Dan McCarthy | University of Maryland - The Unit Economics of AI - Can LLMs Actually Make Money?

Summary

CMO Confidential Podcast – Episode Summary

Episode Title: The Unit Economics of AI – Can LLMs Actually Make Money?
Host: Mike Linton
Guest: Dr. Dan McCarthy, Professor at University of Maryland
Date: April 21, 2026

---

Episode Overview

This episode dives into the financial realities behind large language model (LLM) companies like OpenAI, Anthropic, and Google’s Gemini. Host Mike Linton and guest Dr. Dan McCarthy examine whether these artificially intelligent marvels are truly profitable—or just extremely well-capitalized. Their discussion is anchored in Dr. McCarthy’s expertise on customer lifetime value (CLV), applying it to shed light on the hype, economics, and sustainability of LLMs in the context of sky-high valuations.

---

Key Discussion Points & Insights

1. The Importance of Customer Lifetime Value (CLV)

• Definition: CLV is the overall profitability a customer represents to a company over their lifetime. Not all customers are equally valuable; a small segment often generates the bulk of profits.
  • Quote: “There’s a lot of customers that aren’t worth very much, and there’s a few customers that are worth a lot.” (Dan, 03:23)
• Customer-Based Corporate Valuation (CBCV):
  • Adds nuance by focusing on incremental profitability from acquiring a new customer, after variable costs.
  • Enables understanding not just the quality (value) but the quantity (number) of profitable customers needed to justify a business’s value.

2. LLMs – Free vs. Paid User Economics

Timestamps: 09:22–14:33

• Freemium Model Dynamics:
  • Free users on LLMs (e.g., ChatGPT, Claude) are expensive due to high inference (compute) costs. Unlike traditional SaaS, every query run by a free user incurs real, scalable expense.
  • Quote: “The free users cost these businesses a lot of money...every time you write that chat, it’s actually a real expense for the firm.” (Dan, 09:48)
• Monetization Challenge:
  • Conversion from free to paid is slipping—fell from ~6% in late 2023 to below 4% by late 2025, despite user bases exploding from 100M to 900M.
  • LLM companies must decide: make the free tier worse, improve paid features, or monetize free users via ads.
  • Quote: “Maybe is the free plan too good? People want to hang out on the free plan, and they don’t need to upgrade.” (Dan, 12:53)

3. Unit Economics & Profitability of Paid Users

Timestamps: 16:47–18:44

• Paid Tiers Are Highly Profitable:
  • $20/month subscribers generate ~70% variable margin; margins continue to grow as model efficiency improves.
  • Customer acquisition cost (CAC) is about $30.
  • Quote: “The margin is actually quite good for those people. It’s about 70% and it’s way up.” (Dan, 17:31)
• Retention is "Best-in-Class":
  • Subscribers rarely churn, aided by the tool’s stickiness in workflows and professional life.
  • Multi-homing: Like streaming, many users pay for multiple LLMs, driven by specialized use cases.

4. The Real Profitability Challenge—Training & R&D Costs

Timestamps: 23:04–27:37

• Massive Ongoing Expenses:
  • Unlike food in a restaurant, these companies face colossal, recurring expenses to train new, more powerful models—what Dan analogizes to “the cost of the chef.”
  • Training & R&D has gone from $5B (2024) to $15B (2025), projected at $30B (2026), and possibly reaching $60B by 2027.
  • Quote: “They spent about $15 billion on training in R&D in 2025. ...projecting that those costs are going to go to $30 billion next year.” (Dan, 26:40)
• No Clear Ceiling on Spend:
  • It’s unknown at what point investment in new models will plateau, making long-term profitability and valuations difficult to predict.
  • Quote: “It’s really hard to know when these companies will finally back off on the training and R & D expenses.” (Dan, 28:02)

5. Industry Competition and the Role of Subsidization

Timestamps: 29:34–31:39

• Google’s Gemini:
  • Subsidized by search ad profits—can afford to undercharge and under-monetize the product (Gemini monetizes much less than ChatGPT or Claude) to retain users and protect its ecosystem.
• Competitive Dynamic:
  • Non-Google LLMs (OpenAI, Anthropic) can't “make the free tier shittier” without losing users to Gemini, who has more leeway due to its broader business model.
  • Quote: “If the free tier really sucks compared to Gemini, then Gemini will just take it all.” (Dan, 31:39)

6. Marketer Advice: How to Use & Future-Proof Your Work

Timestamps: 32:07–35:03

• Diversify Providers:
  • Relying on a single LLM vendor is risky; outages or business collapses (with high R&D burn and uncertain viability) could jeopardize your workflows.
  • “Diversification of your providers is probably prudent...imagine if you only relied on them for kind of mission critical things.” (Dan, 32:28)
• Continuously Test Multiple Services:
  • Models evolve rapidly—what works best can change monthly or weekly.
• Treat Experimentation as Education:
  • Keeping skills current and understanding outputs across platforms is crucial for marketers and organizations.

