Azure Cosmos DB Conf 2026 Keynote

0:07 Well, you know what?

0:08 I think we've set you all up and it's now to get you to our opening keynote.

0:12 Yeah, that's right.

0:13 Here's Kirill Gavrylyuk, Vice President of Azure Cosmos DB at Microsoft.

0:18 Welcome to Azure Cosmos DB Conference 2026.

0:25 We're really grateful to have you here and let's take a look back

0:30 at the year since the last conference and what a world year has it been.

0:35 Of course.

0:36 We're all building AI apps at amazing pace, right?

0:40 And for Cosmos DB perspective, we invested heavily in semantic search.

0:44 That is core part of any AI app.

0:47 You have now full-text search, hybrid search, vector search,

0:51 semantic ranker, plethora of capabilities with confidence, we can say.

0:55 If you're using Cosmos DB, you do not need a separate search system.

0:58 Just use built-in capabilities and take advantage of cost efficiency,

1:03 performance, scale, reliability, and precision.

1:07 And of course, more than half of our customers

1:10 are using coding agents today to build apps on Cosmos.

1:14 And for that, we offered great agent kits

1:17 with skills to make your coding agent world-class expert.

1:21 We added MCP servers and other capabilities to make this possible.

1:26 Now, AI does not remove the need for reliability, security, performance.

1:32 And that is our priority number one.

1:34 This is where we invest the most of our energy.

1:37 We've provided throughput management,

1:39 fleet management for customers with large fleets of Cosmos DB.

1:43 We've invested with simplified partitioning,

1:46 with global security indexes, hierarchical partition keys.

1:49 We've invested in per-partition auto failover,

1:52 and now Cosmos DB is the only database

1:55 that offers you five-nines reliability for any consistency mode.

1:58 And of course, security is number one priority.

2:02 and a number of improvements in security space.

2:05 Last year, we've added second NoSQL database to Azure, DocumentDB.

2:11 Unlike Cosmos, DocumentDB is fully open source.

2:14 It's based on an open source project at Linux Foundation where Microsoft,

2:19 Amazon, Google, and 12 other vendors are collaborating.

2:23 DocumentDB has a number of advantages over other document databases and notably,

2:29 more than 40% cheaper than any other document database,

2:33 including MongoDB Atlas on AWS.

2:35 So please come over to Azure, use DocumentDB,

2:38 take advantage of performance and low cost.

2:42 Now, AI transformation is on top of everyone's mind, right?

2:45 And for databases really means, in my opinion, three things.

2:49 One is AI increase the importance of flexible data.

2:55 The data AI processes and emits is all semi-structured, right?

3:00 It's prompts, it's memory, context.

3:02 That's what AI is focused on.

3:04 And Cosmos DB provides the first-class support for semi-structured data.

3:11 It proudly presents flexible data model.

3:14 Now, AI accelerated the pace at which we develop applications.

3:19 And flexible data is paramount for you to enable this pace.

3:23 Developers cannot be locked down by strict schemas.

3:27 They need schema flexibility, which is again,

3:29 number one capability of Cosmos DB.

3:31 That's why we build the database is

3:34 to provide you schema-less flexibility in data modeling.

3:37 Now, AI brought Semantic Search as a first-class operator into queries.

3:43 And it's true for any database that matters.

3:47 And of course, in Cosmos DB, we invested far beyond of that.

3:51 We added Semantic Search.

3:53 full-text search, vectors, hybrid, semantic ranker.

3:56 We invested in GraphRack, and we're about to announce Agentic Retrieval,

4:00 which takes the GraphRack to the next level,

4:03 where the graph, knowledge graph, can be inferred by LLM for you,

4:07 without you having to manually construct it.

4:09 Now, the third important thing about AI transformation is

4:13 we all now are accompanied by our coding agents.

4:17 That's our favorite friends now, right?

4:20 Coding agents.

4:21 speed up development, and for coding agents to be efficient,

4:25 world-class experts in databases, they need skills.

4:28 And that's why we invested in Cosmos DB Agent Kit.

4:30 We'll talk about it later.

4:32 As it provides skills to your coding agent,

4:35 no matter which coding agent you use, along with MCP servers,

4:40 plugins for any favorite coding agents that you may choose.

4:44 Now, let me introduce our guest speaker, Guillermo Rauch.

4:51 founder and CEO of Versel, a popular AI app development cloud.

4:57 And Guillermo knows a thing or two about what AI apps need from databases.

5:04 It's an honor to have with us CEO and founder of Versel, Guillermo Rauch.

5:10 Guillermo, you have built libraries like Socket.io and HGS

5:14 that form the modern web as we know it today.

5:19 On top of it, you've built a highly successful company, Versel,

5:24 that has seen a really good growth

5:27 of developers ah that provides developer cloud.

