OpenAI’s $102B Advertising Forecast, Nebius’ AI Acquisition Talks, Meta’s New AI Model

0:13Gil Luria:Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Thursday, April 9th. We've got a great show lined up for you today. We are kicking things off with exclusive reporting around OpenAI's grand expectations for its ads business. Our OpenAI and Anthropic reporter, Sri Mupiti, will share with us what she knows. We also have an M &A scoop about Nebius and a company that it is looking to buy. We'll bring on our deals reporter to talk about that. Next up, ServiceNow is wading deeper into the customer data wars. My colleague spoke with their chief operating officer, and I'll be chatting with her about what she learned.

0:51Gil Luria:And we will wrap the show with analysis about Meta's big news this week. A new model yesterday and a giant deal with CoreWeave today. It's going to be a fun show, so let's get right on into it. OpenAI has big expectations for its ads business, and my colleague Shreemupiti got a hold of the company's internal projections for that product. I want to bring her on to talk all about it. Shreemupiti, welcome back to the show. It's great to have you here. Excited to be here. So you got a hold of OpenAI's internal projections. What did you find? Yes, exactly. We found that OpenAI hiked its projections for its advertising business.

1:27The company expects to generate roughly$2 billion from advertising this year, but 50 exit by the end of the decade in 2030 to about$102 billion. That's a huge uptick. And so that's not just from only increasing the number of ads that it shows to users and the number of users that grow, but also the average revenue per user for this ad business is also expected to grow. And so that number, for example, will go from about$2 per year this year, the average revenue per user, and then grow up to$60 by the end of the decade as well.

2:01Gil Luria:$60 per user is what they're forecasting. Do we have any sense for how that compares at all in the industry? So Meta, for instance, has an ARPU of roughly$57. dollars and so it puts it in line with meta but it's much higher than other companies like pinterest or snapchat and so uh we will see if openai can also achieve this uh they had actually previously forecast last year in its first quarter 2025 projections that the arpu would only go up to 15 in 2030. so uh quadrupling it to 60 is that means that they have uh increased their expectations for how large this ad business can actually grow. And we see in the financials that the ad business will actually be the largest revenue driver for the entire company by the end of the decade.

2:48Gil Luria:Okay, so what's changed about the product? Is it that much better now? So OpenAI has since piloted ads to companies. What they've done essentially is that they had a small pilot where they charge about$60 per thousand views. They had advertisers pay about$200 ,000 to be able to participate. But now this month, there's an expectation that OpenAI is expanding that to folks beyond. And so as part of that pilot, they actually hit about$100 million in annualized revenue. And so I think these forecasts were actually made prior to the launch of the pilot, but they were sort of noodling on it. And so it's just sort of an emphasis is that how OpenAI expects that this ad business will be a huge revenue driver by the end of the decade.

3:38Gil Luria:Now, you mentioned that$100 million figure. That's about where they were late last month, we reported. Do you think that they will get to$2.4 billion this year? I mean, what's your hunch here? I think it's a big push, but especially given that they sort of stalled at about 900 million weekly active users since the start of the year. I don't have a recent update on where they actually are, but my understanding is that they haven't yet reached 1 billion weekly active users, which was their actual goal for end of last year. And so I think with the number of users stalling, perhaps there is another way to make up for that by having a deeper level of engagement for the users that are staying on and to perhaps that's a way that they still expect to hit 2.4 billion or so this year from ads but i think that time will tell on if they're able to actually pull it off especially since they're still working on a lot of the kinks of putting together this ad business and so right now for example they're partnering with other companies like ad tech firms like trade desk or but eventually i'm my expectation is that the company will try to own more of that process as well.

4:53Gil Luria:Well, I mean, and this is kind of interesting too, right? Because if they're saying that we are going to get all this ad revenue quickly this year, the ramp is going to be rapid. I sort of think about the extent to which the product would change from the user's perspective. And if they might make changes that are not so minuscule, and in fact, might be very significant, in turn, affecting the customer experience. and I don't know. It's just a concern that I have, you know, would they sort of ruin the product in some ways by putting so many ads in an effort to reach this$2.4 billion goal? And the other thing I'm thinking about, Sri, I wonder if you talk to people about this, the ARPU being so much higher than other tech companies, fine, Meta's right in there.

5:39Gil Luria:I mean, what does that say about the advertising product? It's supposed to be like a more premium product, although maybe they're saying we're going to make more money from it. Just help me frame that for us. Yeah, I think that the positioning is like a more premium product just because ChatGPT, for example, has such a deep level engagement with users. It might know, for example, not only where they want to eat, but like a deeper sort of ambitions or goals of the user because they have long type of conversations in a way that's different from like if you just scroll past Facebook. So I think the positioning is to be a more premium offering and that puts it in the sense of rivals like NFL video, but actually even more than meta in terms of how much it costs advertisers to advertise on ChatGPT.

