We're gonna get started. I'm Laura Martin. I'm the Senior Analyst at Needham & Company. Let me introduce our speaker today. Rajeev Goel is CEO of PubMatic. He's the co-founder and CEO, where he spent nearly two decades scaling the company into a leading independent SSP, which has recently pivoted to become an agnostic OS, AI-powered operating system. A serial entrepreneur and tech visionary, he previously co-founded Chipshot.com. Rajeev also serves as a trustee for Johns Hopkins University and a board member for the Interactive Advertising Bureau. Okay, I want to start with a question on leadership. Even the last two sessions, both men have started with the saying that generative AI is transformational. My question is: How do you think about a leader in this kind of period of uncertainty and tech disruption?
Yeah. Well, I think, you know, we've been working in an industry, leading in an industry that's obviously incredibly complex. You know, sometimes people liken it to wartime, and I feel like I've been a wartime CEO for 20 years. You know, we compete with Google. Which is a declared monopolist, right?
Fair.
Okay, you gotta compete with somebody like that. You know, we view our opportunity as roughly a 4% market share leader versus, you know, Google at 60%. There's a lot of upside for us. I think with AI, I mean, I view this as really the third major transformation in our industry, and we're right at the early stages of it. The first was real-time bidding, the second was the consumer shift to mobile, and that obviously, you know, created a lot of disruption, winners and losers.
Now, AI. We're in, I think, the second branch of AI. The first was machine learning, right?
Yeah.
Which works on structured data. That came into our industry about 15 years ago. This transformation with generative AI, I think is gonna be significantly larger. The reason is that, you know, structured data's only been around since really the 1950s or 1960s, right? Think about, you know, data in a database. Unstructured data is the world around us.
Yeah.
It's 100x the size of structured data, right?
Yeah. More.
More, right? It's audio, it's video, it's the spoken word, right? It's the human interface to how the world exists. It's just a much bigger opportunity. We really view it not so much as just a technological revolution, which of course it is, but really an opportunity to revolutionize the value chain in our industry. To really bring the publisher and advertiser much closer together, remove a lot of the complexity in the ecosystem, which for us means an opportunity to drive performance advertising, increase our TAM, and create a lot more value.
You're talking about the company. I wanna know about leadership.
Sure.
What does this say? How do you have to lead differently in a period of, let's say it's only 18 months, of technological disruption, 1 your employee base, the fear factor? How does that change? Are you saying it's just the same as when you went through mobile? Is that the point you're making an internet?
I think those transformations required a significant level of leadership. Specifically, I think number one is honesty. Honesty with yourself, with your team, with your capabilities, where you have holes, where you have gaps, where you have opportunities. The second is to really question every assumption, right? It's really important to not think about what are the things that we've been good at historically, we're gonna continue to focus on, but really, what is it that we can do for our customers? What's the value that we can generate for them? Really hone in on that and then move the business in that direction.
Such a fascinating answer because it's so systemic. It's not about people. You must not see leadership as people. You must think it as strategic from a systems point of view.
I don't agree with that because I said honesty at the beginning, right?
Honesty, you can't tell me you're dishonest. If there's not a transition between mobile. I assume you're honest all the time. That's not a change.
Well, I think leaders aren't always honest with their team, right? Sometimes there are unknowns. You have to say, "Hey, we don't know. We gotta figure it out.
Yeah.
Sometimes you gotta change the skills of the team. Sometimes you gotta bring in new people. I think all of those things require a high degree of honesty in a period of significant change.
Yeah. More so, maybe. I think I would say most CEOs think they're honest all the time.
Well, that's a different issue. Whether they are or not, they may think that they are.
I think that's totally fair. Okay. Let's start. I missed you at POSSIBLE.
Yes.
I came by to your booth, but you were so busy. Everybody was so crazy at POSSIBLE. You guys made a lot of product announcements at POSSIBLE. Let's go over some of those and just, your AgenticOS, your AI Insights, your Creative Innovation Suite. Let's talk about the big product announcements you guys are doing and why you think they're important to the future of PubMatic.
Well, I think the biggest one by far and the biggest growth driver for us, in the coming years is AgenticOS.
Okay.
