Thank you, [ Judith ]. Good afternoon, everybody.

Thank you for joining our investor call. I'm -- so we're going to today talk about the figures, the unaudited figures for the first half of 2026.

And as usual, I will walk you through the high-level numbers. some strategic updates and my colleague, JD, our CFO, will provide more insight to the details, financial details.

And then at the end, we're going to do a question-and-answer session. So we're looking forward to your questions and feedback.

So as I said, we're going to start with a business update, current trading update, an outlook and summary and then questions and answers. So this is our standard disclaimer.

So let me start with the update for what's happened in the first half of 2026 for Exasol . So overall, here are the key figures.

Let me speak to this, and I'll provide some, like I said, high-level overview, and we will dive into more details later in the presentation. So we have -- the year-over-year recurring revenue is slightly up, driven by substantial churn reduction.

So we actually have a significant less churn in the first half of 2026 than we had last year. So the churn rate decreased to about 10%.

Last year, we had 24% -- so -- and we expect that ratio to continue also for the remainder of the year. Our revenue and EBITDA year-over-year comparison is affected by nonrecurring items in FY '25.

So last year, for those of you who have been with us for a longer time, we had a large onetime hardware deal with one of our largest customers in the financial services space. So that brought in EUR 2.9 million of ARR with roughly a margin of EUR 900,000.

So that effect we didn't see this year. So therefore, we have a revenue impact of the aforementioned EUR 2.9 million and also an EBITDA impact because we didn't get the margin from this onetime deal.

So that basically explains the difference between the -- on the revenue side. So you see here on the top left revenue, EUR 19 million versus the EUR 21.5 million we had in 2025.

It's worthwhile saying that we also had some smaller one time revenue as well in June and July this year, but not at the size of what we had last year. So the total ARR is close to where it was last year.

So at the end of 2025, we had EUR 38.1 million. Now we had EUR 37.9 million.

In our outlook, you will see that we are projecting ARR growth to be in the range of minus 2% to plus 2%. So that's basically less than what we had guided previously.

And we -- therefore, we had provided an updated ARR outlook in our press release last week, and we will share some more updates on where we think this is going to play out. So on the new business performance, we were able to secure 11 new customers.

I'll share a little bit more detail on the next slide on those, leveraging our key messaging around data sovereignty and Agentic AI. So one of our key challenges is that we've been seeing existing customers have been postponing expected and planned expansion investments in their data infrastructure, which had a negative effect on ARR.

So we actually had 2 larger deals in the plan for the first half of 2026 that did not materialize. And therefore, we basically short of the deals that we had initially planned.

We also had some positive upside. So we closed one deal with a financial services customer, and we were able to avoid a larger churn.

So that compensated to a certain degree. But I think the key element here on growth, we were not able to bring in these 2 larger deals.

One particular, I think, has to do with our expansion strategy for the Lakehouse Turbo product that did not materialize. That was roughly [ EUR 800,000 ] to EUR 1 million.

And then there was one other customer who basically didn't expand the system because of cost factor, both on Exasol license costs, but also, as pointed out here, increasing hardware costs in the market. So the positive effect from new customer acquisition and churn reduction didn't offset the weak investment behaviors among our existing customers.

That said, we are -- when it comes to the EBITDA, net liquid funds and net income, we're on track. So as I explained, EBITDA is lower than last year, but this is just because since we didn't have this onetime revenue deal.

Net liquid funds increased to EUR 21.6 million and our income for the first half was at EUR 1.2 million. So here, you have a deeper view on the ARR development, and we've done the comparison between the first half of this year versus the first half of last year.

So you basically see the starting point this year was EUR 38.4 million. We ended the first half at EUR 37.9 million, which is a decline of 1% overall.

We were able to secure 11 new customers with a total of, I believe, roughly EUR 350,000 ARR, where 8 were in our focus verticals. We had upsell of EUR 1.1 million with 86% coming from focus verticals, and we had offsetting downsell and churn in total of EUR 2 million, which basically led to this more or less flat development from the beginning of the year to the end of the quarter.

