket booster, but also the fact that we could only do that because of partnerships where we draw from the private sector innovation that's out there from companies like, in that case, SpaceX. [01:21:24] So, early in my tenure, I invited Elon Musk out to CIA. [01:21:28] I did the same with leaders from so many of the other great technology companies that are out there and that are here today, Amazon, Google, Dell, so many others. [01:21:38] Another big movement that we've recently undertaken, that some of you are already tracking, is the transformation from our Directorate of Digital Innovation to now the Directorate of Mission Systems, or DMS. [01:21:52] So, unlike DDI, DMS doesn't have offensive cyber or open source duties and responsibilities. [01:22:02] Instead, we've streamlined our efforts, focusing on core functions like cybersecurity and advanced data and infrastructure services. [01:22:10] Why is this important? [01:22:12] Simply put, it will dramatically strengthen the foundation of our entire information technology architecture. [01:22:20] This focus is really an evolution for the CIA. [01:22:24] We recognize that we must have the most advanced and resilient technology foundation. [01:22:30] We must draw from the innovation that exists in the private sector and quickly integrate that into our agency systems. [01:22:38] And that we must give our new tools to every officer in every position at the speed of mission required to do the job well. [01:22:48] As part of this transformation, we're undertaking an aggressive data sprint right now, as we speak, to enhance the discovery and exploitation of all of our mission data. [01:23:02] We will drive data standardization across the entire agency, increase our ability to better integrate all of our holdings, and train our officers on how to use all of our new capabilities. [01:23:16] If all of this sounds to you like, that are listening, sounds like these are just redrawn lines on an organizational chart, I promise you it isn't. [01:23:25] This is the fundamental reshaping of the CIA's entire approach to technology. [01:23:31] We have to be better positioned than ever to put forth a strong defense and offense against our adversaries. [01:23:39] Which is why we have also, as many of you are tracking, also recently elevated the Center for Cyber Intelligence into a mission center. [01:23:49] Look, we need to protect not only our physical borders in this country, but just as importantly, our digital borders. [01:23:57] By wielding both a sword, with regard to CCI, and a shield with regard to DMS, to deter, to degrade, and to disrupt attacks on our critical infrastructure. [01:24:11] At the end of the day, you can only be the greatest kinetic power in the world if you can protect your command and control systems. [01:24:21] If you don't control those, you may find that your adversaries are the ones in charge. [01:24:27] Finally, our widespread adoption of AI is literally transforming the way we do business. [01:24:35] And it must. [01:24:37] Because worldwide advancement of AI tools will only continue to raise the stakes in our competition with all of America's adversaries. [01:24:45] In conversations with many of the president's other national security and economic security advisors, we're talking about the impact of these frontier AI models. [01:24:57] And it would be, as we've talked about, not misplaced to refer to their capabilities as akin to digital nuclear weapons. [01:25:07] While I can't predict just how fast AI and the other emerging technologies will advance or how far they're going to take us, [01:25:17] what I can say is that we're going to do everything we can to deliver all of the top tools necessary for our officers to succeed to keep America safe. [01:25:27] What we're not going to do as we test the limits of what is possible at the CIA is to let perfect be the enemy of good. [01:25:38] We're going to take smart risks, we're going to experiment, and then we're going to course correct as we go. [01:25:45] We simply can't afford to wait for a risk-free approach when it comes to emerging technologies. [01:25:52] It doesn't exist. [01:25:53] We have to move fast, we have to be aggressive, and we have to take full advantage of the ingenuity that sets America apart. [01:26:02] That's the only way to make sure that the CIA continues to operate at the cutting edge of technology. [01:26:09] It's for these reasons that the emerging technologies, and AI in particular, is a domain in which the CIA must excel, [01:26:18] because every algorithmic decision has implications for US strategic advantage and for the national security of all of our people. [01:26:29] When President Trump first talked to me about joining his cabinet, I didn't hesitate to say that CIA is where I wanted to serve. [01:26:36] From my time previously as the Director of National Intelligence, I knew that CIA was a national treasure, [01:26:43] but I also thought that it was being underutilized, that it was being held back by structural challenges, [01:26:50] and by a culture that too often emphasized process over speed and bureaucracy over agility. [01:26:59] I knew that, properly deployed, CIA would deliver for our country in a way that no other agency can, [01:27:08] and that it was truly indispensable to the strength and the security of our country. [01:27:13] 18 months in, I'm more convinced of that now than I was before. [01:27:18] I firmly believe that if the United States is to be the dominant global superpower, [01:27:25] the CIA must remain the best intelligence agency anywhere in the world. [01:27:32] And that's why the changes that I've shared with you today are so important. [01:27:37] So, we must not just incorporate emerging technologies, we must master them better than our adversaries. [01:27:48] For us to be able to conduct and to enable both human and technical collection of intelligence, [01:27:56] more CIA officers are going to have to become just as comfortable handling lines of code [01:28:01] as they are with handling human assets and sources. [01:28:06] Now, to be clear, while technology is critical to our mission success, [01:28:11] the choices made by human beings will still determine the direction that we go. [01:28:16] Good intelligence is always going to require good judgment, [01:28:21] and only people can and should decide which is the right way to go. [01:28:27] So, we must always embrace technology as an embedded and lasting part of the work that we do at the CIA. [01:28:37] One that spans every mission and every location where we are and do our work around the world. [01:28:44] One that helps us best our adversaries at every turn. [01:28:48] And one that will remind our officers, when they survey the global landscape years down the road, [01:28:57] that they weren't just trying to be somebody. [01:29:00] That instead, they remain focused on that far more critical mission of doing something. [01:29:07] And that in the end, they did something. [01:29:10] Something very important. [01:29:12] And that they did it very, very well. [01:29:15] So, thanks so much for having me today. [01:29:28] Thank you, Director Ratcliffe. [01:29:30] It's inspiring to see how CIA is leveraging technology to meet the complex challenges the intelligence community faces today and tomorrow. [01:29:42] AWS is proud to be a partner in that mission, and we're grateful for the trust you've placed in us. [01:29:49] We covered a lot of ground today. [01:29:51] We saw infrastructure purpose-built for the mission from commercial all the way to the most sensitive workloads. [01:29:58] We saw how AWS enables you to move AI from pilot to production securely and at scale. [01:30:07] We met a new class of agents that reason, plan, and act within guardrails. [01:30:14] And we saw how modernization can help you transform legacy systems into modern platforms. [01:30:20] But if I step back and I think about what ties all of this together, I keep coming back to Philadelphia, 1876. [01:30:32] 10 million people showed up that summer because the culture said the future is here and it belongs to anyone who walks through the door. [01:30:41] Bell didn't ask permission to transmit a voice over a wire. [01:30:45] Shoals didn't wait for someone to tell him the world needed a typewriter. [01:30:51] They saw a problem, they saw tools, and they said two words. [01:30:56] Why not? [01:30:59] That was America at 100. [01:31:01] At 250, we have more capacity at our fingertips than every exhibit in that hall combined. [01:31:09] So the only question left to answer is, what are you going to build? [01:31:16] Thank you and enjoy the summit today.