What we're doing with AgenticOS is MCP, Model Context Protocol, you know, generative AI enabling all of the core use cases on our platform. That's everything from inventory discovery and planning to media activation, to reporting and measurement, fee transparency, you know, guardrails for buyers. This extends well beyond purely the SSP capability set, and it moves into many more buyer-connected workflows. Many of the examples that I just gave you are actually things that buyers wanna do. We've been building towards this moment for the last four or five years, and we'll get into our data platform. We launched a couple of years ago a product called Activate, which allows buyers to buy directly in our SSP.
By doing that, what we're doing is, back to what I said earlier about revolutionizing the value chain, via AI, we're collapsing the value chain in the industry, removing complexity, and allowing buyers to buy directly in our SSP.
AgenticOS brings that all together, where what we see is that software, the legacy software user interfaces, those are very quickly becoming obsolete. I think one of the big advantages that incumbent DSPs have had over the years is, they've trained the agency trader or the advertiser, you know, trader to use their platform. What we see is that, with AI, the expectation of the customer is, "Hey, I don't need a complex interface. I don't wanna take two hours to set up a campaign. I wanna use AI and be able to do it in minutes. I don't wanna have to log in and troubleshoot a campaign. I want the system to tell me, 'Hey, this campaign has an issue. Here's three options to resolve it. You can pick, or maybe an AI agent will do that autonomously. That's exactly the value proposition that we're delivering with AgenticOS.
AgenticOS is a product, but then you need action on the side of the ad agency, presumably, to create an agent that can talk to your AgenticOS? Is that right?
A buyer will use an agent. They can use an agent that we're providing them. They can use an agent they've built or an agent that they've licensed. Using any of those buyer agents, our agent is part of AgenticOS, or they can connect in with their own agent to our system.
They can't use a DSP. The Trade Desk doesn't have an agent that can talk to your AgenticOS on the SSP side?
Yeah, as far as I'm aware, they do not. I don't know if they will in the future. If a buyer, if a DSP wants to connect with us via Model Context Protocol, we're open and available to do that.
Okay. The idea is to streamline and cut some of the fees. This is another, like, aspect of, trying to pull fees out of the ecosystem? That's the AgenticOS objective?
Our objective with AgenticOS is to deliver performance in the open internet.
Okay.
The key part of delivering performance is to simplify the ecosystem.
Okay.
When you have a DSP and you have an SSP, it's very difficult to actually deliver performance outcomes because there's a huge data divide between the platforms.
Okay.
If we think about, you know, what do walled gardens look like, right? There's a single layer of technology that connects t he ad campaign with the inventory and the audiences.
Yeah.
Right? It's Exactly. They're not living in this world of, "Hey, we have these disparate systems that have limitations in how they can connect.
Right.
Instead, they're one platform that combines everything. We're taking the same approach with the open internet. With Activate, you can buy directly in our SSP. With AgenticOS, y ou can do it using an AI agent. We expose all of the services agentically. That product is growing incredibly quickly.
I have a question. I guess I don't understand something. Activate is your ClearLine equivalent where a DSP can go directly to you through Activate and buy directly on your SSP, thereby disintermediating a bunch of the fees in the open internet. How is that different from AgenticOS?
Sure.
It seems it's exactly the same except one has to interact with an agent and one has to interact through a DSP.
I wouldn't compare Activate to ClearLine. I think there's some huge differences.
Okay. It's the same idea. It's search. It's, like, optimizing.
Well, our focus is on collapsing the value chain in the ecosystem so that a DSP and an SSP is not needed. Traditionally, that's done via software interface.
Okay.
A user will, you know, come into a UI or Activate UI.
Yeah.
They'll set up a campaign and they'll buy.
Okay.
With AgenticOS, they don't need to use the user interface. They can do this all via AI. They can log into Claude if they want to.
Right.
Anthropic solution.
Right, like Claude does have a buyer agent.
Correct. They can log into there.
Yes.
They can connect to our Model Context Protocol service. They can discover all of the services.
Right.
They can use that. They can do it via OpenAI. They can do it via Gemini. They can use any system that they want. They can deploy their own agent that they've built.
I guess my question is, aren't we just moving money from one pocket to another? The brand is the same, and you're just giving a brand, let's call it Coke. Instead of going through a DSP and paying the DSP fee, you're saying you can go through an agent, skip the DSP, and come straight. It's the same. All we're doing is moving money from Coke from one place to another, right?