Maybe one comment on what happened last year. I think we had some very positive upsell with existing customers.

So we -- at the June 2025, we had a very large upsell with one of our financial services customers, but also we had contrasting downsell and churn, which in total at the first half 2025 was EUR 8 million. So you see here, there's some good news and some bad news on this slide.

We were indeed able to reduce the downsell and churn substantially. But on the flip side, we have not been able to drive the significant upsell that we have been last year or in the first half of last year.

So that's, I think, as I explained, has certain reasons. Ultimately, here's the explanation.

So again, the delta between the bars on the left and on the right has gone in the right direction. We will talk about also later on how we predict the business to go for the remainder of 2026.

So on the next slide, let's talk about the strategy update. This is a slide that I already showed in our last earnings call.

As you can see here, we have been going through a significant clarification of our overall positioning and strategy. So we do have with our core products, 4 key use cases that customers are using us and that we're promoting.

So the classical use cases is data warehousing, so where we are being run as the central data warehouse at our customer. We have analytics acceleration.

So that's a use case where we accelerate an existing third-party data warehouse like an Oracle, for example, or Teradata at customer site. Now we've identified 2 additional use cases.

Let's go through this, and I'll have some more details on the following slide. Sovereign AI -- so sovereign, meaning that customers run AI and data analytics combined in a controlled and governed environment.

So that's a key positioning for us. Specifically for the German market, it means we are a German company.

We can run on local cloud infrastructure like the STACKIT, for example, but we can also continue to be strong on running on-premise. So customers can and are using us to run not only data analytics, but also more and more AI workloads on-premise.

So that's a key element of our strategy. If you go to our website, a new updated website, you will see that we have changed the top-level positioning, and we have also added key use cases on our website that basically underpins or illustrates what I'm describing here on this slide.

So Lakehouse Turbo, we have been talking about this in the past 1.5 years. Lakehouse Turbo basically means that customers have a modern lakehouse in place, such as a Databricks or Snowflake and that we would run Exasol on top of these data Lakehouses.

So we've been working on that as a separate product. We had some milestones that we gave us end of Q1, where we wanted to achieve certain adoption business numbers.

We haven't not achieved that for a stand-alone product. However, it's an integral part of our overall product functionality.

So as you see here on the deck, we're integrating the Lakehouse Turbo core capabilities into the different use case-driven approaches. So we have several customers that we're actually working with who have the need and the desire to run Exasol on top of existing data Lakehouses.

That's what we're doing. But the decision that we make is we no longer market a stand-alone product outside the existing Exasol value proposition chain.

So let's dive deeper into customer benefits for Agentic AI installments. As I said, we are moving and centering our messaging around that.

There are 2 key market opportunities for Exasol that we are actually very excited about. First of all, data sovereignty increases the demand for European database infrastructure solutions.

So we've seen and heard this due to the overall geopolitical situation. There's a lot of concern about using U.S.

hyperscalers and running U.S. software, database software on those respective hyperscalers.

So we are -- actually, I cannot disclose the names here, but we are engaged now in 3 POCs with companies that have explicitly looked at us together with regional partners and building solutions that are sovereign by nature. So that is a key element of our strategy, and we will continue to drive that forward.

The second aspect is a more horizontal aspect. Agentic AI results in exploding data requests, analytics and number of data requests.

What does that mean? Today, traditionally, in a BI or a stack, people would generate BI dashboards.

So you would have a centralized BI team usually that creates dashboards for the different departments and all the data gets consolidated underneath these dashboards. So that's one way.

The other way to access data was to use programmatic interfaces so that it can be fed into other systems. For the past 6 to 9 months, we've been seeing the world changing around us.

So more and more agent agentic use cases are entering the enterprise of today. We're seeing this with all our customers.

We're seeing this internally in Exasol. And what that means is agents are going to access data and they want to access heterogeneous data sources in enterprises.

