Nebius Owns 25% Of This $15B AI Company

Added: 2026-05-31

[00:00:00]Hi everyone and welcome back to another episode on the Indicot Invest channel.

[00:00:04]As promised from yesterday, I wanted to go ahead and do a massive deep dive on Click House today. So the goal today is to give you guys a little bit of background of Clickhouse, how they came to be, um obviously a little bit more about their financials, also give you guys a sandbox environment, show you what it looks like, how it works, and a very very very dumb down aspect for myself. Um because I am not an extreme engineer. I understand technology a decent bit, I guess. Um, but I wanted just to show you guys a little bit of a sandbox, get you guys thinking about how it works and kind of show you guys just a little bit around of, you know, what you can do with it and what it means.

[00:00:41]Now, as I like to do, we'll go ahead and look at the stock just real quick. Um, here's the public account, but we go ahead and look at Nebius real quick.

[00:00:48]It's currently at 226 um 70. It did hit an all-time high, so congrats to Nebia shareholders. It did hit an alltime high, I think of 2430 something. Let's take a look. Uh, well, it shows 233. It hasn't updated, I think, at 24039 or something like that.

[00:01:04]Um, but yeah, congrats to us Libya shareholders, you know, super super awesome. Um, but yeah, let's go ahead and hop right into it.

[00:01:16]All right, so quick context if you're new around here. Um, Click House is a database. And look, I know I know databases, they're not really not aren't exciting at all. Um, but stick with me.

[00:01:26]I'll read through it for you and give you guys an idea. Um, right now every ad company generates, you know, an, you know, an absurd amount of telemetry data. So, every agent decision, every tool call, every token, um, every valuation, every prompt change, they all create rows of data, right? Enthropic does it, OpenAI, every AI startup with like actual customers and traffic does it. Um, and they all need to query the data really really fast like subsecond fast you know within milliseconds if not under a second across billions of row rows that landed in the last 60 seconds right so they need to query really quickly and grab the data snowflake wasn't really built for that when they first originally built it big query wasn't really built for that Postgress definitely wasn't built for that back in the day u but click house was right and so as of today the thesis you know obviously it's now shown up in the numbers in my opinion um 250 million of AR growing 3x 4,000 paying customers and they're out of 15 15 billion. Um now obviously for us Nebius investors, you know, we only care about this for two reasons. One, Nebius, they own a 25% of the company. Um and so if you own Nebius, you own a piece of ClickFouse and that piece to me is just going to be increasingly more valuable um as we go on. two is to me and I feel like again I've said this a million times but like this is one of those infrastructure stories that defined the next few years of AI buildup right um and I think most retail investors and honestly even more so the analysts sleep on this subsidiary um so let's go and hop into it so where did this thing come from so quick history honestly like a pretty insane story when I was doing more research on it for you guys so let's start the year 2009 this is when this Russia searched in Yandex. We know Yandex is the the wide neb spun off. Um they have a problem, right? Their analytics product, Yandex, I think Meritra or Mer Merrica, I don't know how to pronounce it. Um is processing over like 20 billion events per day. This is in 2009 by the way, which is insane. Um they need to give their website owners, you know, real-time dashboards of analytics. Um none of these existing databases they work with can keep up.

[00:03:38]So, one of our engineers, a guy named Alexi Milovidov, Milividov, I think I'm terrible at Russian, sorry. Um, builds a new one from scratch. He calls it Click House because it processes clicks. It's literally what the names come from comes from. That's literally where this company came from. Um, for seven years, Click House runs internally on Yandex powering their own analytics. In 2016, Yandex open sources it. the benchmark the benchmarks that in database community sees um so the benchmarks the database community sees are honestly so good that people think they're lying like this is really powerful at the time um the dev community is like holy crap this is insane Cloudflare picks it up in 2018 and scales it to six million rows per second of log ingestion eBay rebuilds their entire observability stack on it um and the thing kind of just spreads like wildfire right that was sort of their early days so again an engineer at Yandex built this company and they spun it out. Crazy, right? Um, now fast forward to September 2021.