Well, when you collapse the.
We didn't add. I guess, are we adding value to the ecosystem?
Yeah.
Are we just moving money?
Let me give you a couple of case studies around this. When you collapse the ecosystem like what we're doing.
Okay
You're creating a lot of value for the advertiser and for the publisher.
That's true.
That's what we're focused on.
The two endpoints get more value.
Exactly
You're pulling out fees. Okay.
You're pulling out fees. You're removing operational overhead and complexity.
Sure.
You are increasing the amount of data that's available for targeting. You are building solutions that are privacy safe. You know, big problem in programmatic is, you know, dispersion of data across the ecosystem. You're reducing the amount of time that's needed to set up and manage campaigns.
Yeah.
I'll give you a couple of examples of this. We did a demo, live demo on stage at a MarTech event a couple months ago with MiQ as the buying platform. Their trader got on stage and said, "Using PubMatic's AgenticOS. I can cut down the campaign set-up time by 98%. So, i t would take me a couple of hours to set up a complex campaign in a traditional DSP. Now there's no user interface. I just type in or I upload.
You talk.
Planning brief.
Yeah.
Or you talk, whatever you like to do. We can set up that campaign in 2% of the time.
If you compound that across the ecosystem.
Right.
You're talking about, you know, thousands of millions of hours of labor that's saved.
Um, but we also-
Millions of jobs lost.
Well, I think millions of jobs focused on higher value add activity, right? That's how I would define progress. Second is, by collapsing the value chain, we're able to deliver much higher ROI. Abovo Maxlead, largest independent ad agency in the Netherlands, probably nobody here has ever heard of them.
Sure.
They play an important role in that market. They're able to generate 40% lower CPMs for the media that they buy, and 30%-40% improvement on return on ad spend. That's because we are removing the DSP from that equation.
Yeah.
Compressing the supply chain.
That saves you 12%. Mm-hmm.
Well, I don't know anybody that's paying 12%. It's usually much, much higher than that.
Okay.
When you look at the all-in fee that DSPs are charging. It also removes a lot of that operational complexity. When a deal doesn't scale, for instance, now a trader on the buy side has to log into the DSP. They're working with somebody, let's say PubMatic at the SSP, or the publisher trying to figure out, "Okay, am I sending you the inventory? Are you seeing it? Why are you not bidding on it?" All of that complexity is eliminated when you're in a single platform.
Okay. All right. You say you've launched this cool product, AgenticOS. You say it's growing fast, but I don't get that. Really what you're waiting on is ad agencies, who are not typically very fast. They need to create a buyer agent. Let's call it a set of rails. There's programmatic rails that have DSP and SSP. You're saying, and I think people agree with you, we're moving to agentic rails. Those rails are different. They're a different caliber of railroad size.
Correct.
You need a buyer, and a traditionally slow-moving buyer. To create a agent that can talk to your AgenticOS. Who is it that is driving the growth in this product?
Sure. We launched the industry's first fully autonomous agentic campaign in December.
Okay
With an independent agency called Butler/Till.
Okay.
The advertiser is a malt beverage, Geloso Beverage Group.
Yeah.
Some canned cocktail, I believe.
Okay.
We shared in our earnings call last week that we now have 30 campaigns live.
With that guy or with that ad agency?
Across the ecosystem. Dentsu Amnet in France.
Okay.
We have some holdcos in Asia. We have Abovo Maxlead, independent agency in the Netherlands.
Okay.
I agree with you that the big agency holdcos, they tend to move slowly. However, that's not our sole focus. We have a big focus on the indie or independent agency market.
That makes sense to me.
They move very quickly.
They move fast.
Yes. That's also where a lot of the ad budget growth is. If we look at, you know, who are the clients behind the agency holdcos, specifically Fortune 1,000 advertisers?
Or 100.
100. Okay. What's the growth in media budget, advertising budget at a Unilever, at a Coca-Cola, at a BMW? It's flat, right? Maybe even down. When you get into the mid-market, you look at challenger brands, right? Vuori instead of Levi's. Chime Bank instead of Citigroup. That's where the growth is, and that's where the growth in ad budgets are, and that's a huge driver, of our growth.
Okay. That's who's adopting this AgenticOS?
Correct.
