Delivering Valuable AI Insights Requires Protecting AI Data Sources

Avaya syndrome. I run products for Haiku. I've been with the organization from the start of Haiku.

I think I've had the opportunity to talk with some of you in the past. So glad to again, connect the gap. Satya covered quite a lot about the overall, what does it mean to do a true AI infrastructure and how could we protect and are different deployment models customers do in ai.

I'll cover some of them and how we protect. At the end of the day, our focus is data because we think that's the core, which, uh, drives the whole operation, right? So that's what we are looking core when we talk about data.

I want, again, one of the things I want to add to what Satya was saying earlier, when we talk about data, it's just not the data alone, it's the data, the configuration of the different services which the customers actually have. And so when, when you think in terms of recovery, when you think in terms of data resilience, it's the entire set. Not just the, not just just as sources alone.

So wanna let you know before going specific into some of the ai. I just wanna give a quick recap about our platform itself. Brian did a very high level overview early on.

Now I'm gonna double click a little bit and then share with you. As all of you guys know, there is quite a lot of data sources. Customers actually have a big problem, which our customers have told us is half the time they don't know what they actually have.

Sounds a little crazy, but that's a reality, right? If many of you might remember, that's why in the traditional on-prem world, people used to have the application discovery, uh, mechanisms. This, especially with the adoption of cloud services and the adoption of SaaS, this has become a big problem for customers.

The first thing we help customers, we don't even charge for that by the way. We help customers discover their entire data estate, which is all their SaaS, pass, DBAs, ias, all of their services. We iteratively discover.

That's the first element to the customer. We discover, help them visualize auto categorize and tell them what's protected and not protected. That's the first step, especially when you think in terms of, uh, uh, ai, I mean, and the number of the breadth of data sources the customers have.

This helps them quite a bit. That's the first one. And this is not a one time thing, as you guys can imagine.

It's a continuous operation. We help them. That's the first part.

Obviously, once you do that, we are a company of focus on data protection. We help them protect the data. The key thing, I'll probably tell you, our focus is again, protect, protecting it extremely efficiently.

And that's the big thing we focus on, um, how we, the breadth of coverage, which we'll talk about, but it's also the efficiency we deliver in. The third part is around recovery. At the end of the day, the only reason customers are protecting is to be able to recover.

One of the beauty of Haiku workload platform is that we are able to recover the data granularly because the Keith was asking, Hey, what happens if there are some individual J files corrupted? Things like that. We can get you the entire data set.

We can get you even granularly the single full JS file, which you messed up. So we can do at different levels, we do that. This is just not at the file system level, even at the SaaS level, we can do the granularity which the customers, uh, desire.

The other one, which there was a lot of discussion on in the last one, was around the migration portability for customers. We don't think of it as a migration. We think of it as a data portability for customers, because customers want the freedom of choice and we are there to help them with that particular aspect.

That's what we do. We do all of this capability for a wide range of data sources. We have over 90 plus integrations today, right?

That is something wise. This actually important, especially if you are doing a two enterprise, you will actually have it. I'll show you, share with you some data points, which we actually have as we go further.

All of this, one of the constant requests we get from customers is that data resiliency when in terms of, um, a ransomware, things like that, especially even in the world of ai and some of the SaaS customers are right now extremely worried about, uh, protection against ransomware, malware, things like that. The good thing is we have Haiku R Shield. This is our essentially the way our cyber secu cyber, um, cyber recovery solution.

It is not an independent product. It is something we have built into the fabric of our HighCo R cloud platform, right? So our, our shield, which is a cyber resiliency solution, is built into the platform itself.

So it's not an independent solution. Just a little diversion here, but I wanted to you guys to have a notebook since all of you guys are not very familiar with Hico. I wanted to give you a little deeper into what we do.

So now let me get into, uh, specific in one of the ai, you guys are all experts here in the AI space. Um, such I talked about one of the things with building AI models and what customers are doing, things like that. One of the other more popular use cases you guys know in a lot of customers is creating a rag workflow, right?

Essentially, how do you train your existing LLMs, not train, enrich your existing LLMs with additional internal data sets? And this is pretty much you go into most large organizations, they're running an in-house, uh, in-house system, a rag, uh, system so that we can actually enrich it with their internal data sets. This is a very common one, but one of the core elements in this thing is the web database, right?