5:31 What do you credit Versel's success?

5:34 I think if I look back on the principles that we created for SellOn,

5:38 it was really about the ease of use.

5:41 At the time it was ease of use for humans.

5:43 Now, lately I've been calling it agent ergonomics or ease of use for agents.

5:48 It's about, you know,

5:49 you have an idea and you need to bring it online and you actually

5:54 want it to scale and work really well and be a really high quality product.

5:59 if it could be summarized as something simple,

6:02 would be the focus of ease of use, but in the service of quality,

6:06 not just getting slop, as we call it,

6:09 this taste online that crashes and pages you at midnight.

6:13 Makes sense.

6:15 As you transition to agents as your customers,

6:18 ah what surprised you is what's happening in the industry?

6:23 It's really the scale.

6:27 As an entrepreneur, as a founder, you start doing like...

6:30 market sizing estimates when you're preparing your early

6:33 decks for your investors and would be investors.

6:36 And the common wisdom was that, you know, even being generous,

6:41 maybe there were 30 million people that could deploy applications because

6:46 we targeted the easiest programming language that we could think of JavaScript,

6:51 TypeScript, trying to make the cloud super accessible,

6:53 but it was still tens of millions of people.

6:57 And Next.js also lowered the barrier to entry, to React into this universe.

7:02 But when you think about agents,

7:05 they're empowering everyone on the planet to create software.

7:09 They might not even be realizing that they're creating software.

7:11 They may just ask for a solution to a problem.

7:14 And in the process, the agent writes up

7:17 an application and deploys it on our cloud.

7:20 So a lot of the growth that we're seeing these days is Literally,

7:23 by the way, happened yesterday.

7:24 A high school friend of mine came to visit me.

7:27 He'd never written a line of in his life.

7:29 And he's like, by the way,

7:31 I know what Vercel does now because Claude Code deployed

7:35 a Vercel and told me the application was now online.

7:38 So I could have never imagined that we would

7:42 see an increase in, let's call it a hundred times,

7:46 the addressable market of creators of software, ah maybe even higher.

7:51 ah I think will be soon in the future where software is like water.

7:57 It's just a normal thing to create, consume, sell, ah and it's everywhere.

8:03 That's amazing.

8:04 And as the platform that is uh so frequently used by the agents,

8:10 what do you expect of the modern database?

8:13 How can we help you?

8:15 Well, I'm very thankful.

8:16 I always tell the team.

8:17 because part of growing up as a company is

8:20 telling the stories and scars of the early days.

8:23 We started out using a database that was very managed by me in the early team.

8:30 And it was a horrifying experience.

8:32 And I reached out to you uh and a very thankful Kirill,

8:36 through an introduction from Nat Friedman,

8:38 because I was looking for a solution that would help

8:41 me focus on growing Vercel and not being a DBA.

8:45 We had very few people on staff.

8:46 This is literally the garage story.

8:49 And one of the convictions that I had that helped me

8:53 choose Cosmos and be savvy about this, because I'll tell you,

8:57 some of the less familiar developers look at query languages

9:00 and like they might have a little bit of culture shock sometimes.

9:03 But one of the things that convinced me is that compute was becoming serverless.

9:09 The value of Vercel was you deploy,

9:12 it scales to zero, It scales to billions of people.

9:16 I literally am supporting applications these days that were vibe

9:19 coded and go viral all over the Internet and have traffic

9:22 that even teams of engineers would have never dreamed off and can

9:26 now be sustained by someone that barely knows how to prompt.

9:30 And the magic there is that it was that bet on serverless.

9:34 Because I also tell you,

9:35 the reality is that a lot of this software is quite ephemeral.

9:38 That's being written by the agents.

9:40 Maybe it's useful for a day.

9:42 Maybe it's useful for a presentation.

9:44 We're seeing a lot of sales engineers,

9:46 including people at Microsoft use vZero to vibe code an app for a sales pitch,

9:51 for a demo for pre and post sales to demonstrate an integration of a product.

9:56 And so if software is ephemeral,

9:58 serverless is scaled to zero and the flexibility there is extremely useful.

10:03 So I'm very thankful that we made the right bet.

10:06 The other thing that I wanted to be super mindful

10:10 of that serverless compute helps a lot with is fault isolation.

10:14 So I didn't want to get a machine hot that could create noisy neighbor problems.

10:20 Imagine if the misbehavior of one tenant could impact the reliability,

10:25 optimum availability of other customers.

10:28 But what I just described is the everyday database.

10:30 It's kind of nuts that there's so many people that kind of live in the shadows,

10:36 I think, because they put systems online because behavior they cannot predict.

10:40 They might become a wave of queries or users or whatnot.