7. The Crucial Role of AI Literacy and Play

Timestamps: 35:03–39:59

• Hands-On Experience is Key:

  • Institutions (and individuals) need to make time for direct interaction. Success is higher with "foundational knowledge" in the subject area—AI amplifies human expertise rather than replaces it.
  • Quote: “People who had very strong foundational knowledge in the area...were getting the most out of it, because they knew all of the right questions to ask.” (Dan, 38:38)

• Keep it Fun:

  • Engage with AI playfully to make adoption less intimidating and improve learning.
  • “LLM use...can be really fun.” (Mike, 37:47)

8. Practical Takeaways & Final Tips

Timestamps: 40:20–41:45

• Talk to Your Phone:
  • Use voice, not just typing, when prompting LLMs. Providing richer context by speaking leads to better results.
  • Quote: “Talk to your phone...the LLMs crave context...you end up with this humongous amount of words...and the results are just so much better.” (Dan, 40:30)
• Treat LLMs like business collaborators:
  • “Talk to it like a coworker or an agency or consultant. Don’t talk to it like a piece of technology.” (Mike, 41:36)

---

Notable Quotes

• On CLV and Core Economics:
  • "When you roll it all together, that gives you a sense of, when this company acquires its customers, are they actually generating a return?" (Dan, 03:32)
• On the Challenge of Scaling Freemium LLMs:
  • "The free users cost these businesses a lot of money. And that's one way where this is pretty different actually from like a pure software business." (Dan, 09:48)
• On Profitable Paid Tiers & High Retention:
  • "The margin is actually quite good for those people. It’s about 70% and it’s way up." (Dan, 17:31)
• On Eye-Watering Investment in R&D:
  • “ChatGPT had spent $5 billion on training and R&D when they generated about $4 billion in revenue. In 2025, about $15 billion, projected $30 billion next year.” (Dan, 26:41)
• On Defensive Pricing by Google Gemini:
  • "Gemini monetizes about 30 times worse than Claude...they’re doing that intentionally to keep people coming back to them." (Dan, 30:45)
• On AI Literacy in Practice:
  • "People who had very strong foundational knowledge...were getting the most out of it because they knew all the right questions to ask." (Dan, 38:38)
• On Practical Prompting:
  • “Talk to your phone. ...If you want to get a really good answer, what they crave is context.” (Dan, 40:30)

---

Key Timestamps for Reference

• 03:16 – Dr. McCarthy explains CLV & CBCV.
• 09:39 – Start of discussion on LLMs’ free vs. paid economics.
• 12:31 – Insights on conversion rates for paid plans.
• 16:47 – Unit economics & paid subscriber profitability.
• 23:04 – Limitations to profitability: training and R&D costs.
• 26:41 – Concrete numbers: LLM training and R&D expenditures.
• 28:02 – Challenges in “justifying” LLM business valuations.
• 32:07 – Marketer recommendations: diversify and continuously test.
• 35:03 – AI education, hands-on learning, and the value of play.
• 40:20 – Dan’s practical tip: talk to your phone; richer context yields better AI results.

---

Actionable Takeaways for Marketers & Decision-Makers

• Diversify LLM providers to mitigate risk and avoid vendor lock-in.
• Regularly test multiple models; provider performance and output quality can change rapidly.
• Encourage teams to experiment and “play” with LLMs, reducing learning friction.
• Invest in foundational knowledge: Deep understanding of your field significantly enhances AI’s usefulness.
• Provide context: When prompting, use voice or elaborate descriptions for better results.
• Monitor the competitive landscape and vendor business health—LLM ecosystem is volatile, and today’s leader might not last.