Because how do you actually enrich your information? It's the information stored in your Vector database. The thing we are very happy to say is that Haiku is the only one, which actually today is one of the first enterprise backup vendors to be able to deliver backup four vector databases.

When we talk about vector databases, we talk about things like Pine Cone and Redis. This is in addition to even some of the classic databases actually have, uh, vector information stored right now. We can actually predict them too.

But these guys specifically Pine Corner, as you know, was specifically focused on Vector database by this, even though it's an in-memory thing, has a significant, uh, vector, uh, database focus. So that's vector data focus, and that's one of the reason I think we started with those things. In addition to those two databases, a big part for a lot of our customers is that the additional data sources which actually play apart, right?

It's not just the public cloud data sets, it's a lot of the other SaaS data sets which also come into picture. Let's talk about it here. Uh, Can I interrupt you for one second?

Sure. There, on, on that previous slide. Uh, so this is Dave Graham again from Emel Commons.

So how do you start to handle this idea where lots of these storage companies or storage platforming companies are starting to embed their own vector derivatives into their databases? I'm thinking about the vast of the world, the Pures, the NetApp or whatever with inside X and all the things that they're ending up doing, because that starts to change your locus of data. It starts to change your locus of where these, you know, where this method is.

Where these embeddings are is something that very, very circumspect of data itself, right? And it's now embedded in a system that you may or may not have direct access to a IM to you name it. So how are you looking forward to that?

Or how are you trying to understand that, especially in a, I'm hosting my own cloud kinda world? Totally. So it's an excellent question there.

So the way we, Omar, one of the strengths of Haiku platform, I'll tell you is that even though we talked a lot about the cloud, we actually not just actually do public cloud and SaaS. We also do a lot of the on-prem infrastructure, talk about the large data file systems and object storage is running on-prem and the virtual machine application infrastructure, we protect that and that is one of the benefits of us. So it's when we talk about our overall solution, we can cover the breadth of data sources for our customers.

That's how we look at it. And there was a question early on to saying, Hey, what happens when the customer data is distributed across lots of different, uh, lots of different large, just one place, not just one cloud. It's across the spread.

The beauty of iku is that from a single pane of loss, the customer is able to manage the entire dataset. That is one of our unique strengths, and that's why we have had a literally very rapid growth in this space, uh, as because a customer is able to protect data spread across multiple different, uh, locations. And that's how we look at it.

That said, I mean, every storage vendor is starting to add it. As they start having new APIs, we are starting to integrate. I would not say that we are doing everything possible, but that is, that is a significant effort going on in in this space.

Yeah, Thank you. I have one more question on that. Just open Hendrix, Parker, CTO, six feet out, the solutions.

You mentioned you've talked a lot about odd native type aspects, but what about if someone's doing kind of roll your own or Kubernetes, they're doing Postgres and PG Vector, how does that play in with these tools? Okay, so for example, if customers want to run a, a Postgres on their own and run Kubernetes, for example, let's say they're using GKE and it could be on the cloud and using GKE, it could be on-prem tool, it doesn't matter. And if they're running Postgres along with GKE, uh, that is something we absolutely support today that is natively supported in the platform for customers who wanna do that, right?

We are using a services as an easy example to talk through, but we absolutely support customers running their homegrown stuff. So, so Ray Lucchese Silver Drink Consulting, what's, what's the integration between Iku and Pine Cone or Redis? I mean I, so post PG Vector and things of that nature, you know, you can do a lot of things from a standard database perspective to download the database and things of that nature.

You have something special with res with, with respect to Pine Cone and Redis integrate. And from an integration perspective, we are using the standard APIs and using their standard export methods and as they, as each of these solutions offer. And that's what that currently we are doing.

We have a lot of other integration discussions going on, but I would not venture to discuss the future, but as of right now, that's the way we do. Okay, Cool. So we talked about the rag.

I mean like even when it comes to data warehouses, right? People typically, typically think of it us as one data monolithic data warehouse, but as you guys are very familiar with it, there are tons of data sources which feed into this day, big data warehouses, the cloud. And the thing is, as Satya mentioned earlier, there's not just one thing.

It's multiple streams of data coming at t from different locations, different data source there. Some of the things Satya covered earlier, things on object stores and databases and you covered there. The thing is that we do have a lot of customers using pass and they are, uh, extremely critical data source because for example, BigQuery has over 80 different SaaS applications, uh, pumping data into it that becomes very critical for customers.