[00:04:40]Click House spins out. This is before the entire Russia Ukraine war actually, which is pretty lucky timing on their side to be honest. Um, ClickHouse spins out as an independent US company headquartered in the Bay Area. They raise 50 million from, you know, index ventures and benchmark. One month later, they raise another $250 million at a $2 billion valuation um from Altimter and I think uh I forget the other one. Um Aaron Katz, who used to run sales at Elastic becomes a CEO. Then in 2024, Yandex gets split into Russian half and Western half as we already know. I've kind of gone over that in another episode if you guys are interested. Um the western half which they called the index NV gets renamed to Nebius Group and as a part of that split Nebus retains the index stake in ClickHouse and that's the Nebia stake as we know it now. Um then in May of 2025 they have a $350 million series C at 6.3 billion. In Jan 2026 they have a $400 million series D at 15 billion. Um, and so again, that's sort of the back history of ClickHouse, how they started and their funding rounds. So now I'm going to show you guys a little bit of how this thing works, right? So I want to show you guys like just how fast this product is. Um, because I think it's pretty impressive and I get a little nerded out by it. So traditional databases like Postgress, MySQL, um, I think even NoSQL store your data rowby row. um every row of user's table sits next to the next row on the disk. So every row of your user's table sits, you know, next to the next row on the disk, right? So that's fine if you want to look up look up like one user.

[00:06:19]Um but it's abysmal and annoying to say, you know, what's the average revenue across 50 billion transactions? And so Click House stores your data column by column. All the time, you know, all the time stamps live together, right? all their user ids live together, all their revenue numbers live together, and when you ask an analytical question, you only read the columns you actually need. Um, and again, they've been told it's, you know, a 50 to 100x reduction than a row store. So, column storing versus row storing. Then on top of that, Click House uses something called vectorized execution. So, instead of processing one value at a time, it processes batches using um SIM CPU instructions. Now, guys, I know I'm getting really technical. I am not some crazy technical engineer. I'm reading things I've learned and kind of spitting out to you guys. Um, one CPU instruction can do 8 to 16 operations in parallel. And again, that's another 10 to 30x speed up. And then on top of that and everything else, click houses compresses your data 12 to 20 times. Snowflake compresses maybe four to five times. Um, and less data on the disk means less data to read, which means faster queries and lower storage costs. So again, just a benefit all around. Um, now I actually want to show you guys how this works. And so I'll go ahead and pull up um their playground is what it's called. Okay. So again, I am on Click House's playground. This is the free public ClickHouse instance. You can anyone can pull this up and kind of go with it if they want and play with on their own. It's literally just called play.clickhouse.com.

[00:07:56]Um, they have a few real data sets they load in for people to play with to understand it. The one I'm going to use is the UK property sales. So, it's literally every single house sale recorded in the United Kingdom going back to 1995. About 30 million transactions, real data from the UK's government land of registry. Um, and again, this is the kind of data set that would absolutely crush a really traditional database. And so, we'll go ahead. I'll just talk, you know, kind of talk while this one runs. The first query is just a, you know, the simplest possible thing. Just count how many rows. Oh, I guess it doesn't want to do that real quick. What the uh interesting. Hold on.

[00:08:37]This should work.

[00:08:40]There we go. I don't know why I did that. I might have had an extra space at the end. Um so again, this just counts the amount how many rows are in this table. And again, as you can see here, literally in 0 I mean 0. Zero seconds who like insanely fast. um literally milliseconds again, right? And so this is kind of, you know, not impressive yet. I just want to kind of show you guys what this does. Then I'll go ahead and I'm going to scan across the 20 million rows of data we have. And then I group them by year, calculate the average sales price for every year, and count how many sales happened. So it's doing math across the entire data set.

[00:09:21]Again, it's going to be something with spacing. I have a feeling. Uh, see if I can fix this.

[00:09:29]There we go. Sorry for the spacing again. And so you can see here, three decades of UK um, housing market history computed in literally 02 seconds. Um, just absolutely insane. And you can see prices average sales around 60 to 70 back in 1995, 1976. You can see over time how it increases. 2022 being the the highest. And you can see here the little graphic rates too. You can see the movement over time. Um, and also the amount of sales here on the right side.

