Marcus Johnson (00:04):

Hey, gang. It's Thursday, February 19th. Jeremy, Nate, and listeners, welcome to Behind the Numbers, an EMARKETER video podcast. I'm Marcus, and joining me for today's conversation, we have two normally New York-based folks. We have senior director of Content, Jeremy Goldman.

Jeremy Goldman (00:17):

Hey there. And can I just say, happy birthday Nikola Jokic, who turns 30 today?

Marcus Johnson (00:22):

Oh, God. He's only 30. I was hoping he'd be out of the league soon. Guess not. And we're also joined by principal AI analyst, Nate Elliott.

Nate Elliott (00:34):

Hello. No place I'd rather be.

Marcus Johnson (00:36):

Hey, fella. That's not true, but thank you anyway. Today's fact, how did the Super Bowl halftime show get so popular? So we just had the Super Bowl, just happened. Congratulations to the Seahawks. Congratulations to the Patriots. Not really because you won a lot in the last couple of decades. Super Bowl I was at the LA Coliseum where Vince Lombardi's Green Bay Packers destroyed the Kansas City Chiefs. Playing the halftime show was the University of Arizona Symphonic Marching Band, Grambling State University Marching Band, American trumpeter and band leader Al, I think it's pronounced Hirt, H-I-R-T, and the Anaheim High School Ana-Hi-Steppers Drill Team and Flag Girls.

(01:26):

So they used to perform, used to be marching bands at halftime, but folks weren't particularly enthralled with the mid-game entertainment it seems. And so, in 1992, Fox tried to take advantage of this by airing a live episode of In Living Color, a show during halftime it worked. Millions of viewers turned from the Super Bowl. It was 17-nothing to Washington, so maybe all the more reason to see what else is on, but this panicked the NFL. And so, in 1993, they went and they booked Michael Jackson, whose performance drew higher ratings than the game itself with over 100 million people watching, ushering in the era of the pop star model.

Nate Elliott (02:03):

Fire Marshall Bill.

Marcus Johnson (02:06):

I'm sorry, what?

Nate Elliott (02:07):

Fire Marshall Bill, the best In Living Color, well, my favorite In Living Color character.

Marcus Johnson (02:11):

Oh, I see. Didn't know who Bad Bunny was until two weeks ago. Most streamed person only on Spotify, Marcus. That's who. Anyway-

Nate Elliott (02:22):

Just a casual multi-Grammy winner.

Marcus Johnson (02:26):

I knew him as an actor. I saw him in Bullet Train.

Nate Elliott (02:27):

[inaudible 00:02:28] superstar.

Marcus Johnson (02:29):

Well, he also does music, apparently.

Nate Elliott (02:31):

Yeah.

Marcus Johnson (02:31):

Anyway, today's real topic, the three big questions surrounding Google. All right, we set the table first. What happened last year? Google made over 400 billion for the first time in 2025, growing 15% more than the year before, and then nearly 300 billion of that 400 was from advertising as up 11.4% a fraction above 2024 Q4. What did it look like? Ad revenue was up nearly 14%. That's very, very fast. It's the fastest growth in nearly four years, three and a half years, so very good Q4 for ad revenue. Network was down year-over-year. No surprises there. It's the 14th consecutive quarter that that's been down, but still made $8 billion, so it's not nothing. YouTube, that grew 9% in Q4 year-over-year, down from 14%, I should say. There's some of the numbers, the backdrop for Google at the moment, but we're talking about the three big questions surrounding this company. Nate, I'll start with you. What's one of the big questions you think Google is asking themselves or analysts, journalists are asking of Google?

Nate Elliott (03:41):

I think that folks want to know whether Google wants AI search to replace traditional search. Well, they've got lots of different tools available to people who want to use AI to find information online including AI Overviews, AI Mode, Google Gemini, and a variety of other tools as well. You've seen them pushing some of those AI tools into the general search experience, and really, the question is one of, how far down that path does Google want to go? Do they want to get to the point where every search result that's served up by Google.com or the other services where people search for information on Google is responded to with an AI response? If so, what would that look like? Does that mean an AI Overview on every search result page? Does it mean just defaulting people into Gemini or AI Mode as a result of every search? It's a balance that they need to get right. I'm not sure they have an answer to it, but as someone who watches them pretty closely, it's one of the things I'd love to know.