Okay.
Buyers don't need to wait. Again, they can log into chat ChatGPT. They can log into Anthropic. They can use that chat interface. They can use our chat interface. Some of them are, you know, hiring an Accenture or somebody like that to build a buyer agent for them. Some of them are building it internally, so they can connect their agent to our platform. There's no obstacle right now, other than I would say the human component of change management.
Leadership.
Leadership, exactly. That is preventing buyers from executing this way.
Four years from now, five years from now, which feels like three forevers in a generative AI ecosystem. What percent of the like of a programmatic advertising has moved to agentic, would you estimate?
I said two earnings calls ago that I estimate, I expect that by the end of 2028, 25% of media transacted in our ecosystem will be agentic.
No way.
By the end of 2030, it'll be 50%.
No way.
In the two quarters since.
I would take the under.
I think I may have been too conservative.
No.
Yes.
That implies the big ad agency holding companies, who are 75% of ad spend, have adopted this.
No. The big agency holdcos are 40-ish% of ad spend today. They were 55% about 10 years ago. Their share of the budgets has shrunk. It does imply that the big holdcos will have adopted this, and I fully expect that they will by the end of 2027.
Oh, boy. I'm taking the under on that.
Let's see.
I went to an upfront yesterday.
I'll come back.
Wall Street's been trying to kill upfronts for 30 years. I still have upfronts this week.
For sure.
That is the ad agencies.
More and more of the upfront is being transacted programmatically.
True.
Right?
I'm just saying these ad agencies.
That'll become agentic.
Are still doing upfronts.
Sure.
It's two decades since they've been relevant.
I mean, upfronts are salesmanship. That's all that is, right? You gotta sell your wares.
I'm just saying, you know, infrastructure and systems are ponderous, and they're slow-moving.
Sure.
By decades.
Upfronts are going to drive more programmatic selling and more Agentic selling and buying, not less.
Oh, maybe. We'll see. I think you're being a little optimistic. It probably doesn't matter to your company. You're small enough. You can grow no matter how much moves.
We can absolutely.
You know, if 10% moves.
Yes.
You're gonna grow.
Yes.
It doesn't matter.
I mean, our Activate product is growing 3x year-over-year. This is the direct buying in the SSP.
Right.
Now greater and greater portion of that is gonna be done agentically.
Yeah. I do wonder whether one cannibalizes the other, though. It feels like it's the same guy that would adopt your Activate ClearLine product as the Agentic product.
Agentic is a great on-ramp for our Activate solution, so I see them as going hand-in-hand.
Why would that be? Seems like they're on different rails, so they're substituted.
No, they're not on different rails. Under AgenticOS is a way to agentically access a variety of capabilities on our platform.
Right.
One of those capabilities is buying directly in our SSP.
Another set of capabilities which is Activate. Another set of capabilities is the SSP itself. Another set is Connect our data platform, so 300+ data partners. It's the ability to attach data to any campaign. All of these are now exposed agentically via AgenticOS. Think of AgenticOS as a, as a wrapper, as a front end to agentically access any capabilities on our platform.
Okay. Interesting. All right, fair enough. Okay, you don't want to talk about the other announcements you've made, like AI Insights or Creative Innovation. That one sounded interesting to me. You think AgenticOS is the number one driver of financial upside, revenue upside.
Correct.
For PubMatic?
That's right.
Cool. All right, one of the questions I think a lot of people, investors are particularly worried about is this take rate question. If both sides of the equation, both DSPs and SSPs are trying to cut out fees of the middlemen, doesn't that hurt everybody's take rate? I mean, what do you see happening to your take rates over the next three years?
We see our take rates as stable to growing.
Okay.
The reason is that we are bringing more value across the ecosystem to our customers. I mentioned earlier that with Activate and AgenticOS.
We are reaching deeper into buyer workflows.
Okay.
Right? As we do that, we're creating more value for publishers and buyers on our platform. That means we're able to charge incremental fees.
Okay.
For instance, when a buyer uses Activate, we charge a fee for that. When a publisher or a buyer uses our Connect data platform, and they activate data from any of 300 different parties, then we charge a fee for that.
Okay.