So it's no longer just I build a Tableau dashboard on my data warehouse. It's actually I build a set of agents and agents have different skills and different tasks that are being designed by different departments that basically do data analytics combined with AI use cases.

So what that means is that more and more data will be needed that is stored locally in environments that the user can control. And I think this is where Exasol comes into play, where AI agents will take over and automate a wide range of processes and analyzes within organizations.

Now think about it, if you have large databases in your enterprise with millions of rows with very large data sources, your AI agent that works on these data structures needs to store data somewhere. They need to put it somewhere so that they can do fast, efficient and cost-effective analysis about it.

And it's not just one agent that does this. Enterprises see a plethora of agents spawning up left and right.

And that's the opportunity for us. We're making it extremely easy for companies to develop and deploy those agents.

And of course, we have our engine, our Exasol engine where data gets being stored. So that's what we are working on.

So this is from a road map perspective, things we are developing. At the same time, and I'm going to share this in a second, we already have customers where we're working on with those customers on some of these key Agentic use cases.

So back to the overall position. So Exasol, we are positioned with our new positioning as the sovereign Agentic database and it addresses the challenges posed by AI agents with its unmatched speed, massive compute power and excellent price performance.

So you see the same arguments that we've been using in the past, speed, compute power and price performance absolutely also apply to the new Agentic world. They even become more important than they were in the, let's say, traditional world where BI dashboards were being generated.

So the 3 key pillars of this positioning is sovereign by design. So we let the user, the customer control where data runs, how data gets accessed and how governance is enforced.

It is built for AI agents. So our system is designed to handle many AI [indiscernible] consecutive in parallel concurrent AI agents, applications and users working on.

That's inherent to the Exasol core engine, and we are extremely well positioned to handle multiple hundreds, thousands of concurrent queries that are being run on our core engine. This is the heart and soul of what our product does.

And then over the last 6 to 9 months, we also work on making it easier for customers to get access to our product, the faster to deployment path. So we've launched a free product, Exasol Personal.

The initial launch was in December last year for deployment in AWS. But the big breakthrough for us now actually was in June when we launched the Exasol Personal local, which is geared towards developers who use the product on their own computer.

So you don't need a cloud environment. You can basically run agents on your own machine.

And only when you go into production, when you go into deploying it in your enterprise, you would then -- we would upsell those customers from the local version to the enterprise version. So we're seeing traction there.

So this traction is measured in active users. So now we have about 100 active users using the local version.

This is not revenue yet, just to be very clear. We have a target for this initiative that by the end of the year, we want to substantially grow this by another factor of 10.

But that is basically seeding our product into the market and making it as easy as possible for developers to build agentic data solutions. And I can tell you, actually, I literally did it last night with -- together with our Chief Product Officer.

I used ChatGPT, and I prompted it that it automatically downloads the free version that we have on our website, on our developer site and created an agentic system for a certain use case where data is being generated and then retrieved. So it is easy for any developer to do that.

So we overcome the burden of installing a complex system and also signing up or putting your credit card information into a complex system. We strongly believe that this is going to be a major game changer for Exasol.

And it addresses one of the key issues that we ultimately have in the market, which is the lack of awareness and visibility in the global market. So with the faster production pillar here and the underlying Exasol product, the personal version, we're driving awareness in the overall market.

So let's talk about a few use cases. I think it's important, of course, that we get traction, as I just described.

But ultimately, we need this to also translate into revenue and for our customer. And I think what we can share here is there's a large U.S.

pharmaceutical life science company that we're actively working on that are basically have a super compelling use case, which is an Agentic AI environment where they gather all unstructured data, clinical trial data and put it in a structured database, which is [ Exasol ]. And then they can use agents to run analytics on this database.

So that's one of the areas that over the past months, we've been actively been involved with working with these customers to see a few other names that you're probably familiar with because they've been existing customers like the T-Mobile, Piedmont Healthcare and Helsana. But we put them on this slide because these existing customers are evolving on their existing use cases and basically adding Agentic use cases to what they have already been doing with Exasol in the past.