Jeremy Goldman (04:55):

I mean, the great thing Nate, as you know is they have such an insane install base, which I think gives them a lot of... Let's say if they say they answer to that question is, "Yes, we want to push people in that direction," well, then great. They start to do it, and if they notice that that's creating more friction, then they start to pull that back. They obviously are going to do that in the way of light experimentation with their user base. Also, I think that as we all know, time spent with various platforms really does matter. If you can keep people on platform more and serve them up more ads, that's generally a good thing. So I mean, my thinking is they're going to be leaning into that, and then they're going to just see how far they can lean. Does that take off some users, and if so, is that just part of the cost of doing business? Because I do think their whole entire mode of making money is in a really transformative spot right now.

Nate Elliott (05:52):

Yeah, I mean, I don't want to get too far into my next question that I have for Google, but the point you make about ads, Jeremy, is a really good one. Yes, their model is selling ads, and most of that 400 billion that Marcus mentioned was advertising revenue. And the thing is, if people kept using traditional search the way they have been for a quarter century, Google knows how to make money from that. They're very, very good at making money from that. What they don't know yet, because no one knows the answer to this yet, is how do we make reliable, repeatable, and enormous advertising revenue off of AI search or AI interactions in general?

(06:32):

We're just at a point where OpenAI is starting to test ads in ChatGPT. We've seen Google move very slowly in introducing ads into AI Overviews and AI Mode and there are no ads in Gemini just yet. So there's a question of which version of a search response, whether it's TenBlueLinks or MAPSearch or ProductSearch or AI Overview or AI Mode or Gemini, which version of all of those options is actually going to best serve the people who are looking for information? Also, how can Google move people to the thing that best serves their users while also not endangering that $400 billion that Marcus just mentioned?

Marcus Johnson (07:18):

Yeah, they're in no rush, are they, to turn it back on the search line item. As you mentioned, Nate, three quarters of the ad business coming from Search, it grew faster in 2025 than it did in 2024, and in Q4, it grew faster than it had in three and a half years, so a line item that they don't want to mess with too much. Jason Aten of Inc. had a nice line. He said that the question of the last few years has been what happens to Search when AI gets good enough to replace it. Turns out, Search isn't losing to AI. It's funding it. I think that's a good one. That's a good one to start with. What else do we have? Nate, did you have something to follow on and something else to throw in that's kind of related to this?

Nate Elliott (07:55):

Yeah, I mean, the ads question that Jeremy brought us into, which is how quickly and how, in terms of mechanism, format, and everything else, is Google going to start introducing ads into their AI products?

Marcus Johnson (08:09):

Yeah, that's a good one. Jeremy, what do you think is top of their mind?

Jeremy Goldman (08:15):

I think that the question about... So we talked a lot about ads there, but I mean, I think in terms of can Google monetize AI commerce indirectly while others try to chase direct checkout? There's a lot of different approaches of how people are going to be making money and whether or not they can make enough money to fund all of this AI expenditure. That's a whole nother story, but I think that in terms of how people want to shop in the future and how much of an opportunity agentic commerce might wind up being, I don't think anybody really has the answer to that right now.

(08:51):

I'm curious what you both think, but I kind of believe that there are some categories where this could be really, really great, where it's just you're doing research on something and you want to auto replenish something that you have a high degree of familiarity with. And then, there are some things where you're going to do a ton of research, and you will never necessarily ever check out through like an AI search mode, let's say. So I mean, I think that Google has a strong POV here, but I think that we really have to be looking to see how do the consumers of today react to these newfangled opportunities to check out, and I don't think anybody has a clear answer to that yet.

Nate Elliott (09:30):

Jeremy, we have the answer to that because we just published our EMARKETER AI-

Jeremy Goldman (09:34):

Hello.

Nate Elliott (09:35):

... Commerce Forecast a few weeks ago. But before I get into that, I mean, isn't this just an advertising conversation as well? I mean, you talked about indirectly monetizing AI-driven commerce, and OpenAI has planted their flag on this one, and they're charging a 4% service fee on Shopify transactions to the merchant, which is actually more than Shopify itself charges those merchants. So they're trying to make money from instant checkout by just charging a relatively large and odious fee to the merchants. Whereas Google, when it launched UCP, has said, at least at this point, they're not planning on charging anything for transactions through the Universal Commerce Protocol. So the only way I can see them making money from that is either finding other ways to charge merchants or by selling ads into those interactions.