10:44 They just ruin the day for everybody.

10:46 And people start citing slow query logs and all that junk.

10:49 And so I wanted a system that gave me an economical

10:54 thinking where the developer writes a query and they understand its cost.

10:59 We can project out its cost.

11:02 A lot of what I think has made Cosmos successful is

11:05 that it kind of makes sense in a token world, right?

11:08 Like when I use a coding agent,

11:11 I know that certain traces of reasoning costs more compute.

11:16 When I make a cross-partition query in Cosmos,

11:18 I get back an RU count of how much I spent on that query.

11:22 And so it allowed me to scale Vercel from a few developers that didn't

11:27 want to have a DBA on staff and to actually hundreds of engineers

11:33 that are supporting a growth in usage that we would have never imagined

11:39 with deployments which is one of the key collections that we host on Cosmos,

11:43 having grown our weekly deployment counts as three X'd in a few months.

11:48 And I think there's probably no database system

11:51 in the world that I could have homegrown,

11:54 for example, that could have absorbed that kind

11:56 of growth that agents have gotten us.

12:00 Thank you.

12:01 We're very grateful for you to be our customer.

12:04 We have many uh developers online right

12:07 now watching this that are early in career.

12:10 What would be your advice to them?

12:14 Well, one of the things is that the agent is like the new computer.

12:19 And this is advice that I give myself every day

12:22 is that I really have to update my priors every day.

12:26 I give a presentation to the whole company saying, look,

12:29 these are the things that I used to believe AI couldn't do.

12:32 And I was wrong.

12:34 I'm coming out and saying it like I thought AI couldn't,

12:38 you know, contribute to large code basis.

12:41 I was wrong.

12:42 I thought AI couldn't do good designs.

12:45 I was wrong.

12:45 And so it's very important to realize

12:48 that the things that you've learned, have to hold on.

12:53 Of course you have to appreciate the skills that you've developed and be proud,

12:57 but also you just be very light on your feet is my advice.

13:02 I like to think of the engineer of the future as someone

13:06 that is very full stack in their understanding of the business.

13:11 When I hire an engineer,

13:12 I don't expect them to just be typing code or even prompts for that matter.

13:17 I want to understand the full problem.

13:20 ah Again, it's a little bit like I mentioned that economy system that Cosmos

13:24 has given us where I was able to shift the burden of understanding,

13:29 you know, what our database performance and cost

13:32 is going to be to the everyday developer.

13:34 They didn't toss the problem out to, oh,

13:36 some other team will worry about scaling my database.

13:39 Right.

13:39 No, you are participating in that process.

13:41 I think the original ethos of DevOps is very much in that direction.

13:45 I think the difference now is that everyone is super full stack.

13:50 I do design, I generate videos, I generate SVGs,

13:54 I create front-end code bases, I create backend code bases, I write CLIs.

14:01 During the holiday break, I vibe coded my own Swift menu bar app.

14:06 And so I think the antidote here, like the way to really stay relevant is

14:13 to be super open-minded and uh super full stack.

14:18 Thank you, Guillermo.

14:19 Really grateful for us to have you here.

14:23 Appreciate everything what you're doing.

14:24 Love your posts.

14:26 Love you everything that Vercel companies.

14:28 Likewise.

14:29 It's been a great journey and excited to continue collaborating

14:32 with you all and uh do more with Vercel in the future.

14:45 Thank you, Guillermo.

14:46 Now, Guillermo just covered a few of the reasons why they chose Azure Cosmos DB.

14:51 But these are the same reasons why thousands of other

14:54 companies choose Cosmos as a database for their agentic apps.

14:58 Its schema-free flexibility enables your team to run fast, iterate fast.

15:03 Its search built in into your query engine so that you don't have to build,

15:08 deploy different systems for search, for database, et cetera.

15:12 Its reliability, low latency,

15:15 and serverless elasticity to enable your software to spike,

15:19 to enable your app to grow with demand without you touching anything.

15:24 And of course, it's five nines reliability that makes your sleep

15:28 sound while your software runs and serves millions and billions of people.

15:34 Now, when it comes to AI use cases, we have many, but the...

15:38 Common ones with Cosmos DB, of course, include vector database, right?

15:42 That's why we invested in vectors, multimodal vectors,

15:46 uh the speed and cost effectiveness of vector search.

15:50 Semantic retrieval, billing on top of vectors,

15:53 offers you ability to do full-text search, hybrid search,

15:57 precise, increased precision of your retrieval, uh billing,

16:01 taking advantage of the knowledge graph of your system.

16:05 But a recent trend is...

16:07 A new use case that tripled in size

16:10 over the last six months is agentic memories.