6:24The thing I will say is that it's not just the pilot so far, the 100 million annualized. Of course, it's annualized revenue, not actual revenue, and it's still a pilot. The pilot itself was only for about 20 % of logged-in US users for free ChatGPT as well as the ChatGPT Go tier, which is about$8 for US users. But they're now expanding it beyond to Australia, Canada, and New Zealand based on the success of the pilot or so. And so I think that to be able to hit 2 billion or so, it's just a matter of being able to expand it further and further to the potential of the user base. So there is maybe a way to do that.

7:05I think though advertisers themselves and execs at ad tech firms and ad agencies have actually been somewhat disappointed of the ROI of ChatGPD ads so far, just because they haven't been able to actually see a measurable business outcome. But of course, it's still early, so perhaps things will change.

7:24Gil Luria:Well, and we've also reported that other companies like Amazon, who also have their own giant ad tech businesses already, I mean, Amazon has started to look at what it would mean to implement ads not directly in the Rufus chatbot that it has, but inserting sponsored advertisements, incorporating Rufus around its shopping pages. And what we've reported is that Amazon would actually be offering more data to its advertisers compared to what OpenAI is offering. And so I don't know, what this sort of feels like to me is OpenAI had this slate of dozens of different products that it was piloting. It picked one of them and now it's saying, hey, we just want to take this all the way, but it's not going to be a phased approach.

8:12Gil Luria:It's just going to be sell, sell, sell. And I don't know how that's going to turn out. Yeah, exactly. We'll see, especially since post the side quest getting squashed, I think they're putting all their eggs now on odds. Consumer business is also, as well as like their codex and enterprise business. And so this is an avenue that we'll hopefully get more data on by the end of this year. Great. Well, Shree, I want to thank you for coming on. That is Srimapiti, our OpenAI and Anthropic reporter here at The Information. Neocloud company Nebius is in talks for a major acquisition. The information's deals reporter Valida Pau and our deputy bureau chief of venture capital, Katie Roof, broke the news with details on who Nebius is looking to buy.

8:55Gil Luria:Valida joins me now to share with us what she knows. Valida, welcome back to the show. Hello. Valida, who is Nebius looking to buy? So AI21, it's an Israeli startup that was founded like eight years ago in 2017 by three serial entrepreneurs in Israel. And one of them is still operating as the CEO of like Mobileye, which is a self-driving car technology public company. And they make kind of large language models and they also make customs AI systems for, you know, businesses. And it was one of the six startups that the information highlighted back in 2023 that OpenAI or Google should be watching.

9:37So it's a$1.4 billion business that was last valued in 2023.

9:44Gil Luria:Okay. And so AI21, which I have to remind myself every time that it is AI21 because sometimes I read it as A121, which is not what the name of the company is. I also make some mistakes. What does the company do? Oh, they make large language models, and then they also make custom AI applications for enterprises. Like, you know, they have a flagship product, Maestro, and they also can help, like, you know, businesses create specialized large language models or AI applications for, you know, industries in finance, in healthcare, et cetera. Okay. And so why would Nebius, a company that is in the NeoCloud business, you know, they're in the business of renting out GPUs, essentially, why would they be interested in a software company?

10:32Well, NetBS is actually more than just a NeoCloud. You're renting out servers to your AI customers. You already do other types of businesses. But this is something that cloud providers have always wanted to do, and then they're trying to build out more of their software capabilities and to diversify their business rather than just so they can build more AI tools for their customers or developers that are already using that platforms. So what is special about AI21 is that they have these enterprise-ready custom AI solutions they can provide for AI developers. So this could fit into what NetBS wants to do to sell more of their toolings and also their hardware.

11:16So these developers are using their platforms already. Okay.

11:21Gil Luria:Now, this kind of reminds me, we've seen other NeoCloud companies like CoreWeave, I mean, they have also acquired software companies. And so I take your point, this is kind of a push that these neocloud companies are making to expand the stage of the value chain that they're playing in. Has Nebius made other acquisitions along this line in the past at all? Well, they recently also acquired like another startup that do similar things, not similar things like it's got teboli and then um so this is not like something they have not done in the past but just like takes it a bit more like upstage for them right and and we should say that ai21 i mean it's been reported that uh they were in talks to have sold themselves to nvidia that was a possible buyer as well that obviously fell apart so the idea that ai21 fits inside a larger company seems to be something that a lot of people are seeing.