So we will -- I know this is a lot, and it's -- the purpose of this call is to provide obviously an overview on where we stand and where the first half of the year ended. But I think it's important for investors to hear where we're going.

where we see opportunities, also where we see challenges for the business. But I just wanted to share my personal excitement on actually where we're moving and what we're seeing in the market.

And we are going to provide deeper insights, both from us, from leadership team as well as some selected customers in our Capital Market Day on, I believe, October 15, JD, correct? And yes, it's there.

And where we specifically do a deeper dive into our strategy. We bring our Chief Product Officer, for example, our Chief Commercial Officer, who will also be able to explain both the technology, how we're doing it, but also how do we implement and monetize this on a going-forward basis.

So with that, I will hand over to JD, and I'm happy to take your questions at the end of this call.

Thanks a lot, Joerg. As usual, let me give you a little bit more granularity on some of the ARR numbers, but also then on our P&L numbers and liquidity and comment on the updated guidance that we provided last week.

So starting with kind of the quarter-by-quarter development. As Joerg already mentioned, we have a largely flat development since the beginning of the year.

with a mild negative decline in Q1, which I commented in last quarter, which was driven mostly by the fact that the churn that happens, happens typically front-loaded in Q1, whereas Q2 was largely flat with a slightly positive development in our focus verticals, which grew by EUR 300,000 net, whereas in total, we only grew by EUR 100,000. What you can see in here, and I'll comment on that a little bit further is that this diminished impact of upsell dynamics primarily hit our focus verticals where we have been historically very strong in driving upsell, and that has had a decidedly lower momentum this -- so far this year and particularly for customers who have on-prem installations where an extension of our license with them often goes along with investments in infrastructure.

Those are impacted right now by the general higher prices in the hardware market and reduced investment behavior. And as a consequence, we see many customers focusing in the use cases that are there.

And existing to kind of limit the data growth because a lot of our growth momentum in the past was about just increase of data sizes because our license model works on total amount of data under management. And in the existing use cases that we have with customers, the large ones, customers are focusing on trying to diminish that in order to avoid additional hardware investments.

The AI topic is a little bit separate of that because it often comes from different investment pots. It's also smaller tickets, and that also partly explains when I talk about the ARR bridge later on, that obviously, the higher number of new customers that we're seeing come along with smaller tickets, smaller initial invests.

They cannot compensate for the large upsell effects that we typically had and that we haven't been seeing in the first half of this year. So as a consequence, we came out at EUR 37.9 million.

If we look at that on a full year perspective compared to Q2 last year, it's also a largely flat development. If you remember how we started the year, we ended last year with a kind of negative ARR dynamic of minus 8%, which improved to minus 3.5% last quarter on a year-on-year basis.

We're now at minus 0.5%. So the kind of the growth momentum that we had last year is diminishing, and that's mostly due to the fact that churn that we've seen, as you saw from Joerg's slide earlier in the presentation, the churn dynamic has been significantly lower.

And a more significant improvement in growth dynamic was kind of hampered by the fact that the upsell dynamic has gone down. So overall -- sorry, wrong button, you can see that gross upsell rate has declined, and this is where the upsell -- reduced upsell dynamic comes in from [ 113% ] to [ 106% ].

At the same time, ARR churn has gone down from 24% beginning of year, 20% last quarter, now 10% so that the net ARR retention rate has somewhat improved, but it's not on the level that we've seen in past years, and it's still below 100%. Looking at the churn rate, and I've commented on that already, we've seen a continuous decline from the peak level that we've reached in Q2 last year.

So what you see here is depicted the 12-month churn rate on a rolling basis. depicted on quarter-by-quarter level.

And you can see that we've been in an increasing churn dynamic almost 2.5 years up until Q2 last year. And since then, have been reducing that now back to the level where we were roughly end of 2023.