16:13 We have more and more systems choosing

16:15 Cosmos DB as agentic memory for their apps.

16:18 And there is a reason for it.

16:19 Memory is semi-structured.

16:22 Memory is dynamic.

16:24 Memory needs to be reconciled over time.

16:27 And all of these capabilities are built into Cosmos so you can focus

16:31 on what app needs to do and less about the infrastructure that it runs on.

16:38 Now, let's get a couple of words about the search in Cosmos DB.

16:43 The vector search is based on this kind

16:46 of algorithm built by Microsoft Research.

16:48 And the two key things about it is one, it's extremely cost efficient compared

16:53 to the traditional uh separate standalone search systems.

16:56 It's up to 50 times cheaper, as you can see on the charts.

17:01 It is also low latency and scales, no matter the scale of your app.

17:06 It provides you the same millisecond latency

17:09 for your vector search and full-text search queries.

17:13 And it scales well from few vectors to billions of vectors in your data set.

17:18 Now, when you think about an AI app built on Cosmos,

17:22 of course, OpenAI Charge GPT comes to mind.

17:24 That is well known.

17:26 It's a tremendous one-of-a-kind application.

17:31 It executes more than 1.4 trillion transactions daily against Cosmos DB.

17:35 It stores more than 45 petabytes in Cosmos by now.

17:39 And at this scale, it's the fastest growing app on the planet.

17:42 It still grows tenfold every year.

17:45 And to dive deeper into what enables OpenAI to scale

17:51 that fast and maintain the rapid pace of innovation, let me invite John Lee,

17:58 Staff Engineer from OpenAI to share a few thoughts with us.

18:03 Well, hi, John.

18:04 It's really great to have you here.

18:06 um Would you mind introducing yourself?

18:09 Yeah, of course.

18:10 It's good to be here.

18:12 My name is John.

18:12 I'm on the online storage team at OpenAI.

18:16 Awesome.

18:17 Well, OpenAI is a unique company in many ways, right?

18:20 But one of the things that everyone knows

18:23 about it is that, and it's really visible, is that it moves really fast.

18:27 You guys are doing many things concurrently and very fast.

18:32 What is OpenAI?

18:34 look for from databases to enable users' fast pace of innovation?

18:39 That's a great question.

18:41 I think there are lots of standard database

18:43 things that we care about uh at scale.

18:46 So predictability, being able to scale to per-byte sizes,

18:50 uh being very flexible with how our use cases can be built.

18:55 And then, of course, observability is very important.

18:57 But I think from the scale perspective and being able to move really fast,

19:03 uh The most important thing here is being

19:06 able to scale from zero to millions of QPS,

19:10 being able to scale from zero bytes to petabytes.

19:13 Features are launched on the regular,

19:15 and they go immediately from no usage to being used by hundreds

19:20 of millions of people on a daily basis kind of thing.

19:24 So the way that you might design a normal database

19:27 and then scale it out as your use case grows,

19:30 that just doesn't work at this kind of velocity.

19:34 And I think we have thousands of developers that are actively building products.

19:39 So with Codex, now things are iterating even faster.

19:42 It's really important to make it easy to onboard to databases really fast.

19:47 we have thousands of tables ah that ah we have to back by Cosmos DB.

19:54 And we have a system in front of Cosmos

19:57 that helps you be able to do schema-less design.

20:00 uh allow these product developers to onboard very quickly.

20:07 Amazing.

20:07 um Could you share any of the interesting patterns

20:10 that you guys have done on top of Cosmos DB?

20:14 Yeah, you know, think predictability is really key for how we use Cosmos DB.

20:19 ah We have a very well-defined API that we have and expose.

20:26 And all of these clients, they build on top of this, right?

20:30 And that allows us to keep the system stable.

20:33 em We don't want every client to be writing their own Cosmos DB queries,

20:37 creating their own Cosmos DB accounts, creating their own tables.

20:41 So we try to build a multi-tenant system on top of Cosmos.

20:44 uh Interestingly, at this global scale where we have

20:47 a lot of Cosmos accounts in a lot of regions, region failure is a big problem.

20:54 It doesn't happen often, but with dozens of regions, it will happen.

21:00 And so we definitely use Cosmos DB multi-region write.

21:04 We replicate our accounts.

21:06 so that we can always shift reads

21:08 and writes to other regions in case of outages.

21:10 ah But generally also for good performance,

21:13 you need to be able to allocate your data

21:16 in Cosmos DB accounts that are close by.

21:20 One particularly interesting feature that we use

21:22 for Cosmos DB ah is that we have tiers

21:24 of Cosmos DB accounts where some of these accounts

21:27 that are being used to store globally accessed metadata,

21:31 things like account information and user settings,

21:35 We actually replicate that to dozens of regions

21:38 as opposed to just a single region.