[00:10:00]So you can see here 2006 and 7 for obvious reason massive sales. You also can see again here in 2021 massive sales and it's slowly gone down since. Now obviously 2025 not as many sales but I don't know if it's like exactly updated to the exact day. Um, but you can kind of see the battles the 2000 housing bubble 2008 dip and the postcoid surge. Um, and again, this just took you, you know, just did it in under a second because it's column oriented.

[00:10:29]You know, only had to read three columns, date, price, and row identifier instead of scanning every column and row at, you know, at a million times per second or whatever it may be. Um, now the last one I'm filtering out just the London sales. And again, I'll kind of it's probably going to be a spacing issue here, so I'll go ahead and fix it before he even barks at me real quick.

[00:10:48]So we can see here grouping by district of within London.

[00:10:53]So computing the average price per district and then ranking them from you know expensive to least expensive. As you would expect the city of London is the most expensive at I think this is 2.1 million if I could do my math right.

[00:11:05]Um 2 million 1.2 million 1.1 million.

[00:11:08]You can see it all the way down. Um you can see Waysworth has the most sales of all time. And you can see here again 0.02 02 seconds um for 28 million rows of data. Um and again, this is on like a free public sandbox, so this probably isn't even optimized with, you know, an actual infrastructure you would need.

[00:11:28]And again, you know, no pre-mped view, no special index, no warm-up or anything going on there. Um, now imagine, you know, instead of it being 28 million rows of UK houses, it's 28 billion rows of agent telemetry from Anthropic or every Cloudflare HTTP request from the last week or every transaction of Capital One in this section of this city. That's what ClickHouse does, right? So that's why companies use this platform.

[00:11:57]And so that is what's happening here.

[00:11:59]And I I hopefully you guys can see this pretty well. Um, but again, pretty nuts just how it works and pretty cool to play with. So, if you guys are interested, go ahead and go to playclickhouse.com and you can play with it a little bit. Super fun. Um, now I want to go ahead and also show you guys just another fun one just just kind of out. If you guys are interested, feel free to stick by. And this is doing as a percentile of sale price for every year since 2015. So if we can see here um the median sale price, the top 95% sale price and the top 99% sale price. Um just a super fun nerdy one to do as well. But let's get back to investment thesis. I don't want to stick here and nerd you guys out and bore you guys for too long. Um, but I just want to give you guys a little idea of what it looks like, how the sandbox works, and then how you can just how you can correlate that to how businesses would be using this in their day-to-day.

[00:12:56]Um, now on the financials, because that's what we actually care about as investors, right? Revenue was 250 million ARR as today's Techrunch piece up 3x year. Um, their last reported number was, I think, 88 million for full year 2025. Um, customer count at 4,000 paying customers. Sorry, the error was 160 million 4,000 paying customers. Um, which again they've 4xed that over a course of like 11 months. I think last June it was a,000 and it's just growing like crazy valuation of 15 billion. Um, but I want to give you guys a little bit understanding of who they compete with, right? Because I think we hear names like Snowflake, Data Bricks. Um, but I want to give you guys more of an overall picture of what that looks like. So, the overall like kind of blunt framing is they don't really compete head-to-head with Snowflake. Snowflake is built on like governed enterprise business intelligence. Um, what that means is like the CFO's dashboards, quarterly board reports. Um, it's slower, more expensive, and not built for like real time. Click house owns the real- time analytical segment. So, as you guys saw, sub-second queries on streaming data. I mean a sub point second queries. Um now again that's for millions of records. If you're doing billions it's way different category though in a different category exploding because of AI right that said the the benchmarks to me are still a little fun in comparison. So clickous's own published benchmarks show two to 3x faster aggregation queries than snowflake 12 to 20x compression versus snowflakes 4 to 5x and on a recent click bench benchmark at 100 billion rows.

[00:14:40]Click House Cloud was somewhere between hundreds and over thousands of times cheaper per query than the other major warehouses. Um now obviously those are I would say biased numbers from Click House since it's their own benchmark but even discounted significantly the gap to me is still pretty real. Um now the closest competitor used to be Apache Druid or Druid um and Apache Reno or Pino um they're also real time OLAP databases.