The question is that a lot of people think SaaS is actually fully protected by the SaaS vendor. As you guys know, they do protect the infrastructure, but the data is customer responsibility. You guys are very familiar with it and the SaaS data is very vulnerable.

When I talked about vulnerable people sometime, really, is that a problem for customers? I'll share with you one of our recent, we actually did a study, um, this is like done across like 500 different customers and this is part of the SaaS, uh, resilience report, which actually we was published, uh, done with independent by independent study here. The thing which pretty much in, if you look at this number of SaaS application, 90 is a portion of them have definitely increased their consumption of SaaS.

That is, it's not a question if it's the question of, uh, how fast is the growing And that is something they're in every customer, right? And when we talk about for 6% of them have significant increase because one of the beauty of SaaS, it's easy to use and it also means it's also easy to turn on and lots of customers end up u turning it on. And this is, uh, this is just IT tracked SaaS, I would prob probably say because there are other SaaS which is shadow, shadow it, shadow SaaS, that's still absolutely there going on.

And the customers are trying to get under, get them under control, and we are helping them with, that's with the SaaS adoption there. The second part, people, one is the adoption itself is growing. Second thing people ask us, is security incidents really a problem in SaaS?

This is real data coming from customers, not us saying something, right? 65% of the have said, uh, some data breach, 65% of them. I said, there are some data breach in the past year.

That's what they said. And uh, the number of average versus SaaS application data breach instance in the past year, just asking the customer, how many have you had? The average was two, right?

It's a significant impact. Customers that are having, when we talk about the number of sash and people say, is it really growing? Here's just a data point for you guys.

Last time when we ran 139 was the average. And right now this year it's 159. The number of average SA application is growing.

The rate of growth is absolutely there. The largest, the number of data SA SaaS store sources, the more risk the customer actually has. That is a problem.

And when things go down, what is the level of disruption will actually happen? And, uh, person have said, because 60 of the 68% that have experienced a data breach in the past year, 87% of them have some level of disruption, right? And one of the things customers told us is that on an average, not even a high average, they said it takes the minimum five days to recover and each day on an average costs a minimum $400,000.

This is just the cost of the recovery thing. It doesn't include all the people, uh, time lost and so on. That's something like that.

So that's the level of impact customers actually having. So what is that have people generally done here? Um, if you look at the backup vendors, we on, on an average people, the, or the entire, all the backup vendor, enterprise backup vendors, there are five is the number of, uh, as they've added in the last year.

That just is doesn't, because there are over 30,000 SaaS applications in the world that are over 35,000 exactly to be the right number, over 35,000 SaaS applications and around five being added that pace, you cannot do that. That's why Haiku r Cloud, we have built a platform that customer can add new data sources very, very quickly. And, uh, I, one thing I probably want to add here is that the way we leverage is we actually leverage AI to build our SaaS modules.

If people are interested in that particular one, we built the entire model so that we can accelerate the development. That's one of the reason, for example, there was a vendor who said, oh, we are gonna support Atlassian. So Atlassian as you know, is one company, but they have lot, I think some of you are, uh, focused on the DevOps side of the house.

I can tell you this, Atlassian is just, is company which actually has lots of different modules, JIRA, sir, JIRA Service Management, Trello, Bitbucket, and so many other things. They actually have the vendor who actually said they're gonna support Atlassian supported one SaaS, one Jira alone over the LA 12 month period, right? That is in Haiku.

If you look at it, this is the number of protected SaaS applications which we actually have today. And so the thing which we have actually added, right, the breadth that the speed at which we add is significant here. And this just to give you the power of our platform and why it's critical here.

If if you have any questions, always stop me. Sorry, I get excited and continue to talk. So the big part here is that especially with some of the things with people get, how does, how does the, some of the data warehouses get fed, right?

They have information coming from some of the SaaS applications. Let's say for example, they have tickets coming, industries coming from Jira. They have a service management from Jira service management, or they have data coming from box being fed into that, or they have, uh, some of the logs fed in from the different applications into the infrastructure into some other services.

The question is that how secure and compliant they are, we can actually keep the data very secure. Uh, we obviously protect as we, I we talked earlier, we protect against ransomware, the original source as well as your data warehouse. We can actually keep it safe against ransomware.

That is something we do. And we support a wide range of clouds, which the customers actually cloud for the data sources where the customers keep the data. That is something we do.

And this is this, um, a key part here is that it's done at, at scale, the big part for customers is that can you actually deliver it at scale? And what is the level of SLA can you actually provide? That is something we do.