Jeremy Goldman (10:29):

The very fact that they're going to be charging people through view-based pricing, essentially not conversion based for ads, I think what that tells me is that a lot of people really believe that AI has a strong ability to influence and to do research a lot more so than this is going to be the way that people check out and engage in commerce in the future or at least in the very near term future. I know we have the forecast.

Marcus Johnson (10:56):

What were some of the takeaways from that?

Nate Elliott (10:57):

Well, I mean, two big takeaways. One is that this is going to grow very quickly, but on a very small base as you would imagine, so that even at the end of the forecast period, which I think is 2029, less than 9% of US online retail commerce will be driven by AI engines like ChatGPT and Gemini. The other part of that is throughout the forecast period, the vast majority of AI-driven US online retail commerce sales, say that five times fast, will happen when people click a link in an AI chatbot and complete the transaction on a retailer site or app as opposed to happening inside ChatGPT or Gemini. 95% of the AI-driven transactions this year will happen on retailer sites and apps. Only 5% will happen inside the chatbots. Even at the end of the forecast, I think we're forecasting an 80/20 split. 80% of those transactions will be on retailer sites and apps, and only 20% will be inside the AI engines.

Marcus Johnson (12:06):

Okay, interesting. I think it's a good one. What else do we have? Any other main questions for that? I've got one here about... This one just seemed to be the headline of most articles written about Google in the last week or two, which is that are they planning to spend too much on CapEx, which I think seems to be a big question for a lot of folks, especially investors. So this year, Alphabet expects CapEx spending on data center servers, AI infrastructure, things like that in the range of 175 to $185 billion, about 30% higher than the Street was expecting. Jason Aten of Inc. thinks that spending 185 billion that you made selling search ads in order to make sure you stay in the lead seems like a reasonable move. Dan Gallagher of the Wall Street Journal though says, "Spending what could be 40% of annual revenue on AI chips and related infrastructure is still a sizable gamble."

(12:58):

Shannon Carroll of Quartz thinks there could be two outcomes here, kind of leaning into what Jason's saying and then also what Dan's saying here. There's a version of this story that ends in triumph. Cloud keeps compounding, AI products become default habits, and spend looks prescience the price of buying the future before someone else does. Says there also a version where the returns arrive slower than depreciation and Wall Street starts treating AI like a lifestyle choice with bad unit economics. Nate, what do you think about this being a big question for them?

Nate Elliott (13:29):

I think all of those takes are correct. It is an almost absurd amount of money to spend chasing this one thing. Also, Google uniquely perhaps can afford to. I mean, we just talked about their enormous and still growing revenues. They're not spending 175 billion of investors' venture capital. They're spending 175 billion of the money that they have earned through other parts of their business. And so, they can make it make sense in a way that, say, OpenAI is struggling right now to make it make sense. OpenAI is trying to make money from subscriptions and ads. They don't have a technology business per se, and Google is spending similar amounts to OpenAI per year on AI related CapEx, but they actually have two distinct ways of making money from artificial intelligence. The first is subscriptions and advertising, and the second is they sell chips. They sell enterprise services, and not just the LLM as an enterprise service, but they sell the technology that powers other companies, LLMs. They have a lot of ways of making money from that capital expenditure.

Jeremy Goldman (14:44):

I think that part of the whole entire thing here is that it's not just Google. It's a lot of other companies here, particularly Meta and Amazon, I would say, that are spending quite a lot in this area. Maybe you can say Apple as well and Microsoft, where if it leads to efficiencies in your existing businesses like advertising, great, but also, it might create new lines of business and new sources of revenue. If that happens, then that's kind of almost what you need to justify this. It is worth noting that some of these companies like Google and Meta, their overall share of the digital advertising ecosystem was quite big before these major expenditures. This has helped them continue to be very dominant, but I still think that essentially, at some point, Wall Street might look at it and say, "Hey, was this all worth it considering the fact that you had a major lead on a lot of the second and third tier players before you were spending this amount? So was it really necessarily entirely justified?"