And that is the level that we expect to continue for the rest of the year where kind of further decline is mostly hampered by the fact, and we alluded to that in our news flow last week as well that the collaboration that we started with MariaDB last year is ramping up slower as expected. So, so far, the advance payment that we received has obviously been recognized as revenue, but we haven't seen business ramp-up because the technical implementation of the joint product has taken much longer than we thought.

and it's only really starting to be marketed by MariaDB now since middle of the year. There are some POCs running, but there is not a sufficient amount of commercialization yet.

And as a consequence, that will reduce ARR a little bit in Q3. And that's why for the full year, we will not see a further decline of the 10% churn, but like that's the level where we expect to end up towards the end of the year as well.

So these trends combined that I've outlined also mean that there hasn't been a further shift in our portfolio from the nonfocus to focus verticals. If you compare the development since end of last year or beginning of this year, yes, focus verticals have slightly inclined since then -- increased since then and non-focus verticals have declined, but that has -- it wasn't a major shift.

And that's also where you see the impact of the reduced upsell dynamic because the focus verticals per se have been largely stagnant so far. What does it all mean in terms of P&L development?

Joerg already mentioned some high-level figures to you in the beginning. On the revenue side, despite the ARR development, generally, recurring revenue has been largely flat compared to last year, a mild increase, but that's mostly due to the significantly reduced churn.

What you can see, obviously, that total revenue is significantly lower because we had a very high one-off revenue in the first half of last year. That doesn't translate fully into EBITDA because one-off business has -- and remember, those are hardware appliance sales, which are kind of bundled businesses together with license sales, they come at significantly lower margin than the software product.

So this revenue decline doesn't only translate to a fractional -- with a fractional impact into EBITDA. So on gross margin level, you overall compared to last year saw a decline of EUR 600,000.

Other operating income was slightly up mostly due to a slightly improved FX rate of U.S. dollar versus beginning of the year in terms of liquidity impact and our U.S.

dollar component in liquidity and a little bit of a special effect from dissolutions of provisions for long-term bonuses for our colleague, Mathias Golombek, who departed as a Board member, where fractions of his LTI are no longer [ avoided ] with his exit, which influenced other operating income to a small degree. As you can see, in terms of total cost base, we are largely -- we are basically exactly on the same level as last year with marginal shift in structure.

Personnel expenses were slightly lower than last year, which is in part due to a changed structure in our workforce. I think we've talked a lot about our developer site in India that we've ramped up, which allows us to invest in development capacity at a lower unit cost, and that kind of influences the personnel expense picture as well.

This has been offset by slightly higher IT infrastructure costs, mostly for internal AI usage, but also for internal replacement investment in server capacity. So the price increase that I've talked about that have affected some of our customers' investment behavior.

Obviously, we are as a company that also offers an on-prem product, we operate a fleet of on-prem servers ourselves to develop the product, and those have been impacted by price increases as well. But overall, we were able to keep the total cost base constant.

And as a consequence, you can see that EBITDA was lower compared to the same time last year, but only by EUR 400,000. So the lower revenue impact doesn't fully translate to down there.

Net income was also slightly lower than same time last year to the tune of EUR 200,000. As a consequence, liquidity also compared to same quarter last year was basically on the same level.

And yes, it's EUR 600,000 lower than same quarter last year, but liquidity as a point-in-time metric is heavily affected by timing of certain payments. Overall, this is very much in line with what we expected given the EBITDA that we're seeing.

And you will see on a full year basis, ultimately, our change in liquidity being very close to the total full year EBITDA once you discount for seasonality and certain timing effects. So overall, that still puts us in a healthy liquidity position to kind of get through this phase with lower upsell dynamic.

And I'll comment in a moment a little bit on what our perspective moving forward for that is. So if we go to the outlook and summary, as you've seen from our news flow last week, we adjusted our guidance, and that is mostly due to the fact that while we expect the dynamic for new logo ARR to continue to develop favorably, for example, we are -- and my apologies for the slight flipping, that must be something related to my [ mouse ] here.