21:42 That's fascinating.

21:43 What is the...

21:44 You've done a lot of innovations on top of Cosmos DB.

21:47 What's your favorite?

21:52 I would say that uh being able to move data across Cosmos DB.

21:57 So we have a lot of products that will start

22:02 on a single Cosmos DB account and in a single container.

22:09 But as they grow or as their needs change,

22:12 iteration is very important at OpenAI.

22:15 As needs develop, maybe we realize

22:17 that they have much higher performance constraints.

22:21 And so being able to transparently migrate their data from one Cosmos

22:25 DB account to a different Cosmos DB account that is globally replicated,

22:29 is something that we've spent a lot of time trying to build around.

22:33 And the Cosmos DB team has uh given us a lot

22:36 of the building blocks that we need to build to do that.

22:40 That's awesome.

22:41 I wish at some point you joined the publishers,

22:43 because a lot of viewers would love to see this as well.

22:47 Well, we have many...

22:49 app developers that are early in career or potentially are still in college,

22:53 ah what would be your advice to them?

22:57 Yeah, I think this is evident also in the way

23:01 that our system has been built to accelerate product teams.

23:05 There's a big culture at OpenAI

23:07 about shipping things and shipping things quickly.

23:10 I think it's important to have a vision of what you're working towards.

23:14 But don't overbuild solutions to problems.

23:17 Try and get something working into the hands of your users.

23:20 And then you can evolve and iterate on it.

23:23 Almost.

23:24 Never you will get the first iteration correct,

23:26 but you'll learn from that experience.

23:30 Thank you so much, John.

23:31 And thank you so much for being our customer.

23:34 And thank you for joining us at this conference.

23:38 Thanks for inviting me.

23:43 Thank you, John.

23:44 Now, you noticed, John mentioned codex use in OpenAI.

23:49 Everyone uses coding agents, right?

23:51 As I said, based on telemetry,

23:52 more than half of our customers are now using coding agents.

23:57 And for you to get started, let's say you don't know about Cosmos DB yet, right?

24:04 The best thing is to install our agent kit.

24:07 This is a set of the skills.

24:09 that gets your agent up and running

24:11 and becoming the world-class expert in Cosmos DB.

24:14 It has 80 plus skills and growing and convincing

24:17 best practices with Cosmos DB across the entire app lifecycle,

24:21 from data modeling to partitioning, to queries,

24:24 indexing, throughput management, high availability, and monitoring.

24:29 Now, to see it in action, let me invite our fearless product leader,

24:34 Andrew Liu, to show AgentKit in a demo.

24:38 Hi, Kirill.

24:41 So to set the stage, ah I am building an app here.

24:45 It's a travel planner.

24:46 It's a multi-agent Cosmos DB application, and it's going to help me plan a trip.

24:52 So here I'm planning a five-day family

24:55 trip to LA that includes my 3-year-old daughter.

24:58 Now, what's important about this is that it's not just helping me plan a trip.

25:01 It remembers where uh I left off.

25:05 And this is done through by storing memory in Cosmos DB.

25:08 This includes short-term memory, long-term memory, and vector.

25:12 Now, here's the thing.

25:13 It's not about the application, but rather, ah I'm new to building AI apps.

25:17 And the thing I constantly question myself is, did I build it right?

25:20 Because at scale, getting it not right gets expensive.

25:26 What if I picked the wrong partition key?

25:28 What if I have the wrong data model, the wrong index?

25:30 I mean, that's burning RUs in production.

25:33 That's the kind of thing that you don't

25:35 want to surprise you when you launch and scale.

25:38 So, let's jump over into what this looks like in VS Code.

25:43 And first thing I want to show you is

25:45 actually our new VS Code extension or relatively new.

25:48 Many of you guys have seen the portal and we get it.

25:51 We've also gotten a lot of the feedback on, we

25:53 really want to go and improve the desktop tooling.

25:55 And so, we've been putting a lot of effort into the VS Code extension.

25:58 And what you'll find here is you'll get a lot of the same operations,

26:01 but you can do it now in line with your code without having to contact switch.

26:05 So, here I have my Cosmos DB account.

26:07 ah and it works just the same way.

26:10 I can go and query the database right behind this application,

26:13 and what you'll see is I'm storing a bunch of data in Cosmos DB.

26:17 I have my short-term memory in the form of messages,

26:20 I have my long-term memory in the form in this memories container

26:24 where I have things like my declarative memory which is uh durable facts,

26:28 procedural memory where I have behavioral preferences,

26:31 and episodic memory which is going to be

26:34 things that are a bit more trip specific.

26:37 Let's go into the magic on uh agent kits.

26:41 Did I even get this right?

26:43 So installing this is actually very easy.