Nate Elliott (15:50):

I mean, and we talk about the different ways of making money from AI. OpenAI is panning for gold, Meta is panning for gold, Google's panning for gold, and it's also selling shovels, right? It's also selling the equipment you need to go pan for gold. And by the way, in case that doesn't work out, they've got a diamond to mine out back, so they're in a pretty good position.

Jeremy Goldman (16:13):

All of the cloud businesses, I mean, have been growing when we're thinking about the major cloud providers. I know it's a little nerdy, but they've all benefited from this AI, quote, unquote, "revolution." They were growing at a pretty great rate for years. And I would say that this has kind of propped up those businesses and will also for the foreseeable future. So that's also pretty nice. It puts less pressure on Google's ad revenues, let's say.

Marcus Johnson (16:40):

Yeah, I thought that was one of the questions potentially, Cloud fast becoming the golden child. Shannon Carroll, of course, again, was writing Google's Cloud revenue jumped by nearly 50% to around 18 billion, the kind of acceleration that stops being a segment update and starts being a plot twist. Better than that, Cloud operating income surged past the over 5 billion mark. Well, that is 45% higher than Wall Street's targets, more than double last Q4, pushing operating margin to just over 30%. Saying that Alphabet is now comfortable framing Cloud in milestone language, a $70 billion-plus annual run rate exiting in 2025. That's why investors have been begging Big Tech to deliver proof that the AI build-out isn't just a bonfire of GPUs, but something that can throw off real profit whilst it scales.

Nate Elliott (17:29):

A bonfire of the fan of these perhaps.

Marcus Johnson (17:34):

I thought though maybe this number of this CapEx... Circling back to the CapEx number, maybe it's high because Meta said we're going to spend 115 to 135, and compared to that, it looks like a lot. Maybe it's high because you can take the net income Google made in the last year and a half and say they're basically spending that much, all of their net income for the last six quarters on AI infrastructure this year. And then, the big concern I guess here is the Evercore senior MD, Mark Mahaney had said, "Could there be another DeepSeek moment coming? Could there be another moment where a company says, 'Look, we did a similar thing with way, way less,'" which would probably be the big concern.

Nate Elliott (18:16):

I mean-

Jeremy Goldman (18:16):

Historically, probably.

Nate Elliott (18:18):

Yeah. I mean, I can see that happening, but to your point, Marcus, why does it seem like a lot? It seems like a lot because it is a lot.

Marcus Johnson (18:25):

A lot.

Nate Elliott (18:25):

If you go back five or seven years, you will struggle to find an example of any individual company spending more than, what, $25 billion in CapEx in any individual year. And now, everyone's spending between 100 and $200 billion on CapEx every year. This is not a financial analyst, but this certainly looks, in my experience, like an unprecedented moment of investment. I mean, the stakes, and we've talked about this internally, Jeremy and I, across the desk in our office. The stakes for this are, on the upside, you control how everything in the world works, and on the downside, you lose your investors' money. That's a very big reward and a relatively small risk for the heads of a lot of these companies to be making, and that's why they're placing some of the biggest bets in history. But yeah, these are actually unprecedented amounts of money.

Marcus Johnson (19:20):

Anything else, gents, big questions right now?

Nate Elliott (19:23):

I've got 100, and I'm sure Jeremy has a bunch that aren't AI related, but I'm the AI analyst, and so I've got a hammer and everything looks like a nail, and I want to actually know the answer to why, but I want to know what they're going to do about it. Why does Google have two separately branded full-fledged LLM chatbot models in Gemini and in AI Mode, and what the heck are they going to do about it? And the answer to why is because they have two different teams running these models. The AI Mode piece is run by the Search team, and Gemini is run by the AI organization at Google, and that's why there are two different things, but it's almost nonsensical from a branding perspective, from a product perspective, from a product development and decisioning perspective.

(20:13):

Somehow we've gotten the point where Microsoft is better at branding its AI tools than Google is, and in an ideal world, Google wouldn't have AI Overviews and AI Mode. It would have Gemini overviews and Gemini mode, and they'd have one organization in charge of all of this. But because of the way that they're structured, that's not the case. I want to know if they're ever going to get around to making life easier for consumers and analysts, and oh, yeah, maybe stop duplicating efforts by having two completely separate full-fledged tools.