We expect that dynamic to continue. In fact, there is one deal that we've been working on in the first half of the year, which is very close to being signed, which we had hoped it would still close in H1 with a large pharmaceutical company in the U.S., which is also an AI use case that builds on what Joerg said.

So we expect that to be included in the Q3 numbers then. And we expect that positive development of new logos to continue.

However, in terms of magnitude of ticket sizes, that will not compensate the fact that we expect the more moderate dynamic in upsell to continue throughout the rest of the year. And we would only expect kind of a recovery of that potential next year.

And as a consequence, we adjusted our outlook to a corridor of plus/minus 2% with some additional churn happening in the second half of the year, which we expect to compensate with the new logo momentum that we're seeing. On the revenue side, that means that we also slightly adjusted our guidance from mid-single-digit percentage decline to upper single-digit percentage decline -- as you probably -- as you've seen in the half year numbers, revenue was on a downward dynamic anyway because of the decline in one-off business.

And that's why we guided mid-single-digit percentage decline in the first place with the lower ARR dynamic, that doesn't fully translate into revenue this year, but it also puts us rather on the upper single-digit percentage decline. On the EBITDA side, we then also had to narrow our guidance to the lower end of our existing guidance.

We are still highly confident to achieve an EBITDA above EUR 3 million, but now in the corridor between EUR 3 million and EUR 3.5 million with taking account the impact of the lower revenue that I've just talked about. So as Joerg already pointed out, one of the important learnings from the first half of this year and the mild pivot in terms of which additional growth fields we focus on in terms of -- on top of our traditional core business of data warehouse and accelerator and kind of sovereignty-oriented verticals with this new focus on helping customers implement Agentic AI-driven use cases and adapting to those demands.

we feel we already announced that we're going to do a Capital Market Day mid-October because I think it's important to take more time to go in depth about what our perspective is on how the landscape of analytics infrastructure in companies changes with the shift to AI and how that also changes the role that a software like ours can play in tech stacks of companies that are now facing more and more AI-driven analytics versus classical dashboard and [ typical ] user-driven analytics and the new attributes that are required. And that's why we sent out an invitation for a Capital Markets Day on October 15, where we are going to go in more depth in the use cases that we now today alluded to, what we're doing for customers already, what we're seeing in terms of activity and pipeline development on the Exasol personal product or a version of the product that Joerg talked about and also what that means for our perspective on addressable market and growth perspectives for next year and beyond.

So you're cordially invited to join us there. It's a hybrid setting.

So you have the option to either be there physically or to be dialing in to give you a maximum possibility of attending. Should you -- if you haven't had received an invitation yet from our kind of distribution list, feel free to get in touch with us, and we will make sure that you receive an opportunity to sign up.

Beyond that, we're going to be present in the investment conference of Baader in Munich and then also in the equity forum in Frankfurt in November, where I'm looking forward to meet as many of you in person. So in summary, summing up what we said today, I would say, first half with mixed results for us.

I think we are seeing good response on the conversations we have with customers on both the sovereignty topic and Agentic AI use cases. The churn rates declined substantially as expected.

but that was not able to offset the significant impact that we've seen from the lower upsell dynamic, particularly with our focus vertical customers. And that's why overall, the first half of the year was flat or slightly declining.

New logo pipeline generation, as Joerg pointed out, we're happy with, although it doesn't translate yet into big ARR numbers, but the dynamic that we're seeing and the conversations that we're having and particularly also the use case that we saw implemented in the U.S. with the pharmaceutical customer gives us confidence on the AI angle.

And we hope to push that forward in the second half of the year based on that new positioning. Liquidity remains very good and puts us in a good position to get through this flatter growth phase and give us strategic and operational flexibility.

So that concludes what we want to share with you today. Looking forward to your questions.

And in case of no questions, very much looking forward to seeing you on the equity forum or the Baader conference on our Capital Markets Day.