These are all, you know, take rate additive opportunities. We're growing our TAM, we're growing the value we create for the ecosystem. I also don't think it's correct to believe that as we do this or as others, you know, move in a similar direction perhaps, that there's gonna be a revenue collapse in the ecosystem. What I see is that as advertising gets more efficient and more effective, we should be moving, bringing more dollars into the ecosystem, and that's exactly what we're focused on doing.
Okay. Well, ideally 'cause you love to pull it out of the walled gardens.
For sure.
The more efficient you get, the more you look like a walled garden.
Not only pulling it from the walled gardens, but if the marginal effectiveness of advertising goes up.
Yeah
I mean, those dollars are not capped, right?
Yeah.
So a brand will advertise more in order to sell more product.
Yeah.
That would be the profit maximizing thing for them to do so that, you know, that describes the growth of Meta, for instance.
Fair enough. One of the meetings I had at POSSIBLE that was really fascinating was an SSP private, and he was saying that his curation, his data is growing 80% a year, and it's now like, I'm gonna say, 40% of his business. I'm gonna explain curation in a minute, but Is curation a driver for you guys?
Curation is a driver for us.
Okay. We need to explain what curation is.
Sure.
I just wanna make sure it was important to your business before I brought it up.
Yeah. Yeah, yeah. Absolutely.
Why don't you explain what curation is, and then we'll talk about this 'cause this is a new thing going on in SSPs.
Yeah.
I'm more excited about than Agentic, I want you to know.
Oh, interesting.
Anyway.
I think you have that backwards, but I'll come back next year.
I know, I know. I get it. We disagree.
We'll see.
It's okay.
We'll be honest.
That's what makes markets markets.
We'll see. Yeah, exactly.
Okay. first explain curation.
Sure.
Because I think it's really important.
Historically, advertisers have attached data on the buy side of the ecosystem, meaning in a DSP.
Sure.
Right? If you go into Google or, you know, any large DSP.
Trade Desk. Mm-hmm
Amazon, Trade Desk, sure. They have extensive data partnerships, where you can, you know, layer in Nielsen, Comscore, et cetera data.
Yeah.
Over the last, really over the last five years, there's been a conservative.
I would say two or three years really. It's more recent than that.
I mean, we've been generating revenue there for five years.
Okay.
It's accelerating though.
It is, yeah.
I'll give you that. We've seen a shift in how the ecosystem operates and a shift in where buyers are applying data to the sell side of the ecosystem in the SSP. There's a couple of reasons for this. One is the scale of audiences is larger, much larger in the SSP than it is on the, in the DSP because you have to do audience matching.
Yeah.
The further you get from the publisher and the consumer, there's a, you know, typically a 20%-40%, you know, falloff rate.
Loss. Yeah.
Exactly. Second is privacy. With privacy regulations increasing around the world, there's a need for consent from the consumer. Consent exists at the publisher, not typically with the advertiser. There's more data to be applied there. Third is, it can be done, applying data can be done much more efficiently. In many DSPs, you know, data has become 30%, 40%, 50% of their revenue where they're charging, you know, very high take rates against that data. We built, we've been building now for five-plus years a Connect platform, which is our data platform. I think it's probably one of the largest scaled data platforms in the ecosystem. There are over 300 data partners that are plugged in. Everything from, I mentioned, you know, Comscore, Nielsen, Circana. To a very rich set of commerce media partners. Klarna, Kroger, Intuit, Shopify. We've launched, announced at POSSIBLE Walmart and PayPal.
Wow.
Right. With PayPal there's $25 billion transactions, $400 million Venmo and PayPal accounts. This is, you know, very rich data that's sitting in our platform. What we do with curation, back to your question, is we package the data and the media for a buyer's objective. A buyer might say, "Okay, I'm launching a movie in June, and I need to find these types of audiences." We will take the right data from our platform, 300+ data partners. We will take the trillion-plus ad impressions that we process on a daily basis, and we will figure out, we will curate the right set of media.
To sell to them.
To sell to them. Exactly. To help them meet their objectives. We generate a data fee when we curate and leverage one of our data partners. Our data partners generate revenue, we generate a data fee in addition to our SSP fee.
I thought the big thing about curation was that you guys didn't charge on the data side. It was just free, but you just got a higher CPM because the data makes the CPM that you're selling at a more valuable CPM, so you just charge your take rate on a bigger CPM. I didn't know you were charging.