---

Memorable Moments

• Freemium Pitfalls: “Maybe is the free plan too good? People want to hang out on the free plan, and they don’t need to upgrade.” (Dan, 12:53)
• R&D Costs Reality Check: “$5B to $15B to $30B... potentially up to $60B in 2027.” (Dan, 26:41)
• Fun with Caricatures: “Oh, the caricature thing is hilarious on ChatGPT, if you want to make fun of your friends...” (Mike, 38:07)

---

Conclusion

Dr. McCarthy and Mike Linton reveal the economic engine (and burn rate bonfire) beneath the LLM hype, equipping marketers and business leaders with a pragmatic framework—CLV—to make sense of dizzying valuations, product strategies, and evolving risks. Their closing mantra is clear: diversify, experiment, and keep learning, because in AI, lasting advantage favors those who play and adapt.

---

Transcript

[A]

The CMO Confidential Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit iheareverything.com welcome to CMO Confidential,

[B]

the podcast that takes you inside the drama, decisions and choices that go with being the Head of marketing. Hosted by five time CMO Mike Linton

[C]

Typeface has completely changed the process of large campaign executions. Everyone knows AI can help with images and headlines, but the real impact is making the leap from a creative brief to a live, multi channel large campaign at speed. Typeface just announced its marketing orchestration engine, the first platform built to automate campaign workflows. You can roll out campaigns that used to take months in just a few hours. TypeFace uses Agentic AI to orchestrate the entire process, taking one campaign and instantly orchestrating it into thousands of personalized experiences across ads, email and video. Major brands like Asics and Post holdings are already transforming their marketing with Typeface. See how to move from brief to personalized campaigns in hours, not months at typeface AI/cmo welcome marketers, advertisers and those who love them to Chief Marketing Officer Confidential. CMO Confidential is a program that takes you inside the drama, the decisions and the politics that go with being the head of marketing at any company in what is one of the most scrutinized jobs in the executive suite. I'm Mike Linton, the former Chief Marketing Officer of Best Buyer, eBay, Farmers Insurance and Ancestry.com I'm here today with my guest, Dr. Dan McCarthy. Today's topic the Unit Economics of AI. Can large language model Companies actually make Money? We know they have high valuations, but today we're going to talk about how they make money through a customer lifetime value lens. Now Dan is a professor at the University of Maryland. Previously he was an Assistant Marketing professor at Emory where he created what may be the first ever customer lifetime value course. His specialty, and you want to take notes on this, is the application of statistical methodology to contemporary marketing problems. He has been one of our most popular guests over the last few years discussing cltv, the business impact of privacy policies and and the power of loyalty programs. So it's great to have him back to discuss his latest work. Welcome back Dan.

[D]

We're going to make it a 4P.

[C]

There we go. That's right. So hey Dan, just assure everyone is starting from the same place. Give us a quick overview of customer lifetime value and customer Based corporate valuation and how they're used to evaluate businesses.

[D]

Yes. The customer lifetime value is this number that's kind of over every single person's head. That represents the overall profitability of that customer to the firm. As you'd imagine, there's a lot of customers that aren't worth very much, and there's a few customers that are worth a lot. And that could be really, really powerful to know. And when you kind of roll it all together, that gives you a sense of when this company acquires its customers, are they actually generating a return? A lot of the nuance comes in, like what costs do we include? That's where I think CLV is kind of extended to what we call customer based corporate valuation, or cbcv. That typically what CLV is supposed to represent is when I bring in that next marginal customer, what is the incremental profitability that I'm going to get from them? And so I've got all my accountants, I've got all the lawyers, I've got the headquarters, I'm paying the CEO, Those are all relatively fixed. But when I bring in that next customer, what is the additional incremental profitability that customer will throw off? I need to know how much I'm spending to bring them in. I need to know the revenue I'm going to get. And then I need to subtract out all of those variable costs that are associated with the revenues that I bring in.

[C]

And one of the keys here is that all customers are not even close to equal and understanding which ones are worth more. Say you're an airline and you have someone that flies 200 times a year versus someone that applies 10 times a year. Knowing that and knowing the descriptor really makes a difference. And it should make a difference in how you market and, and also the valuation of the company, right?