26:46 Just one uh CLI command away,

26:48 you can use MPX skills add and go and add the Cosmos DB agent kit.

26:54 It's going to go through a few different uh questions like setting it

26:59 up uh for the project scope or doing it uh globally on your computer.

27:05 When it sets it up, what it's going to do is

27:08 it's going to go and create this on the project scope,

27:11 a .agents container or rather a directory

27:13 and it'll load all of my skills in here.

27:17 Now that I have my agent skills set up,

27:21 let's go and go check out the quick application.

27:26 I am going to go take a look here and I'm going

27:29 to have it just go review my Cosmos DB data model for performance issues.

27:35 Now, this isn't just a generic LLM looking at the code.

27:38 ah The agent skills is a package set of specialized expertise modules, right?

27:42 We're taking a lot of context about Cosmos DB

27:46 and bringing that into uh modules uh just for this.

27:51 Now, there's no services, no accounts, it's a repo of skills.

27:54 So, the agent gets smarter about Cosmos.

27:56 It's teaching the agent how Cosmos DB actually works.

27:59 It'll include data modeling,

28:01 best practices, partitioning, indexing, RU economics.

28:04 and make sure that that actually matches up with my data access patterns.

28:07 These are things that it took us decades

28:11 to learn over building real-world Cosmos DB applications.

28:14 Now you can get that directly in the form of agent skills.

28:21 Now, ah this is what's really cool about this.

28:23 It's not just doing some generic pattern

28:26 matching and giving me generic NoSQL advice.

28:28 It's reading my actual code.

28:31 It's taking a look at my code, infrastructure as code bicep templates,

28:35 correlating my partition key, my index policy,

28:38 my document shapes to the different data access patterns.

28:42 And it knows what quote unquote good looks like specifically for Cosmos TP.

28:47 This is expertise that used to live in one senior architect's head on my team.

28:53 But it was hard to get a hold of those individuals.

28:56 I had to book them a week out.

28:58 And now it's just in my editor here on demand.

29:03 So let's take a look at that summary here.

29:05 ah It was able to go and find an order of priority in highest ROI.

29:11 Oh, shoot, I missed some composite indexes for my order by query.

29:15 That can shave off a bunch of RUs for each of my different queries.

29:20 And when I'm running at scale with high QPS, that's really going to add up.

29:23 It's also going to go and uh prioritize additional things like,

29:27 hey, I should go and fix my memory's partition key.

29:30 Once again, hey, if.

29:32 I get the wrong partition key uh as this application scales to many partitions.

29:37 If I have a lot of cross partition queries, that's going to go add up.

29:41 So, what I'm really happy about this is,

29:44 it's catching this uh before I hit live application or live production.

29:49 So, there's two key takeaways I'd like to uh for everyone here in the audience.

29:55 Number one, if you're a new developer,

29:58 uh Cosmos DB doesn't have to feel foreign or scary

30:02 like all the cognitive load of partitioning, data modeling.

30:06 Right here, I can go and interact here

30:09 with my Cosmos DB agent kit and go and get

30:12 a live code review of my system and make

30:15 sure that what I'm doing is well architected by default.

30:18 Also, if you're an existing Cosmos DB uh developer, well,

30:22 this also lets you go and scan your code,

30:25 scan your past projects and go look for opportunities for.

30:28 optimizing the performance as well as cost characteristics of that application.

30:32 Now back to you, Kirill.

30:40 Wow, thank you, Andrew.

30:42 I think the last point resonated with me so much.

30:45 This is not just allows you to increase your productivity and build apps faster.

30:50 It's actually a world-class expert in Cosmos DB as your code reviewer.

30:54 You can plug it in into your CICD processes in your code review flows.

30:59 And it's right there catching any bugs that you may

31:03 introduce or helping you tune your code to be more efficient.

31:08 Now we have.

31:09 Great coding agents, skilled on Cosmos DB, we have fantastic AI apps.

31:14 But with all of that, we need systems to continue to be reliable,

31:19 secure, and performant.

31:20 And Cosmos DB is built for that.

31:23 That is the number one priority for Cosmos DB as a database.

31:26 Cosmos DB is a database that can

31:29 take your applications anytime from gigabytes to petabytes,

31:31 from hundreds of transactions per day to trillions of transactions per day.

31:35 You don't need to worry about it if you hit on overnight success at some point.

31:41 Cosmos DB is the only database in the cloud that offers

31:46 you guarantees not only five nines uptime across any consistency modes,

31:51 including the strong consistency.

31:53 It also provides you money back guarantees for zero data loss,

31:58 thanks to the consistency SLAs financially backed.

32:02 And finally, it's the only database that gives you full autonomous resiliency,

32:07 zero touch for your system.