It's the breadth of coverage. It's the question of can we actually give you freedom from where the data is stored and how fast can you actually recover? Dub big Go ahead, Guy Courier here.

Um, you true. So, uh, data protection in SaaS, uh, I didn't, it's kind of obvious now, but I didn't realize how crazy it must be. Um, it, but it sounds like the key to addressing the CR by crazy, I mean just you alluded to it earlier, 10 different ways, 20 different, 30 different ways to connect to access the data formats, like all this other kind of stuff.

Like it's, every SaaS vendor is gonna be a little bespoke. They have their own infrastructure anyway, blah, blah, blah. So if you're using even four or five major SaaS applications as an enterprise, good lord, the exponential complexity of it is boggles my mind in this way.

Um, but I feel like you, you sprinkled a little magic AI pixie dust on this when you said, uh, we use AI to, you know, form our connections, is that, uh, maybe that's an unfair way to state it, but is that more or less how you are addressing that craziness? That's, that's a great question. So AI actually helps us accelerate it, but a fundamental thing which we have to do here is to build a platform which allows us to, it's extendable modular platform, right?

There are two aspects to that particular, we have made the platform, Now you've used the magic pixie word platform, so explain what you mean by that. Totally. Yeah.

So the way Haiku RCloud is actually built the core data services meaning things around like the policy, the scheduling, the databases, how do we manage track the retention? All that stuff is done in the core platform itself. Given that, let's say tomorrow you have your own magic application, right?

And if you wanna protect it, the way we have done is that we have modularized, you can create your module. A customer can create the module, the partner can create a module, or Hiku can create a module at that's the level of flexibility we have actually provided. It's a very simple small set of code, which only focuses on the Apple SaaS application itself.

They don't have to worry anything about data protection, they just have to say what's the data structure of the SaaS application, what the APIs they have, and things like that. And then we will automatically be able to create the module for that, right? So we will automatically, because once you understand the data structure, the reason we need to know the data structure, then we build a backup solution.

We want to build a granular deep, uh, data protection, not necessarily just the dump and restore. That's easy, right? I mean, that's not good backup solution normal.

Okay, so that's awesome. So then next week the vendor, the SaaS vendor makes a change, and then next at the end of the year, they do a major upgrade. Each one of those can break that, that integration.

That's, that's, that's an excellent question there. So one of the things as part of the, one of, a lot of times you, you, you hit the nail on the head, a lot of people think is adding SAS is very easy. It's just a one-time operation.

It's not a one-time operation, right? It's a continuous integration with the new services which are coming up and keeping it up to date and making sure nothing breaks. The good thing is that, I'll tell you of working with so many of the 90 plus integrations in general, people have the, the fundamental principle we do is that we only use APIs to do integration.

We don't do any backdoor because once you steward backdoor is the problem. So only we do only go through the APIs. That's number one.

We do follow the, uh, vendors' best practices when we do that thing. That's secondly there, there are, even with that, there are some special cases when we do have issues with the, with the partner changing the API without recognizing the impact on the side impact there. Um, but in general, this is one of the big effort which ICO has taken as the continuous validation that nothing breaks, especially in SaaS because those are continuously updated and we are do that.

That is, um, that's one of the, one of the beauty of how we have done it and at the scale at which we have done is, um, is magical with the platform. Yeah. So do you have an internal SLA or, or something similar so that, you know, you can say, okay, you know, we, we, we detect the break because it's just inevitable we detect the break within this amount of time or, uh, within an hour or something.

And, uh, we pledge a repair within a day. And so if your snapshots are at this interval, you may have this gap, but, or, you know, whatever That that's, that's a great question. So the way we do is that, I wish it was every hour.

The good thing is that we, as part of our cloud operations continuously, we do actually monitor, um, any breakage in backups. Well, it's actually because typically the first thing you see is the sign that something backups start failing in a particular environment when things happen. So we do actually continuously track that.

That is one thing we do. Um, I wish the answer was like, within an hour, I'll fix it. No answer.

Oh yeah, I, that was my magic pixie dust. Go ahead. That is, that is not the case.

But the thing is that because we are continuously monitoring it and because we are using API, the number of times we have seen this, because we have done this for many years now, number of time we, this is like probably two or three in the entire last 12 plus months I would probably say. So the times it breaks is small. The key things would be is that when it breaks within, typically within a couple of days, we are able to update because it's completely transport to the customer except for the initial failures.