[D]

Yeah. And instead of thinking of kind of like your typical customer or I've got my ideal customer profile, there's these Personas and we have like the. It's this type of person that I want to acquire, you know, that they're my best customer. Typically what you find is that the customer base from a value perspective is like a barbell, that 80% of the customers are worth very little, nothing, or maybe even costing you money. And then you're making all of your money on like the top five, 10% of the customers. And so knowing that, it's like, well, what's so different about them? You know, what are they, what do they like? You know, what's bringing them in? The Door, you know, where, where are they geographically located? You know, what was it, what was the acquisition channel that I acquired them through? You know, suddenly it's like, oh, that, that, that stuff that I really want to know. But yeah, I think the, the key in a case like these AI companies is that there's customer lifetime value, which is like the quality of the customers that you bring in. But there's not, at least with CLV on its own, there's no notion of quantity. You know, how many am I going to bring in? And what is the kind of like the cost of the running of the customer acquisition machine? You know, so every period I need to spend a certain amount of money to pay all of my overhead expenses, whatever those might be. And, you know, ultimately that really matters. Like, if it's really, really big, then you need to earn a much higher rate of return on your customer for the whole business to eventually be able to grow its way into profitability. And so, yeah, so that's, that's kind of when we, when we talk about the LLMs in particular, you both need to think about lifetime value, but then you also need to think about, you know, well, how much do they have to spend to kind of keep the whole thing going and how many, Especially

[C]

when you have a valuation of, you know, hundreds of billions of dollars on something where there's only several years worth of data. Is that what made you want to dig into this analytically or what, what made you want to go look at this through a CLTV lens?

[D]

It's really, it's kind of two things. One is, I think that when you're thinking about it from the perspective of is this business going to justify its valuation? The CLV lens is the most valuable when you have a business that's losing money but growing quickly. Because then the big question is, are they just buying growth and they won't ever be able to grow their way into profitability, or is it a business that's actually doing very well, like they're acquiring customers profitably, but they're not yet overall profitable yet because they're still scaling. And that's an incredibly important distinction. And some companies that we run the numbers on, we find that they're good. Some companies, we find that they're bad. And so for one, it's a really nice industry where this sort of analysis is specifically suited to it. But then I think the second thing is I am like the, a voracious user of all of these services. And yeah, you're like a tire. So it's like I want to know, you know, should I be worried here? You know, is this company going to have, does it have a viable business model? Because, you know, if it doesn't, then it's going to have real implications for me because either the price is going to need to go way up and then it's like, oh, well, that's kind of a bummer. Or you know, in the worst possible scenario, you know, is this company going to crash and burn and I should really make sure that I'm splitting my usage across multiple services?

[C]

So yes, because if I get dependent on one thing and I build it into my, say, entire Martech stack and all my work and then it disappears, that would be a big problem. Right.

[D]

I think a lot of companies should be thinking about that, you know, that they, they're going to need to know because a lot of companies are, and a lot of people within those companies are building, you know, workflow after workflow around certain things that they do. And, and if this, you know, if the company is going to run into major issues, then they're going to need to, to think of contingency planning.

[C]

Got it. So let's talk about the work you did with, because you looked at all the major AI companies, right. And, and tell us what you, you learned for products like Chat, GBT and Anthropic and everything else.

[D]

Yeah, the big thing, I'd say one of the big learnings was that among the paying subscribers, the economics are actually pretty good. So the way that I would kind of conceptualize these businesses is that they have free plans and the free users cost these businesses a lot of money. And that's one way where this is pretty different actually from like a pure software business, because a software business like Slack, there are costs associated with their free tier, but software is cheap. And so, you know, the marginal cost is going to be very small. But here, every time you write that chat, to create that Annie Mae picture of your family, it's actually a real expense for the firm. And so these free tiers are actually quite expensive and I would consider them to be kind of like the on ramp to the paid plans.

[C]

And when you look at expense you're talking about, I mean, I have to have everything working and then I'm using up huge amounts of energy, etc. I have a lot of money I'm spending just to be on. Right? Is that what you're thinking?

[D]

Well, not only that, but even just the cost of running the queries every time a query is run, there is what's called kind of inference costs. And inference costs are one of the biggest expenses right now of the LLMs. And so, yeah, so those are purely variable. You hit, you know, you type in.

[C]

That's an example of how inference costs work. Like, just take us through one where I ask about something about the University of Maryland.

[D]

Say, yeah, all of the tokens that are used. So there's tokens that are, you know, you're asking the question, and that's basically, they pay per word. And then everything that you might have had in context, you know, like, keep this document or keep these images, keep this paper as context to help, you know, to help answer this question that I'm asking you. All of that. That corpus of data basically represents a cost to OpenAI or to Claude or what have you, and then it's going to spit out its answer. And the length of the answer, again, represents kind of token usage. So you can think of the direct costs associated with inference as being a function of the number of tokens that are used by the service.