We do both.
For data separately. I would say most SSPs are not charging for the data piece.
Well, last quarter was our 40th consecutive quarter of profitability at adjusted EBITDA. You know, we like to generate fee revenue where we're adding value to our partners.
Okay.
Yeah.
Okay. Interesting. Okay, that's the first time I've heard it. Anyway, this trend, how much of your growth is being driven by attaching data to, How much of your revenue is now because you're attaching data to the sale?
We don't break out that specifically, Connect. A lot of the products that we've been talking about, Connect Data Platform with curation, Activate, AgenticOS.
Okay.
We put that in a category called, emerging revenue streams.
Okay.
That category, we just shared, last week, is 14% of our revenue, and it grew 80% year-over-year last quarter.
Okay.
This, you know, collective set of newer products.
Which is where we see the industry evolving to, is growing very quickly.
Okay. A year from now you're gonna sit on my stage, and I want you to break out this 14% and see which is bigger, and I bet it is curation and n ot Agentic. Betcha, a year from now.
Let's see.
We'll see.
Curation today is bigger, because it's been around longer, right?
Fair.
Agentic.
That's fair.
You know, two quarters. Two quarters old.
It's two. I was gonna say two months old.
Exactly. Exactly. Let's see. I think there's a decent chance.
I'm gonna be right, and you're gonna come in and go, "Oh, but it's 120% growth in Agentic from $2 million-$3 million.
Let's see. I think Agentic will be bigger by next year, but let's see.
I disagree with you.
Let's see.
We'll see.
We'll see.
It's discernible fact.
Yeah.
One year from now. We're gonna be here.
Either way we win.
Next time.
Either way, PubMatic will win.
Yeah.
Our shareholders will win.
I really am very high on this curation stuff, but okay. Okay. Where is AI already driving measurable ROI today, either by yield optimization or matching or cost efficiencies or you're not hiring as many people? Tell me where it's actually delivering financial returns today, and what you see the biggest upside from it, specifically generative AI over the next year.
Sure. I think a good framework, let's say for our customers, and then we can turn to internal application of AI.
Okay.
Is really just efficiency and effectiveness, right? I think whenever we're thinking about innovation in advertising, we can think about, like, those two categories. Efficiency we've talked about already quite a bit, right? Significant reduction, 7x time reduction in campaign troubleshooting and management. These are all, you know, efficiency measures.
Yeah.
Where a significant amount of labor is saved. Effectiveness is where there's still a lot of work being done. As AI scales up, right, from very early stages. A couple of examples of this, right? We've been able to reduce fees taken, you know, in a media buy by 40%, right? That means lower CPMs, 30%-40% more working media for an advertiser. That's a, you know, great example of increased.
You're an SSP. All you should care about is higher CPMs for your publisher.
The point is how do you generate higher CPMs? It's not a zero-sum game to say, "Hey, we just wanna get the advertiser to pay more." What we focus on at PubMatic is delivering higher CPMs by delivering more performance for the advertiser. When we deliver more performance, the advertiser is happy to pay more. Which is how we connect the dot to higher CPM. That effectiveness and performance is really important. We talked earlier about 300 data partners in curation. One of the biggest challenges in curation is what data and what media should I apply to this advertiser's campaign? This is where I think the big wave of AI and effectiveness is going to come, which is using AI to figure out which dozens of the 300 data partners.
Of the 2,000 publishers on our platform, you know, which hundreds of those publishers. For instance, today, when that RFP comes out, hey, a studio's launching a movie, the agency's working on behalf of that studio to launch that movie, they typically will send an RFP out via email. It's very old school, right? Send an RFP out to 20 or 30 publishers. There's no reason that that shouldn't be happening across hundreds of relevant publishers. That's where a seller agent for a publisher and a buyer agent for an advertiser can do that matching job much better. This is yet to come, but this is where I think there's a lot of opportunity around effectiveness.
Okay. I think the question though was where is AI already driving measurable ROI? Is there anything on the cost side? I mean, usually CEOs sit up here and they talk about cost 'cause they can see those.
Sure. We talked about that already, the efficiency piece. That's where I started the answer, right?
Okay. Is that helping your FTEs not grow as fast?
Okay.
Margin improvement somewhere?
Everything that I talk about is for customers.
Right.