My first question is on your partnerships. Are there still technical integration works ongoing with MariaDB -- or are these problems solved?

And maybe you could also comment on the development from the partnerships with adesso and STACKIT in the pipeline.

Yes, I can take that. Yes, with MariaDB, let's start there.

As we said, it's going slower than we would have hoped it goes, but it goes. That's the good news.

So they actually have 3 customers now in the U.S. that they are going to the POC of the product, and they expect to get that closed this quarter or next quarter.

So that's basically, I would say, forward progress. They're continuing to be committed to working with us.

And remember, they gave us also an upfront payment. So it's progressing, but it's progressing slower than we initially hoped.

That's why we basically not taking the [indiscernible] while we have to show it as churn at this point in time, but it's not something that disappeared. So it's again, slower and 3 customers will be -- are in the process of being onboarded and to generate revenue then for MariaDB and consecutively also for us.

And we are in talks with some other larger customers with that as well. So I would rate this.

I mean, so we were actually very excited last year. I think the excitement has calmed down, but it's, I think, moving in a slower but steady pace right now.

That's how I would describe it. Second, partnership with adesso, yes we are actually evolving.

So we've attended their Digital Day in Düsseldorf in June. We got some leads from there.

And we're particularly working with their financial -- the banking team right now, both in Germany, leveraging on our sovereignty message and in other countries as well. So specifically Turkey, adesso Turkey is going really well.

We just closed a deal, smaller deal, but I think somewhere in the 40,000 to 50,000 range. But I think it's a good first step.

And we have a good handful of other deals with adesso Turkey in the pipeline right now. The team is very committed for what it's worth.

I'm also going to go to Turkey in early October. So it strengthens the partnership with them.

So I think we're actually on a good path with adesso. And then last but not least, you asked about STACKIT.

Yes, we're actually working on 2 opportunities right now together with STACKIT in the German public sector.

And one other question is on the deal you're working with the U.S. pharma company, which will maybe close in the third quarter.

Could you give us any light on the deal size?

Several hundred thousand dollars to start with. And in addition to that, there's also a substantial 6-digit onetime revenue we get for providing services.

So we're not just selling the product to them. We're also providing consulting basically to implement the use case.

That can become a larger account next year. So there are more use cases for them.

So it's a step-by-step evolution. I think we are on a good track right now.

We have actually fully negotiated the contract, you might imagine with large U.S. companies, that's a bit of a longer game, but we've reached agreement on all terms and conditions with them.

And right now, we're actually bullish that this can become the beginning of a longer partnership that both gives us onetime revenues [ through ] consulting, but also a recurring product revenue. I think the other thing is once we get it to a point that we can talk about it in public, it also gives us a tailwind from a from a market perception, specifically centered around the Agentic AI use cases that we're building.

With regards to the cost of infrastructure, you mentioned that some clients are not willing to pay that, let's say. What's your feeling with regards to the midterm perspective on that?

I personally do not see that there will be any better situation in the midterm with regards to the costs, but maybe the clients realize that they can't change this. What's your view on that?

Yes. I think I would comment on that, Joerg.

I agree. I don't -- I personally don't see the price situation in the hardware market change anytime soon.

On the other hand, what we're currently seeing this kind of attempts -- the levers that are being pulled to postpone those investments, you can't continue those forever either because a lot of times, it comes down to data cleanup in the system, removing some old data. But once you've done that, it's not something that you can pull off again next year.

So ultimately, the investment dynamic will come back and ultimately, companies will have to invest in infrastructure because from a use case perspective, the data amounts are still growing. So the base dynamic of both compute and storage are still the same as before.

If anything, they are going to be reinforced by AI. That's the reason why we're seeing this investment boom in infrastructure and why the prices are going up.

And as you can see, there's also a little bit of a -- I wouldn't call it schizophrenic budgeting logic, but a little bit of a dual approach to how companies think about implementing their AI use cases where they do take some investments, whereas how the existing kind of more classical use cases are run and being managed. But -- so when this will come back, I think that's something that we are trying to evaluate for ourselves as well right now.