[C]

Got it. And so I have this free thing, and then I obviously want people to trade up to buy stuff. Tell us what you're learning on the acquisition and then the retention and the customer behavior, seeing across your cohorts.

[D]

I think a lot of people find the free tiers to be quite attractive. And I think one of the issues that OpenAI has been facing is that there's like all these users and then there's some percent of them that actually bite the bullet and say, I'm going to pay 20amonth or 200amonth. And that percentage back at the end of 2023, it was almost call it 6%, and then it went to 5%. And right now, late 2025, it's below 4%. So they've had a really big increase in the number of users. It's gone from like 100 million to like 900 million over that period of time. But the number of paid subscribers has not gone up to nearly the same degree. And so, so there's this question, like, maybe is the free plan too good? You know, like, people want to hang out on the free plan, and, you know, they don't. They don't need to upgrade. So again, there's this whole question that a lot of companies face that have freemium models of how ultimately you want people, you want to make money. Right.

[C]

We had the same thing in Ancestry. Yeah.

[D]

Yeah. So it's like, well, what could we do? Either we can try and lose less money on the free people. Or we can make the free tier worse. We can make the paid plan better. Yes. And then there's like a whole bunch of other things that we can kind of inject in certain parts of the process that really, you know, will make it the most likely for people to want to take the jump. And so again, when we think about ads, you know, everyone's talking about ads right now. Ads would be one way that they can both simultaneously lose less money on the free plan and they can create more of a gap between the free plan and the paid plan.

[C]

Right. Because I can take ads out.

[D]

Yeah. It's like, this is kind of annoying. I don't want to be exposed to this. So it kind of kills two birds with one stone. But, yeah, I think there's been a lot of articles written about how OpenAI in particular their gross margin. Actually, this was true of both OpenAI and anthropic. Their gross margins were a lot lower than they even were saying it was going to be. You know, it's going to be. It was 40%. Now it's down to 33%. They're like, why the heck is this getting so much worse? And part of the reason why is because they have all of these free users that are just consuming so much cost. And. And so I think the ad plan is a way for them to help kind of defray some of that expense associated with a free plan.

[B]

We are taking a short break from this show for a word from our sponsor, Typeface. To orchestrate workflows across channels. Meet ARC agents, your AI teammates handling complexity so you can focus on what matters all within a reimagined workspace where you and AI create as one spaces. Works the way you do. Create anything from documents to videos to entire campaigns at scale. Our agents help you craft campaign storyboards powered by your brand hub, giving you the perfect starting point. Transform one asset into countless variations for every audience and market. Join industry leaders already reimagining their content lifecycle with typeface. Welcome to marketing's next chapter, where every story finds its voice. Now back to our discussion.

[C]

How does this play out? We're talking now about valuations. A lot of these companies have enormous valuations. Tell us, take us through kind of the unit economics, where they end up and how people should be thinking about it.

[D]

Yeah, so the one. You know, the big question to me comes down to cac.

[C]

Customer acquisition cost, everybody.

[D]

Customer acquisition cost. The value of customers after they've been acquired. And then just how many can you bring in the door? And what happens with your overhead. And so I think the CLB perspective, it's actually pretty optimistic that the paid subscribers are actually pretty profitable even when you factor in pretty generously, you know, the cost of the free tier.

[C]

I mean, even at 20 bucks, just 20 bucks a month, I'm still profitable, even given massive token usage because, yeah,

[D]

the margin is actually quite good for those people. Okay, it's about 70% and it's way up. So obviously in a lot of other businesses, that margin, it doesn't change very much. Like if you're selling food, you know, and you're twice as big, you're probably still still paying the same amount on food. Here. The efficiency of the models has been improving and improving and improving. And so that margin has been going like this. So if they can get people to paid, they're getting a nice kind of 70% ish variable margin on those people. Now they're not only getting 20amonth, for some people, they're paying 200amonth. And then there's the API usage as well. And the customer acquisition cost ends up being around. I was estimated to be around 30 bucks.

[C]

Okay.

[D]

And retention is amazing.

[C]

So credit cards, once you have this tool, you're probably not going to dump it, right? You're just. Even if you have more tools, like if I have clot and anthropic and I can use anything I want, right, Gemini?