We will, as soon as we detect it, we fix it, update the platform, and then the next set of backups automatically catch on. That's the way we do. Okay.

I wanna be clear. I think all of that is really reasonable and I think you have to have this cascade of events. Well, maybe it's just two steps, which is number one, um, there being a break in the integration and then number two being an incident at the time of a break, that it would really be a, a case of, uh, of data loss.

And that's probably a really low probability occurrence. But that is as an application manager, uh, in, in an enterprise, the sort of thing that I would be thinking about is all the bad experiences I've had in the past with, you know, various solutions or they're homegrown or otherwise of actual data loss because there was an integration that was broken and no one noticed for a month. Totally.

And I think you hit nail had a lot of people, that's why they think I can write a script and it'll dump it every month and it's good enough. Like, nope, it's actually more complicated than that, right? You, you, you set it there.

And the other thing for us, I'll tell you also is that let's say typically the vendor doesn't break everything. They break one or two APIs or the behavior of any APIs typically what they do. So what, even when we do, we back up everything we can and if there are one or two things failure, we mark it and we continue on because that way we can come and catch up later.

That's how the smarts are built into the system. Okay, thank you. Yep.

Cool. That's a quick thing. Uh, two other things I'll mention and I probably will give time for questions.

One of the things which Haiku does very differently than many other people, uh, many other people is the data itself, right? When we talk about data, typically people, especially when you're running ai, people are very, as you all know, classic enterprises are right now paranoid and they won't have control of the data they do, which is good for the production data. They don't think in terms of backup.

We are a fully managed service. So people say, oh, so you own the data. What we tell customers is that you should always own the data, never give up the access to the data.

That is a big thing which Haiku does seek. Give the customer, make sure that the bag, even the backup data is always with the customer. Obviously you don't wanna put it along with a protection production data.

You want to keep it separate and make sure it's warm, locked, make sure it's really isolated from, uh, any access and things like that. That is something Haiku automatically does it for you. So that is something we actually do, but we make sure the data is always with the customer because we don't want the customer to ever take our risk.

I mean, they should be completely under their control and that's what we do. But that's, it's a di big difference compared to many others because a lot of other people take the storage from public cloud vendors and repackage it and sell it to customers on a premium. We don't want to do that.

We say customers, you can get it much more cheaper from directly from the cloud vendor, go at it, get it, and then we will help you keep the data safe in your own control. That is something we do. Yeah, this is different than most others.

That's why we explicitly call it out. Okay, so this is, again, when we talk about that, people say, where can you store the data? Where can you keep the backup data?

Uh, you guys heard Satya talk, uh, Satya talk about, uh, data domain as running on the public clouds. That is a choice. But in addition, we can support all of the hyperscalers, we can support other, uh, SaaS, other S3, uh, comparable vendors like Wasabi Cloud and OEH Cloud and so many other, uh, vendors we actually support.

And the good thing here is that in all of these vendors, we support not just one tier of scalers. We support all of their tiers of storage for the customer. That is one of the big benefits here.

And we can support all of these different SaaS being able to protect it onto these, uh, storage there. Yeah, just coming to a conclusion here, and we probably will summarize and leave it open, uh, with the floor, AI is hot and lots of customers are think the good thing we are seeing is that as data production is not the first thing many customers think when they go use their AI projects, but as they're going into production, as some of these large data warehouse are going into production and the cloud customers are realizing that they have to protect them. That's what, that's the trigger right now, if you guys probably, I'm sure many of data don't talk about it because this has not been top of mind for customers, but we are starting to see that as customers going production, they're seeing a need to do that, especially in a compliance in this data, being able to get to a good data model, uh, data model, data, data library, being able to get the copy of the model, what it was, along with the data sets along with it, being able to accrue to, uh, compliance for their regulator needs.

That is something there and we gotta keep it safe. And that is there Haiku is there to help them. Many of you guys asked and you get to a particular checkpoint.

Absolutely, we work with it. Everything we do in Haiku, haiku is through the rest. API, this is why it sounds a little probably crazy for some people.

We, we backup up all of their data repositories too, like GitHub, GitLab, Bitbucket, um, their orchestration tools. Like, um, for example, if you end up using like a Terraform, uh, cloud provider, Terraform providers, or if you end up doing things like Circle CI for some of your DevOps management, we protect all of those infrastructure too. Not just the core data.