32:10 This scale and the reliability is made

32:12 available to you not only just by software, but also by an awesome hardware.

32:17 And to discuss how we run the fleet of Cosmos DB,

32:21 I would like to invite Steve Berg,

32:23 Corporate Vice President and General Manager of the Service

32:26 CPU Cloud Business Group at AMD on stage.

32:30 Welcome, Steve.

32:31 Thanks for having me.

32:32 Good to be here.

32:33 Great to have you.

32:36 Now, Steve, from your perspective,

32:38 how does AMD and Microsoft partnership impact how customers experience Azure?

32:42 That's a great question.

32:44 Before I answer the partnership piece, let me give you a little bit

32:47 of history about our working together with Microsoft.

32:50 Back in 2017, we launched our first EPYC processor, Naples.

32:54 And Microsoft was actually the first cloud provider

32:56 to offer it to make it available in the cloud.

33:00 That created a really unique bond between us and Microsoft.

33:03 And over the past decade,

33:05 we've worked together to launch five more generations of EPYC instances

33:08 that are available in about 60 different compute families across the globe,

33:12 with our most recent one being Turin, which is an excellent product.

33:17 So we're constantly working to optimize our hardware, along with your software,

33:21 to be more performant, more efficient,

33:23 and provide new capabilities with each generation.

33:26 We do this with a deep

33:29 engineer-to-engineer collaboration on making roadmap changes.

33:31 that improve our products and help them

33:34 get tuned better for Microsoft Azure needs.

33:36 So some of these examples include things like confidential computing,

33:40 uh 3D vCache and high performance bandwidth that's used on some of your VMs.

33:47 This partnership has led to more Azure uh first parties adopting our hardware

33:52 and that is what the partnership provides is Cosmos DB and users

33:57 on Cosmos DB get all the benefits of decades of work that we

34:02 do together and all the hard work that happens within our collaboration.

34:06 I could agree more.

34:07 We actually use AMD quite heavily in our Cosmos DB fleet and that's what

34:12 enables us to scale with demands like

34:14 OpenAI and other super fast growing applications.

34:17 Yeah, exactly.

34:17 OpenAI and all those just get all the benefits

34:20 of us working together with everything under the hood.

34:22 So it's really, that's what the partnership really brings.

34:26 That's awesome.

34:27 Now in today's world with global situation, energy efficiency is top of mind.

34:32 How should builders think about balancing

34:35 the performance needs and sustainability and energy efficiency?

34:39 Yeah, we absolutely focus on that.

34:40 Our product teams and our engineers are

34:43 always looking at what's energy efficiency and sustainability.

34:47 And fortunately, builders who choose to run on Azure get all

34:49 of this benefit without even really having to think much about it at all,

34:53 because we do all of the work for them.

34:55 We work to deliver generation after generation EPYC CPUs with more performance.

35:00 better efficiency per watt, and improved TCO,

35:03 which directly translates into users and builders being able to get

35:08 benefits of cost and maximizing their performance per watt per dollar.

35:13 That's the critical metric.

35:15 So when you consider the global reach

35:17 of Azure with over 70 regions and counting,

35:20 it's absolutely imperative that we maximize performance

35:22 for those users in a sustainable way.

35:24 So focusing on that performance per watt per dollar is where we concentrate on.

35:30 And when Cosmos DB team rolls out new improvements to your SaaS offerings,

35:35 builders implicitly get all those benefits

35:37 just by using the modern infrastructure.

35:40 And when end users roll their own infrastructure as a service,

35:44 if they gravitate towards AMD EPYC VMs, they also receive those benefits.

35:49 That's amazing.

35:50 That's exactly what we all need to do, right?

35:52 We don't want to think about problems that we can't solve.

35:55 We want you to solve them for us.

35:57 We want to make it easy for you, yes.

35:59 Semiconductor has always fascinated me as a really

36:02 tough industry because you have to plan ahead,

36:05 five years ahead for what CPUs customers might need.

36:08 How do you do that?

36:10 Yeah, it's interesting you asked that because I don't think anyone could

36:14 have predicted what was happening with AI in the past year or so.

36:18 And it's really fascinating to see what you're doing with agentic AI today.

36:22 I think the rapid pace of AI took everyone, many by surprise and...

36:27 No one would have guessed the importance

36:29 that CPU provides the role in this market, right?

36:31 CPUs are becoming what I like to refer to as the new bacon.

36:34 uh Inference workloads, orchestration, pre and post processing, agentic AI,

36:40 general purpose all continue to use CPUs in a very greater detail.

36:47 so cloud infrastructure will continue to move from general

36:52 purpose commodity type machines

36:54 to increasingly customized and differentiated platforms.