[D]

Yeah, I'll put a big asterisk on this. So, you know, as you know, because I've talked about using this data before on your show, credit card panel data is a really nice data source for certain types of activity. And this is a great use for it because people, they kind of build their credit card for this. And so we can directly observe customer retention through the credit card panel data. And companies like OpenAI, they have best in class retention. The people, they just stay. Now, obviously, it's funny, I know we were almost going to do this show a little earlier. I feel like the passage of time has just made this more and more interesting. So just this past weekend we had this humongous kerfuffle where anthropic was deemed a supply chain risk. And then the, you know, it's like right after the Department of War shunted them to the side, it's like five minutes later, open A is like, yeah, I'll take the business. And now everyone is, you know, saying, I'm going to dump my ChatGPT subscription and I'm going to subscribe to Claude. So that is all. You know, I think that is trust is a very big issue with the LLMs because of the nature of the data that they're collecting. But holding all that aside, the base rate of retention is very high in this category. And as you're kind of alluding to people like me, if we find a lot of value in ChatGPT and then we find a lot of value in Claude, it's not like people will only sign up for one. There are a lot of people who do what's called multi homing, you know, where you kind of have multiple subscriptions

[C]

and well, it's no different than streaming services. Right. I mean I have, I have Netflix, I have Amazon Prime, I have HBO Max. It's not like I only have one here. You're saying that there, this could be really a big deal. And I hear almost beneath that is a lot of consumers are probably not going to drop ChatGPT if it's really functional for them.

[D]

You might really like Claude, but let's say you want to generate images. Well, sorry, you know, because Claude can't really, they don't really do images and I think it's smart of them. I think that image generation is extremely expensive.

[C]

Yeah.

[D]

And they've chosen to focus on these knowledge worker use cases and so it makes a lot of sense. But you know, if you wanted to do that sort of thing, well, you're kind of stuck. Also, if you wanted to do like super high end research where you want these models to think about a problem for, for an hour. Well, Claude is not amazing for that. Yeah, it's really good for questions. It'll give you an answer, you know, in a relatively short amount of time. But it won't, you know, it won't think for an hour like ChatGPT Pro 5.4 Extended Thinking Will. And so again, it's not that they aren't really good for what they do, but there are these, you know, kind of, there's like this Venn diagram of the use cases and if you, if you need certain use cases that Claude simply doesn't do, you're not going to just not do those things if they're a value to you. So, so yeah, I think that we're already empirically observing it. Like if, when you look to the data, you'll see a lot of people are using, you know, I'm some, I'm subscribed to probably, you know, four or five different paid plans right now just for different things. And, and I can't see how, you know, for at least three or four of them. I can't see how I would give them up. So.

[C]

Yeah. Yes. If you can't make caricatures with Claude, I mean, jeez, you got to have something. Yeah.

[D]

Where am I going to get my anime pictures? You know, that's right.

[C]

So you're saying all these look like pretty good businesses to you and it's just a matter of time before they have a path to big profitability, right?

[D]

Well, actually, no.

[C]

Okay, tell us what you're saying.

[D]

Yeah. So the clb. The CLB is not bad, but there's a question of quantity and there's a question of, of what they call these training costs. So inference cost is one of the big expenses. And that's again, you do a query, you got to pay that cost. That's like the cost of food for a, you know, for a meal kit company. But unlike the meal kit company, these companies are spending a lot of money on training costs and R and D. And those are, they're not a variable cost in the sense that, you know, if you bring in that next, you know, Claude subscriber, it's not like you're going to need to spend more on training up, you know, Opus 4.7 or something like that, but they need that next model. And in that sense, the tell us

[C]

how the training models work and then come back to this point, which is why they have to keep spending.

[D]

That's the R and D that brings them to the new models. So if you go, if we think about ChatGPT, there's call it 3.54, there's 01, there's 03, there's chat GPT 5 and then 5.2, and then again just over the past week, they just broke chat GPT 5.4. They need to keep coming up with these new models to fend off the competition's models. And they need to be smarter, they need to be better. Everyone's focusing on these benchmarks and the models have come such a long way over the past year, they feel like they're in this nuclear arms race. Really. It's primarily chatgpt, Claude Gemini. And I guess if you think that GROK is part of the set, then grok, but you know, they're all duking it out so that.