People think of object stores, think that it's all the things around it, right? Or could be even things like cloud functions. We protect all of them for a customer so that the entire data set can be brought back, not just the data and then you wonder what the hell to do, sorry to say there, but the question is there.

So it just, the entire stack, we can help the customer protect. That's, that's why keeping the data there. So we talked about data, uh, Lakehouse and Vector databases, that is, we are happy to say we are one of the unique vendor data protection vendors at the enterprise class, being able to protect across my different public clouds, being able to do this thing.

That's a quick rundown. Um, any questions? Uh, Ryan, Satia, Andy, I'm here.

And, uh, any questions we can answer to you guys? Hi, it's Ivan Mcfe from Google and I have one question. Yeah, you've spoken about your customers, but can you give us some indication as to some of your top customers, the size of them, the scale they actually, um, for, for for Haiku?

Sure. That's a, that's a excellent thing. Uh, point there.

We have over 4,600 customers as a company. Our primary focus is, the way we would tell is that from a volume perspective, it's our mid-market and low enterprise. That's our primary, uh, focus area as a company.

But that said, we have some of the largest, the unique thing, interesting thing about Haiku is that even though we are a young company, we have a significant of a number of, uh, federal agencies. All of the US Armed forces uses not just US armed forces. There are a bunch of other countries we can actually, uh, name they use Haiku.

And from, if you think in terms of large, uh, organizations protecting us, I'll throw out some things there. Uh, customers, for example, the large pharmaceutical industry guys who are in the large sip making industry chip or SNGI would say it, uh, technology, uh, in the technology space. Customers with tens of thousands of employees in that particular one.

We have, for example, one of our customers have 200 different sites and they actually protect using haiku. And one of our customers in the cloud actually has data, multiple petabytes and petabytes of data in spread across five different regions backed up through Haiku. Um, those are some of the samples I can tell you that's, that's dimension.

We also have, uh, mid-market customers with, uh, 500, uh, mailbox users to 2000 mailbox users, things like that. So with this quite spectrum, we actually have, and for probably mentioned one other thing and probably stop, sorry. Uh, for example, in some of the SaaS, we have customers running like 30,000 mailboxes and things like that.

So it's a, it's a quite a big spectrum. We actually have Yeah, I I was gonna just add, uh, that I think the customers that care about sovereignty, uh, and coverage, right? It's not just about being able to back up all of these workloads, but I don't want to give my copy of the data to you, I wanna hold it to myself.

We're one of the only SaaS solution that offers customers, you know, full ownership of their data sets, that the control plane is delivered as a SaaS, but the data plane always sits within the four walls of their virtual boundaries. Uh, right? I mean the, and, and the data always stays in.

Customer tenant never leaves. We don't exfiltrate that. And I think the customers that value that are, are typically the ones that gonna jump into this bandwagon and, and, and run with it.

Why I think we have such a strong federal sense as well is that we keep the data, you know, right? Where that, uh, the customer can control and, and and, and see and manage. So Malcolm, you had a question Malcolm looked like you raised.

Yeah, yeah. Hi there. Uh, Malcolm Butan, MNB Networks, um, just a, uh, quite a straightforward question.

Is the service available globally, everywhere, or is there any restrictions in any regions? That's the, that's a, uh, great question Malcolm. So our service is truly available in every part of the world.

It runs on any, anywhere your hyperscale is run, let's say, put it that way, anywhere your Google Azure database can run, we actually support it. There is no restriction on that. Uh, I'll give you guys a real example.

One of our customer partners was using four regions and they wanted, the new region was coming up in, uh, for one of the hyperscalers in Mumbai, and they said, Hey guys, there's a new region coming up and how long will it take for you to support it? I said, the day, uh, when it's available, you can use it. Well, they said, what do you mean?

Because we obviously use their APIs and things like that, which all the cloud vendors, our system is built to truly scale like the cloud. And that's how it has enough intelligence built in to make sure the data movement within the cloud too is done extremely efficiently. And that is, uh, that's a big part for the customer to keep the cost there cost, uh, cost in mind because as you guys know, if you end up ever do, um, within the, even within Hyperscalers, you guys know this within hyperscalers, even if you move the data between regions, there is an egress cost, right?

As a backup. You don't wanna run one region, say, I wanna backup everything. You're like, Nope, you'll end up paying tons of egress cost, you'll just kill the budget.

Don't do that. So that's where the Haiku intelligence is built in so that you can run in the right region, in the right place for the.