[00:15:06]So, but Click House has kind of run away with the mind share and the product velocity there. Duck DB is the other one I was researching that seems like it's kind of an interesting wild card. It's like a single node embedded OLAP database that's taken over um the analytics community. It seems like they're really loving it. Um and Click House kind of responded with creating CHDB. I don't know what that is. I'm sorry. I'm just kind of giving you guys the rundown. Um which is their I guess embedded Click House is all I really know. and the two products are kind of converging in opposite directions. Now, data bricks is sort of a different category. That's like a lakehouse. Um, but the real- time push to, you know, in 2025, they're starting to overlap with um click house. Now, now netting all of this, I think click house has a defensible mode right now in the real time like analytic workload space. Um, the category is growing like insanely fast and I think it's only going to the TAM's only going to grow even faster.

[00:15:58]Um, and they're definitely the technical leader. um they have the open source flywheel, they have the cloud monetization, and I think it's, you know, it's going to be pretty tough for them to beat. Now, again, bring this all back to why it matters. Who cares? Why should we care? Nebas owns a 25% stake.

[00:16:14]I probably said this probably 50,000 times over the past 3 months. Now, that stake is valued at like 3.9 billion. Um now I do think when they IPO in 2027 2028 maybe um their stake is going to be worth over 10 billion um when it's all said and done which is going to be awesome. Now if we take a look at you know the public AI infrastructure companies that traded a higher multiples than the last private round snowflake trades at and you know you can kind of look at yourself click IPOs at 25 to 30 billion and I think that's pretty realistic for a 250 millionaire or company. um if they IPOed right now and that's with 250 million of AR. If they get to a billion of AR or 2 billion of AR, I think they IPO around 50 billion and even a 20% stake is $10 billion. Um now I do think the IPO is pretty pretty soon. Um the CEO said they're trying eye in the next few years and the series D which is the most recent one in January had Dragon leading. Dragon Air is like a crossover fund that invests one to two rounds before IPO. So I think we can assume there's going to be at least one maybe two more rounds before they go for the IPO. Um they also had a T-RO price in WCM investment management both public market investors. Um the Netflix pedigrade president hire they acquired spree. They're kind of running the preipo playbook in my opinion. Um then you also have the strategic synergy. So Nebius obviously as we know sells the II compute. Click house is the observability layer for AI workloads. So they sit adjacent to layers in the I infrastruct. And so even before the IPO this stake has strategic value beyond just the mark of the stake, right? Um and here's really the kicker to me. You know the market as far as I can tell is pricing Neblas pretty much essentially just on the AI cloud business um and the meta and Microsoft contracts. To me, the Click House stake just feels like a footnote kind of in in the business and how they value it. Um, but right now I think it's conservatively in my opinion for projecting around 8 to 10% of Nebis' market cap. Um, and I think if the IPO it becomes 12 to 15 to maybe even something insane like 20% of their market cap, which is insane and something people need to think about.

[00:18:32]Um, and to me it's just kind of one of the cleanest examples of like the sum of parts pricing I found in the eye infrastructure. Um, but yeah, so how should you think about this as an investor, right? Um, click house itself, it's obviously private, so you can't buy it directly. Secondary market shares are essentially almost, you know, inaccessible to retail. Um, and now that makes Nebius the cleanest owner of Click House. So even if you just didn't even give a crap about Nebius or wanted some play in the AI infrastructure, I think having that, you know, that benefit of investing in Nebus get you getting that part of Click House is your probably your best bet since they on the highest amount. Um, but yeah, hopefully that was valuable for you guys. Hopefully that helps you understand Clickos a little more. I figured the back cursor is pretty interesting and also just playing with the sandbox a little bit would give give you guys an idea of like how that might actually work in the space um and how this is going to really propel itself in the agentic world where agents are doing all the queries and all the transactions and thinking and querying.

[00:19:32]Um but yeah, hope you guys found that valuable. If you guys did enjoy, you know, hit the like, hit the subscribe and let me know in the comments what you guys want me to cover next. I hope you guys like this. So, if you guys find some value in it, please share it with anyone that might found value in as well. Um, yeah, and I'll catch you guys in the next one.