But the investment dynamic cannot be postponed forever.

I would like to start with the topic of churn and downsell. Maybe I missed it, but did you show the split between focus and nonfocus verticals in terms of churn and downsell?

No, it wasn't on the charts.

Can you give us a feeling of this?

So did you have other questions as well because then I would briefly look it up and maybe Joerg can look at that can answer your other questions in the meantime.

Yes. Okay.

The second question is also to downsell/churn. Can you explain a little bit the reasons behind the downsell and churn in the first half?

Were these similar to prior years? Or has there changed anything?

So some color on that would be helpful.

Yes. Maybe I can -- I think it's a mix of things.

So there's not -- I mean, when there is a downsell, it usually means that the customer is reducing the usage. And this can have multiple reasons.

But in general, I mean, some companies are just trying to compress their spending. And since our pricing is centered usually done on volume-based pricing, customers could reduce certain use cases or delete some data in the system to save cost.

And because not all data that has been stored in a database ultimately is equal to what the customer needs. So sometimes customers do that to optimize there.

So that's one reason for downsell. The other reason is that sometimes customers have different use cases and then they decide that certain use case either gets end of life or maybe move to a different system.

On the churn side, it's usually more dramatic where customers do consolidation. So one of the things that we've been seeing with some customers is consolidation to bigger platforms.

like a Databricks, for example, where we have been used for certain use cases. But then I mean, it was one of our -- we talked about this in the past with one of our large customer, [ adidas ], for example, they've been consolidating the whole enterprise basically into a single enterprise solution, which in their respective case, Databricks, which means that use cases that have been run on Exasol in the past are being transferred.

That typically then results in churn. And this happens with some larger enterprise customers, but let's say, also some other customers.

So I would say those are the key reasons driver of churn.

Okay. Should I continue with the third question or you would answer the first question.

So in terms of the churn of the first half of the year, that was actually 50-50 between focus industries and nonfocus industries. So the total -- let me go back to the slides to illustrate it.

If you look at the roughly EUR 2 million churn that we've seen in the first half of the year, roughly, it was half and half. The reason was that we had some churn of focus industry customers in the U.S.

for whom the kind of sovereignty argument isn't as strong as for the European customers. So there was some focus.

Just to add that there was one customer in the U.S., which was close to [ EUR 200,000 ]. They simply went out of business.

That was another reason for churn. I mean just to say sometimes companies just -- it was a start-up we had in the financial services space.

We had them as a customer for 2 or 3 years, and they basically shut down their business and therefore, we lost the customer. Third question, maybe third and last question, Spang.

Yes. The third question is again on MariaDB, just to get that right, did you say that you have booked a certain number in the ARR, but now in Q3, you have to cancel some of this again?

Or was it a misunderstanding?

No, that's correct. Yes.

So we received the prepayment from them for the first 12 months, which we showed as ARR that basically, we expected that to kind of ramp up in terms of business volume, in terms of underlying deals from them with royalties up until end of August. As we've pointed out, that has not been the case.

So we will have to true that up end of Q3 to whatever they have achieved up until then in terms of actual run rate. And that influences the churn that we will show in Q3 and then obviously also influences the adjustment of our guidance.

And until end of June, that was how much of the ARR?

That was [ EUR 750,000 ].

And any expectation how much you have to cancel out of this?

That depends on how the POCs that Joerg has talked about will evolve, but I don't think that those are going to be more in total than maybe up to EUR 100,000. So the value that's implicitly included in our projection because what we're facing essentially is roughly a 1-year delay almost or 3 quarters of a year delay in the ramp-up.

So for the purpose of coming up with a new projection and guidance, we factored most of it out.

Yes. Thank you for [indiscernible].

Thanks, everybody, for joining. And as you said, I'm looking forward to continue our conversation.

And I wish everybody a good rest of the summer. Thank you.

Same from my side. Hope to see many of you on the Capital Markets Day.