[C]

And when I train a model, tell us what happens when I'm training, like from 5.3 to 5.4, what is, like, what do I have to do to train the model?

[D]

You know, I'm going to leave that to the people at the company. I wish I could say, oh, All I can say is the nature of the expense doesn't vary as a function of subscribers. What it is, is it's just finding ways to make the algorithms better. Maybe it's some combination of that training on better data. And I'd say the other thing that often gets lumped into that category is building on additional functionality for the services. And so, yeah, I remember back in the day, I didn't used to be able to have Claude just generate any sort of Excel file for me. Now it can generate Excel files with all the formulas in it, and it can analyze all of my Excel files, you know, all the inner workings of them to find errors in them. And so, you know, that wouldn't necessarily. That'd be like some combination of, you know, better model, but also just like better functionality, better interoperability with, you know, with the sort of tools that people use.

[C]

And this costs a lot of money is what you're saying. It's like if I'm running a restaurant, this is the cost of the steaks.

[D]

Well, the inference costs, those would be the cost of the steak. This would be. So like a restaurant doesn't really quite have the same notion because it's like, what's the menu we should have? You know, how are we going to make the next breakout meal? Yeah, maybe it's the cost of the chef, you know, the, the chef that comes up with the ideas for.

[C]

Got it.

[D]

The spring menu. But just to put some of the numbers in context, like in, in 2024, ChatGPT had spent $5 billion on training and R and D expenses when they generated about 4 billion in revenue. And then in 2025. Yeah, it's a lot of money, you know, but I think they spent about $15 billion on training in R&D in 2025. And if my numbers are correct, they're projecting that those training and R and D costs are going to go to $30 billion next year. So we're going from 5 to 15 to 30, potentially up to like $60 billion in 2027. So this is again, a tremendous amount of money that they're spending to come up with these new models.

[C]

All right, then at 60 billion, probably go sentient. And then we will have another issue. So. So when people are throwing around the valuations of these companies, you know, like over 150 billion and it's worth it, where's your head on that? Given all this expense and the CAC and everything else and the retention worth it or not?

[D]

It's really, to me, it's really hard to say because it's really hard to know when these companies will finally back off on the training and R and D expenses. Yeah, yeah, I think that that's a major swing factor. It's like a pharma company and they're all coming out, they're spending huge amounts on R and D for the next drug, but they have no patent protection. And so they just keep coming out with drug after drug after drug. And you'd hope at some point they get to the point that the market has segmented, people have their preferences for the one or the other, and then these companies don't have to spend as much on new models and they can focus a little bit more on actually making a bit of money. And I just know at what point that expense will settle down.

[C]

And also these folks are giving the drug away for free too, to start, right?

[D]

Yeah, they got to figure out, you know, so like for, for ChatGPT, they're the, you know, we're going to serve everyone company and so part and parcel with that is find some way to not lose too much money on the free people. Yeah, so, so I actually, I, I really respect that they, they want to kind of democratize access to this sort of model, but they really need to make ads work or they need to find some way of kind of creating enough of a performance boost.

[C]

So people will pay.

[D]

Yeah, the people will pay, you know, but yeah, I think the thing that's kind of holding them back is Gemini. They're looking to protect their ads business. So they make $160 billion on ads right now. And, and they see all these people, you know, MLMs, it's, it's the end of search.

[C]

Yeah, zero click search. You're just disappearing. It's gone. Yeah, yeah.

[D]

So they're saying we got to defend our ecosystem, you know, and, and if all these people start using ChatGPT for basically what they would have used the Google search for, then we're toast. And so, yeah, I think that they're basically significantly subsidizing their business just to kind of keep people in the Google ecosystem. So we just wrote this, this working paper and we were comparing basically like the mobile app revenue monetization of Claude, ChatGPT and Gemini. And we found that Gemini monetizes about 30 times better than. Sorry, Claude monetizes 30 times better than Gemini.

[C]

Yeah.

[D]

And ChatGPT monetizes like three times better than Gemini. So you can think of it like Gemini is severely unpriced, I think.

[C]

Yeah.

[D]

And they're doing that intentionally to keep people coming back to them.

[C]

160 billion dollar ad business a year is a lot of, you can sacrifice a lot of tokens for that.