We can talk about, you know.
Internal.
Internal. Right. We operate a self-serve platform, so we're not sitting there, you know, logging into our own system and, you know, have teams of ad ops people. Our biggest investments are in product and engineering and in sales. About 80% of the code on our platform now is generated via AI.
Those are good numbers.
So it's a very healthy number. What we've been able to do is ship product, more product much faster over the last several years, even as we've flattened or decreased the size of our engineering team. Significant improvement in efficiency due to AI.
Okay.
Our current focus, 'cause that 80% obviously is gonna hit a ceiling is in every other function: finance, marketing, you know, sales ops, account management. AI enabling every single one of those functions. If we looked at our headcount number, you know, from our K, it's down somewhere in the 5%-10% range.
That's fantastic.
Even as we're growing revenue.
That's fantastic.
Yes.
Yeah, we keep track of FTEs religiously for every one of our ad tech companies.
Yes. I know you publish those stats.
Google Yeah. Meta's down 8% over the last three years.
Yeah.
They sort of lay off 10%.
They hire 20%.
Hire 10% and then they, you know, they sort of do this.
Exactly. We have not been doing that. Yeah. What we have been doing is certainly investing from a CapEx perspective in new AI-enabled capabilities. We historically have invested CapEx, this year's number's in the $15 million-$19 million range. Historically we've invested in two areas. One is increasing capacity. The second is in new capabilities.
An important thing to note is that we own all of our own ad-serving infrastructure, so we do very little in the public cloud. We've had this approach for 10+ years. It's a key driver of the cash that we generate. We own tens of thousands of servers around the world. This year, one of the things that's different is we are putting no CapEx into impression volume growth. We've been eliminating redundancy, across our platform, redundancy of impressions, and instead all of that CapEx is going into new hardware to deliver AI solutions, and we can talk about our NVIDIA partnership.
Yeah, let's do that. Let's do that next. Let's go to the NVIDIA partnership.
Great. We've been working with NVIDIA for, I think three years now, where we co-innovate around solving specific advertising domain problems with the help of their, you know, hardware and software engineers with our software engineers. I'll give you an example of this. I think probably people have heard of inferencing.
Yeah.
Which is, okay, we have a model that's been trained and now, you know, we want to query that model to ask a question or figure something out. One of the big challenges in our ecosystem is I've got 1 trillion ad impressions per day. Which impressions should I send to which DSPs? We have 150 DSPs that are bidding on our platform, some in particular geos. They have different campaigns at any given point in time, you know, different ad formats, different audiences. This is a key inferencing problem. We have worked with NVIDIA to develop inferencing solutions that help us figure out exactly which ad impressions to send to which DSPs.
The reason we have this integrated stack of software and hardware is we're able to control both the software and the hardware level versus layers, versus if we were in public cloud where we can only control the software. We've deployed specific NVIDIA Triton chips. Within our infrastructure to solve exactly this problem.
Hmm. Very interesting. Questions from the audience for Rajeev. Yes, Scott?
Yeah, you talked about Google as the more patient monopolist that you always have to deal with.
Yes.
How do different POSSIBLE remedies that the judge might impose affect your business?
Sure. I think everybody's waiting, you know, with bated breath, at least in this track of your conference, for that remedies verdict to be handed down.
Edge of our seat.
Yeah, exactly. Just to give you some numbers, we estimate every point of market share shift is worth $50 million-$75 million of revenue to us, of which 80%-90% would drop to our bottom line. We have a, you know, high fixed cost, low variable cost business. You know, every point of market share that shifts as a result of remedies. There's, you know, two classes of remedies. There's behavioral and there's structural. We don't know, you know, what the mix of those will be. I think the behavioral remedies, which is where Google has to change some of its practices or policies, those, in my opinion, can be implemented very quickly within weeks to months.
We could see benefit, you know, as early as later this year. Structural remedies are things like Google having to spin, you know, certain entities out or, you know, sell them off, operate them at arm's length in some way. I think those are likely to take longer to come into effect.
I mean, presumably structural remedies would be better for you without any time constraint, but that would take longer. Behavioral remedies might in fact be better in that present policy sense.