Jeremy Goldman (20:45):

Yeah, and I want to know when Gemini's going to have ads. I actually think that on one hand, they could do it tomorrow even though they said they won't. But I actually think that Google's dream would be to not monetize it for as long as possible and for ChatGPT to be so incredibly desperate for revenue that there's a ton of ads on it, and Google just sucks up a whole lot of user data and queries because everybody uses Gemini because it has no ads, and they can afford to do that versus ChatGPT cannot, given the lack of any real major revenue streams.

Nate Elliott (21:19):

How desperate do you think OpenAI is right now, Jeremy?

Jeremy Goldman (21:22):

Eight?

Nate Elliott (21:22):

Relatively-

Jeremy Goldman (21:22):

Is that an eight? Maybe.

Nate Elliott (21:24):

Eight out of 10?

Jeremy Goldman (21:27):

What would you say?

Nate Elliott (21:28):

14 out of 10, I don't know. In one 10-day window in January, they announced ads. They quoted $60 CPMs, which is outrageous, and $200,000 minimum spends for an ads trial, which is pretty high. They claimed that they would be taking a share of the profits that Enterprise has generated by using ChatGPT. They slapped a 4% service charge on Shopify transactions. I'm missing a couple of things. Oh, they said they were going to sovereign wealth funds to try to attract hundreds of billions of dollars of additional venture capital. I mean, it was just a shocking number of announcements, all of which pointed to one thing, which is we are in desperate need of more money over here-

Jeremy Goldman (22:16):

That's a code red.

Nate Elliott (22:17):

... all dropping in 10 days.

Jeremy Goldman (22:18):

That's a code red. A code red for you right there.

Nate Elliott (22:20):

That's a code red on top of a code red, I think.

Marcus Johnson (22:23):

So I've got a question for each of you based on what we've just talked about, and then we'll come to a consensus top three. So Nate, my question for you, based on what you've just been saying, I think it's a fantastic question. My question is how soon do you think they need to figure that out? How long do you think they can keep going with Gemini and AI Overviews separate?

Nate Elliott (22:39):

I mean, they could do it indefinitely. If you play out the thinking that Jeremy just talked about, it's credible to think that maybe they want there to be separate tools because AI Mode and AI Overviews living inside of Google Search, well, this is an environment where they already have an ads business where people are accustomed to seeing ads. It's interesting that AI Overviews got ads about a year ago, and still ads within AI Overviews are exceedingly rare at this point. Now, that doesn't matter so much because of where AI Overviews show up on the search results page. There are already ads nearby. So who cares if they're inside the AI Overview or next to the AI Overview.

(23:23):

But it highlights this point that people are used to seeing ads in search results. And if they can default people into AI Overviews and AI Mode from the traditional search experience, then there might be less resistance to seeing ads in that environment. Maybe they want to keep Gemini completely separate, and they want to keep it ad-free. Maybe they like having them separate, but I still think it's a misbranding opportunity. It's confusing to consumers. Even if they run the product separately, I would like to think that they might rationalize the brands at some point.

Marcus Johnson (23:57):

Yeah. Jeremy, this leads really nicely to the question I had for you as we talked about, should these products have ads, not have ads? When's the right time? You wrote a piece talking about Anthropic committing to keeping Claude free of advertising. You say arguing the ads inside AI conversations would undermine trust in a tool built for work reasoning and sensitive problem solving. What are your thoughts on a bit more on that story?

Jeremy Goldman (24:22):

Yeah, I mean, I think one thing that's really important is that if you're Anthropic, you're kind of like the David versus Goliath. Then, you have to figure out some kind of way to differentiate yourself. So while it's true that like, "Oh, you're giving a lot of sensitive information to a chatbot," you kind of did that with Google for years and everywhere else around the internet. So I think for Anthropic, they do need to have that level of differentiation. Obviously, I think that it might just be that they did the calculus and said, "It would be kind of difficult for us to create an ads business right now, and we make our money in a few other ways, so let's just stay along that route."

(25:01):

I would definitely say that generally speaking, the history of the internet is we have spaces that don't have ads, and then somebody says, "We're going to start showing ads there," and people are like, "I can't believe that. That might not work." And then, it does over time, and we get used to it. Marcus, we're getting to the point that if you have a tattooed ad on your forehead during... Actually, sponsors, if you're listening to this, Marcus is open to this.