12:22Gil Luria:This is a fun one. I just googled this. I'm going to ask you, do you know what the 21 in AI 21 is? I don't know. Okay, according to the AI overview, it's 21st century, which makes a lot more sense to me now. And also why I will never call it A121 again. So anyway, Valita, the last question I have for you is what does this tell us about the the deal landscape at large i mean it seems like deals are are heating up and do you expect that to increase going forward i feel like deals um tech deals especially in mna side has been kind of quiet recently so this is kind of like a sign that like you know some mna is going on but particularly on the neocloss side i felt like you know since last year as they are trying to grow bigger and then grow their you know toolings across like the full stack.

13:11CoreWeave has done like similar purchase buying like weights and biases last year before the IPO'd. And you can see like Crusoe and then together AI also did like some kind of like smaller purchases on like grabbing up these AI startups. And of course, like, you know, these neoclouds, they also want like talents, which, you know, these startups can provide for them. Like, you know, people working at the frontier of AI. So this is the talent angle in AI related M &A is still a big point. that why people are wanting to go after these hot startups. Great.

13:44Gil Luria:Well, Valida, I want to thank you for coming on. That is Valida Pau, our deals reporter here at The Information. Thank you. ServiceNow has a new AI agents tool out today that further builds out its answer to the long-winding story of customer data wars. My colleague, Laura Bratton, wrote about that in our Applied AI newsletter today. I want to bring her on to talk all about it. Laura, welcome back to the show. It's great to have you here. Hey, Kosh. What is this new agent that ServiceNow is unveiling today? So it's less of an agent and more of a product. I kind of struggle to describe it, but it basically pulls in data across customers' different applications and ServiceNow and brings it into one place so that outside AI agents, when they do access ServiceNow, kind of can go to this one place where all their data is updated in real time and really easily readable.

14:41But ServiceNow also made it really clear that it's not going to be an extra charge for customers if they're using agents inside of ServiceNow to use the context engine to reason on their data.

14:53Gil Luria:And tell me about the pricing model that ServiceNow has developed for this program. So they haven't really announced a pricing model for it yet. They just made it clear that it's going to be consumption-based in some capacity. So when outside agents built, you know, with other cloud providers or LLMs or accessing customer service now data with the context engine in particular, they're going to track and meter it and charge based on consumption. So this is kind of interesting because in the past, when we think about the customer data wars, we have talked about the companies operating the agents like Glean and Atlassian.

15:36Gil Luria:It's very much been a story of them against the companies that store the data, companies like Salesforce. And so the idea here that a company like Salesforce would put up a wall to block agents from accessing it, I mean, that was the debate. hey, this is kind of interesting because ServiceNow, in this case, is both the agent company and it's also the company that has the data. The thing that I'm thinking about, though, Laura, is that, I mean, it's kind of a convenient way then to milk more out of the customers, right? Because it's saying, here's a new feature. We have all of what you need, but you're going to have to pay for it.

16:15Yeah, that's an interesting point. And, I mean, I asked them this. I asked their COO, Zabry, about this. And he said, you know, customers can still use standard application program interfaces or APIs to access their data in each individual app. And what they're doing is basically just coming up with a way to bring all the data into one place. So it's easier for an AI agent to read, but then that's going to come at an extra cost if you're using an AI agent that's not built within ServiceNow. But I think it is part of this hot debate where it's like, okay, who owns the data? You know, is it the customers or is it, you know, the platform?

16:54And obviously customers are going to argue it's their own data and argue against any gating. But I guess ServiceNow's argument here is that they're not actually gating anything. They're just putting out, you know, making data more convenient for agents inside their platform to read.

17:08Gil Luria:In other words, if I understand you correctly, ServiceNow, I mean, they have the data. And so customers could argue that, hey, I'm already paying to store my data with you. why do I have to pay more to access it? So it's actually not so much of a different argument than what customers have made otherwise. Yeah, that's true. That's definitely true. Yeah, and it's part of this sort of argument where investors want to see that enterprise software companies are finding out, you know, are finding a way to monetize their trove of historical data because that's really the competitive mode that they have when, you know, AI agents are kind of threatening to undermine the value of their user interfaces if an AI agent can just tap these, you know, legacy systems and access all of customer data.

17:57Why do, you know, companies need to pay for so many seats with seat-based models? And that's why you've seen consumption-based models. But I think, you know, this is just part of the story of enterprise software companies really reckoning with how they come up with new pricing models and in this age where customers might be using AI agents rather than having software engineers locking into software systems in typical ways.

18:24Gil Luria:Do you think this sets any kind of a precedent for other companies that might unveil agents like this or for the legacy storage companies? No, not storage companies, but the companies sitting on the data, housing the data. I mean, does it set a precedent here for how that will go? Sources I've talked to across the industry have said they expect more forms of new products for outside agents accessing legacy software systems, figuring out how to monetize that. They think that there's more that's going to be coming or that more enterprise software firms are going to start gating when AI agents try to access data in their platforms.