I think that is true, I also don't think it's either/or, right? Of course, I don't know what the judge is gonna decide, but it could very well be that there's behavioral remedies in the run-up to a structural remedy. Because based on, you know, my interaction with the judge in the case, she's, you know, very, very aware of the harms that Google has created across the ecosystem and, you know, focused on correcting those harms.
Any other questions? Yes, sir.
I think you just said you work with 150 DSPs. How top heavy is that? Like how much of your revenue is driven by the top four or five today, and how is that gonna trend over the next couple years?
Yeah, sure. There's been a very significant unwind in concentration over the last couple of years. I would say, you know, if we went back like five years, it was a story of DSP consolidation. You know, there's a couple of DSPs that, you know, were driving most of the market share growth. If we look now, you know, I talked earlier about the mid-market advertiser. Many of the two of the largest incumbent DSPs, they're not really focused on that mid-market arena, right? There's, you know, a lot of fast-growing mid-market DSPs, companies like StackAdapt, for instance, Viant, who I think you have here.
Yep.
Later this week. That number is growing significantly, and what we see is that the concentration is reducing on a year-over-year basis. I think one of the metrics, we haven't broken out the concentration specifically, but we did share that mid-market DSP cohort is growing 20% year-over-year on our platform in terms of the spend or activity.
Very, very significant growth.
It's amazing there's such low barriers to entry for DSPs.
Yes, there are very low barriers to enter, and I think with AI it's a huge opportunity to remove what I think is the last barrier, which was software UI incumbency. Right? Again, if you go into, you know, a big agency holdco, the big challenge for them to move off of a DSP if they wanted to expand their portfolio is, well, hey, I've got thousands of traders around the world, and, you know, they need to know Google's DSP in order to buy YouTube, they need to know Amazon's in order to buy, you know, Amazon Prime Video, and for, you know, shopping use cases.
We don't have time to learn, you know, four other DSP UIs. Now with AI, the expectation is, well, if you can type, right, and you know you don't even need to know English.
Talk.
Yeah, or talk. You can do it in any language, then you can give an agent your instructions and execute those. That's why one of the reasons why we see such great growth with AgenticOS and Activate.
Any other questions? Okay. Let's go to your commerce initiatives.
Yeah.
Can you talk about your commerce media platform and your Connect platform?
Yeah.
How your commerce products are driving upside at PubMatic?
Sure. Again, our focus is on performance advertising in the open internet. That for us is the North Star. Commerce partnerships, commerce data is, you know, critically important to that. You know, where most of the growth is in the ecosystem is where you have known user identity, so think of CTV and mobile app, and where you have the ability to measure the outcomes. Commerce media obviously is a key area of that. We have dozens of commerce media partnerships. I went through some of the names earlier. Last week what we announced was POSSIBLE last week?
Two weeks ago.
Two weeks ago, okay. What we announced at POSSIBLE was a partnership with PayPal and with Walmart.
With Walmart, obviously, Walmart Connect, they have great shopper, you know, knowledge, audiences at scale. Absolutely. They're combining their data with the streaming inventory and the mobile app inventory on our platform. Giving their advertisers, whether they're enterprise advertisers or SMB advertisers, the ability to find those Walmart audiences off of walmart.com.
Okay.
Right? Now an advertiser can say, Great, Walmart knows that this user is, you know, high propensity to buy my product.
I wanna reach them when they're watching 'Landman,' you know, on Paramount, or they're watching, you know, 'Bill Maher' on HBO or whatever the case is. That's what we're doing with Walmart, obviously a huge scale and partner. With PayPal, there's a couple of things that we're doing there. One is very similar, taking that data and making that available to any advertiser, not just, you know, PayPal's clients. PayPal is also a user of Activate, so they're buying directly in our SSP when they go out and they sell campaigns. Then the third thing is they launched PayPal Ads ID, their own ID graph.
Identifying who users are, and we're merging their graph, integrating their graph with our graph, so that anybody, any advertiser that's using PayPal's graph can find those users on our platform. All of these are around identifying and delivering high-value, relevant campaigns to the user, which is where we drive return on ad spend. Which is what pushes the CPMs up for our publishers.
Okay. Okay, we're almost out of time. Just wanna make sure there's no other questions for Rajeev before I call it. No. Okay, I will call it there. We have 30 seconds left.
Great.
Perfect timing.
Thank you. Thank you so much, Laura.
Okay, thank you so much.