Marcus Johnson (25:28):

It's stupid.<

Nate Elliott (25:28):

We can do that. Yeah.

Marcus Johnson (25:30):

What do you mean? We can do that. One second. Stuart, don't listen to them.

Jeremy Goldman (25:34):

Even like some kind of 3D thing that we put on and we shift it every single... That might be-

Marcus Johnson (25:39):

[inaudible 00:25:40]

Jeremy Goldman (25:40):

20 years from now, that might be a thing that everybody does. You don't know. So I think that people do get used to seeing things in new surfaces, and Anthropic just kind of had to lean into this differentiation for now, but their circumstances could very well change, and then they'll quickly delete the press release that they put on their website saying that we're basically ad-free forever.

Marcus Johnson (26:02):

So I move around a lot. So I have a capsule wardrobe, and so sponsors, if you want me to wear a T-shirt, I'd do that because then I'd have more clothing.

Nate Elliott (26:09):

No. Henna tattoo, right on the... Look at that forehead. It's dying for a henna tattoo.

Marcus Johnson (26:15):

Maybe, for the right price. Just kidding. Please don't make me. All right. I think this is a good group of questions, gents. I'm just going to recap some of them, and we'll figure out what we think the top three should be. So Jeremy, quickly give me your comments question again. What do you think the headline there is?

Jeremy Goldman (26:29):

How can Google monetize AI commerce indirectly while some others might chase a direct checkout?

Marcus Johnson (26:37):

Okay. We also have, do Google want to replace, Nate, you said traditional search with AI search. You also said when and how will they monetize AI search. We talked about planning to spend too much on CapEx, and then you also had, why do they have two fully fledged AI chatbots and when would they reconcile these tools? What do you think? What needs to go?

Jeremy Goldman (27:03):

Oh, and are they now too big to fail? That was kind of like my other one or too important to break up.

Marcus Johnson (27:09):

Okay, what are we dropping? What are we keeping? What needs to be... Nate, what's your top, top question?

Nate Elliott (27:14):

It's got to be monetizing. I mean, it ties everything together, right? The CapEx, the transition from traditional AI search, it's all tied together by this question of, can you actually make money from these investments and from this change in consumer behavior?

Jeremy Goldman (27:28):

Okay, I agree. I think that's number one.

Marcus Johnson (27:31):

Okay. Jeremy, you want number two?

Jeremy Goldman (27:34):

I think the commerce might be number two because it's like an ancillary. It really gets at the same question, but from a different lens.

Nate Elliott (27:42):

I mean, $144 billion in US online retail commerce in 2029 will be driven by AI tools. So like I said before, it's less than 10% of the total spend in that category, but it's $144 billion. That's not nothing.

Marcus Johnson (27:59):

Yeah. All right. And then, I like the chatbots. You have two separately branded. What are you going to do about this? I really like that question. So we'll go number one, when and how will they monetize AI search? And number two, how can Google make money from commerce, what that looks like? And then, number three is the two separate AI chatbots maybe becoming one. Maybe one takes over, and they sunset the other. Maybe they keep both. Yeah.

Nate Elliott (28:26):

It's weird, Marcus. You invited an AI analyst and an AI-focused briefings director today, and we ended up with three AI questions.

Marcus Johnson (28:36):

Hang on a second. How did that happen? I think it's a good list, so we'll keep it. Thank you so much to my guests for hanging out with me today. Thank you first to Jeremy.

Jeremy Goldman (28:43):

Thank you, and also happy birthday to Millie Bobby Brown, who turns 21 today.

Marcus Johnson (28:48):

21? The Stranger Things person?

Jeremy Goldman (28:51):

Yeah.

Marcus Johnson (28:52):

Oh, crying out loud. I didn't need that. And also, Nate, thanks for being here.

Nate Elliott (28:58):

I don't have any idea whose birthday it is today, but-

Marcus Johnson (29:00):

Thank God.

Nate Elliott (29:01):

... thank you for having me.

Marcus Johnson (29:02):

That's why I invite you and nearly don't invite Jeremy. Thanks so much to the whole production crew and to everyone for listening in to Behind the Numbers, the EMARKETER Video Podcast. Subscribe, follow, or leave a rating review if you can. We're back tomorrow talking about the three big questions surrounding Amazon.