19:09But other people I talked to were like, customers are never going to allow that. They're just going to stop using those software systems. So I think it's really unfolding in real time, and it's a new landscape, and everyone's just trying to figure out what's going to work. Great.

19:24Gil Luria:Well, Laura, I want to thank you for coming on. That is Laura Bratton, author of our Applied AI newsletter, here at The Information. Meta is committing an additional$21 billion to CoreWeave on top of a previous multibillion-dollar arrangement for AI cloud capacity. But we also saw this week that Meta announced a new family of AI models. It is the first release from Meta Super Intelligence Labs division. Shares jumped on that news. Here to break down all of that is Gil Luria, Managing Director and Head of Technology Research at DA Davidson. Gil, welcome back to the show. It's great to have you here.

19:59Gil Luria:Thanks for having me. So Gil, we had you on a couple weeks or months ago talking about when Meta was going to release their new family of models, how high the expectations would be. They poured all this money into it. And we saw that shares jumped yesterday. And so I'm wondering, I mean, what? Is this model amazing? Did they live up to all the expectations? The model is okay. We took a look at it. We used it as consumers. We tried to see how it works within more of an agentic context. And it's okay. It puts a meta in the mix. So if Anthropic right now has the lead, OpenAI and Google DeepMind will probably have similar versions of the most advanced models out soon.

20:45I would put Meta in the mix with your Grox and your Chinese competitors, which is an improvement. Because as of a week ago, they weren't really even in the mix. So expectations from Meta were really low because they've spent so much money on this. Famously, they paid$14 billion to hire Alexander Wang. They made nine-figure offers to multiple people. Some they hired, and some decided to stay at OpenAI and Anthropic, which is a little insulting. And then, since then, subsequently, we've been hearing out of Meta conversations about people leaving, multiple restructurings, issues with the culture.

21:31So with that as a backdrop, expectations were very low. And so the fact that they even have a bottle that puts them back in the mix is a big relief. And that's what you're seeing in the shares.

21:43Gil Luria:Yeah, well, I think low expectations is definitely a good way to put it. You know, the question that I've been thinking about with this model release and also putting that in the context of our story this morning that OpenAI has big forecasts for its advertising business. The main question is, you know, is it harder to build a great model or a great advertising business? And, you know, Meta obviously has nailed the advertising business. Can it figure out the model? Not sure. OpenAI similarly has a good model. Can it figure out the advertising? Which one do you think is harder? The model is harder.

22:22There's a lot of companies doing the advertising business very successfully. None as successfully as Meta, by the way. and Google, but that's a business model that's established. People understand how to do it. OpenAI is just being careful because they don't want to lose our trust. I think the big flashing warning for them were those ads that Anthropic ran at the Super Bowl showing how wrong it could go if you do ads the wrong way within a chat. So OpenAI has to be careful, but they clearly have the wherewithal of how to do ads. And just by piloting, they already have$100 million business there, just by messing around with it.

23:03Having the best frontier model, now that is really hard because there's literally 100, 150 people in the world that are capable of doing AI research at the high level. That's why they're all making millions of dollars, tens of millions, sometimes more, is because there's so few people that can wrap their head around where we're at on AI and advance AI research. Now, those are very concentrated in those three or four places, right? They're at OpenAI, they're at Anthropic, they're at Google DeepMind, and Meta has spent a lot of money hiring them. And then the remainder are in China. But that's how hard this is.

23:45Where advertising is spread around, even digital advertising is spread around several companies. We have decades of experience on that. OpenAI can and will ramp a very good ad business. It will be somewhat at Meta's expense, probably a little more at Google's expense. That's why there were so many concerns about Google last year. Those have mostly gone away because Google has made its own transition to AI well. But when OpenAI ramps up that digital advertising business, it's going to come out of Google and Meta's pockets mostly.

24:22Gil Luria:but let me ask you a question let me push back a little bit on look the models are difficult i mean i i'm not a researcher these researchers are geniuses this is not a shot at them but at the same time there is also a dialogue that the models will be commoditized i mean right now we're in this race between who has the best model in the long run three years from now maybe it won't actually matter and everybody every company will have the model and so the question i have is is the model really more difficult? I mean, isn't it building an ecosystem of advertising products that gives advertisers what they want?

24:57Gil Luria:You know, that's not something that is a slam dunk right away. Well, so the first answer is, as is with a lot of things these days, we don't know. The rate of progress around AI is so unbelievably fast, so much faster than anything we've ever looked at or dealt with in human history that we don't know so this notion that it could get commoditized yeah it's possible but i'd argue that the developments we had this week point to the exact opposite direction so uh anthropic introducing the mythos or mythos model i'm not sure how they pronounce it but yeah good point we're talking about the first model that may not get a full public release we may be at the inflection point where the models are so powerful that they're too powerful to just give people unfettered access.