36:57 And we're gonna focus on what's important, that dollar,

37:00 performance per dollar per watt that is so critical in our infrastructure.

37:04 So we already do a lot of this custom work today.

37:06 provided some of the, with Microsoft,

37:08 I presented some of those examples earlier.

37:10 It will continue to expand in this space in the coming years.

37:14 So if I leave you with a prediction

37:16 of what's gonna happen in the five year span,

37:19 I can tell you that we're gonna launch more and more VMs with Azure.

37:23 And Cosmos DB and user developers in the planet will

37:28 be able to get benefits from AMD and our collaboration together.

37:32 That's exciting.

37:33 So Cosmos DB customers can look forward to more transactions per dollar,

37:37 thanks to the work that we have.

37:39 At a better cost efficiency, absolutely.

37:42 Awesome.

37:42 Well, thank you so much, Steve.

37:43 Thank you for our partnership.

37:45 Thank you for sponsoring this conference

37:48 and made these wonderful sessions available.

37:50 Appreciate you having me and thank you all the time.

38:00 That's amazing.

38:01 ah Now, Cosmos DB is not the only NoSQL database we have on Azure, right?

38:07 Last year, we launched our second NoSQL database, Azure DocumentDB.

38:11 And unlike Cosmos, DocumentDB started from an open source project.

38:15 This is a DocumentDB project that Linux foundations at Microsoft,

38:20 Amazon, Google, Yugabyte, and 12 other partners are collaborating on.

38:26 And both Amazon and us are offering fully managed services,

38:30 Amazon DocumentDB and Azure DocumentDB on top of this open source project.

38:36 This is a fully managed document database that has

38:39 a number of advantages over other document databases.

38:42 It's faster, it has faster AI search, for example.

38:45 You can deploy it anywhere in any cloud or on premises.

38:50 And it's more than 40% cheaper than any

38:54 other document database on the cloud, including AWS.

38:58 MongoDB Atlas, for example.

39:01 Let's take a look at an example of this.

39:03 Let's say I want to send 1 million writes per second to my database.

39:09 On Azure DocumentDB, a single instance, single chart,

39:12 uh MAT allows me to do that, no sweat.

39:18 An equivalent scenario,

39:20 equivalent setup on AWS with MongoDB Atlas with the same skew, MAT.

39:27 would only give me 800K transactions per second according to AWS documentation.

39:32 We have the codes if you want to run it and see for yourself what it does.

39:36 But according to docs, it's only 800K and it cost me 60% more.

39:42 So all in all, I can get twice

39:46 as many transactions per dollar with Azure DocumentDB.

39:50 For every SKU, it's at least 40% cheaper, the SKU per SKU basis.

39:57 And overall, as you scale your application and load more data,

40:01 you get more than 85 reduction in TCO for your application.

40:04 So DocumentDB is really cost effective.

40:08 Take it for a spin and get the cost efficiency.

40:12 Now, when should I use which?

40:17 If you are having a lot of simple queries,

40:21 if you want to have a mission critical application,

40:25 If you want to scale limitlessly, if you like serverless form factor,

40:29 you like global distribution, Cosmos DB is for you.

40:32 This is the only database in the cloud that can do all that.

40:35 But if you prefer open source, if you like MongoDB API compatibility,

40:40 if you want to potentially port your application to other clouds or on-prem,

40:44 and you like the core architecture, then DocumentDB is a good choice for you.

40:53 multi-clouds, open source, complex queries, DocumentDB.

40:57 Lots of scale, serverless, 5.9 reliability, Cosmos DB.

41:04 Now, what if I want both?

41:05 What if I want 5.9 resilience and I want cross-cloud portability?

41:12 As of today, you can do it both.

41:16 We created an open source project.

41:18 We launched it today called MultiCloudDB.

41:21 It's fully open source MIT licensed and it

41:26 provides you cross cloud portability across Azure,

41:30 AWS and GCP by providing one data access layer on top of Cosmos DB,

41:38 AWS, DynamoDB and Google Spanner.

41:43 We provide you cross cloud disaster recovery.

41:47 We will provide you cross cloud replication with Sync.

41:50 as part of this project, you don't have to choose.

41:53 You can have 5.9th availability and cross-cloud portability

41:57 at the same time using this project and with Cosmos,

42:02 Dynamo, and Spanner from each of the hyperscalers.

42:07 Now, these are just some of the topics we will cover in the next few hours.

42:12 And while we have your attention, join us for skill challenge to win free Cosmos

42:17 DB certification and stay tuned for more exciting sessions today.

42:23 Install Cosmos DB Agent Kit, look at more sessions, build an app,

42:27 and maybe you can talk about it at our next Cosmos DB Conference 2027.

42:33 Enjoy the rest of the event.