[D]

Yeah, that's, that's a humongous business to be able to protect. But you kind of feel bad then for Clyde and, and for, for Chat GPT because they got to make their money on the LLM. You know, they're not, they don't have anything to protect, they don't have anything to subsidize the core business except venture capital investors. So yeah, so they got to make their money on this thing alone. And you know, if you're these companies then your free plan, it can't be too much worse than Geminis because if it really sucks compared to Gemini.

[C]

Yeah, Gemini will take it all.

[D]

Yeah, then Gemini will just take it all. So I think that's actually holding down all of these companies ability to. You think, why not just make the free tier shittier? Just make it. Well, it's like, well, I can't do that because Gemini is just going to eat my lunch.

[C]

Yeah, Gemini has the pole position because they have all the search.

[D]

So yes, it's a very tricky competitive dynamic right now.

[C]

All right, Dan, do me a favor. Write marketers into the story. What should they be thinking about? We already talked about, you may not want to put all your eggs in one single basket here. What else should they be thinking about when you look at this and then what indicator should they be watching?

[D]

Well, certainly so I think, you know, diversification of your providers is probably prudent. You know, I think that obviously these companies are proving themselves out and they've got, you know, they've got good value propositions to the users and we can kind of see that through customer retention. But you know, I think diversifying your providers is a variable thing. It's a very valuable thing to do. The other reason that it's helpful is because, you know, like Claude for example, they were, they were basically down for the entire Monday, right. This big flood of users. And imagine if you only relied on them for kind of mission critical things that you're doing. That's kind of a bummer that you just lose an entire day's worth of work, you know. So yes, I think that that's just prudent business practice in this category. Yeah, so I think that that's one. Yeah, I think it really pays if you're a marketer or honestly if you're a consumer to continuously be testing multiple services as well. Yeah, I think that because They've gotten so much better and they're constantly changing. You always want to be running exactly the same queries against different models over time. And I think what it can allow you to do is see the progression of how well it can answer those questions, how that's been changing and how it's different across the different providers. And I've personally found I do this a lot and I kind of have to a little bit because of the testing that I do for one of the companies. It's amazing how much they change. So, so what might have given you the best answer yesterday might not necessarily give you the best answer today. And, and so knowing that is incredibly valuable. It just allows you to kind of stay, stay one step ahead of the game. But when I'll talk to people about it. Yeah, I think it's just that people are busy. You know, you only had so many, you know, so many hours in the day. And so do you have the time to be taking your queries and copying and pasting it, doing this and doing that, and kind of looking at the response and comparing the quality. But I do think because of just how potentially game changing these services can be for our productivity, you, you owe it to yourself. It's like an investment in education to kind of do that, to make sure that.

[C]

Let's talk about the education, because you also did research that said AI education and you can explain exactly what that means is really the key to returns on AI. Tell us what the research said, what it means, and what it means to our users or listeners.

[D]

Yeah, so one of the other hats that I wear obviously is I'm a professor of marketing here and I've been involved with some of the AI initiatives at University Maryland College Park. We've been thinking really hard about how do we train the next generation of leaders and how do we work that into the MBA experience. And it's a really big question right now. And I think the tough part is we all can kind of see it happening before our eyes. But for one, not every single professor is going to be an absolute expert at AI. And so to teach it, you got to know it first. And then for two, it's like, well, how do I weave it into my class and do I weave it in differently for the core classes versus these elective classes? And then there's the question of how are people in industry using AI right now? And actually that comes a little bit to your question about the marketers. Unless you've got that foot either constantly talking with people in industry or you're doing it yourself. You may not just be that aware, especially because everything is changing so fast in terms of how people are using this stuff. So you may be somewhat proficient in your ability to use ChatGPT or Claude code, something like that. But if you really want to help the students get their next job, you need to know, well, how are people using it in the real world in those industries? You know, and that's. Yeah, that. That's just another thing that we have to learn how to do well.

[C]

And there's no playbook. We've had a bunch of people come on and say, there is no playbook. There's no best practices. Everyone's inventing it as they go, why it moves forward at speed. So any tips other than I hear you saying, actually find the time to play with this enough that you have an opinion versus you just see it. Is that fair?

[D]

Yeah, I mean, certainly. That's a big one. Yeah, the playing with it. And I like the use of the word play because that kind of keeps it fun. And I actually think that LLM use. Well, I'll speak for myself.

[C]

There you go.

[D]

It can be really fun.

[C]

It can be really fun.

[D]

Yeah, it's.

[C]

You can have it do all kinds of stuff. That is hilarious.

[D]

Yeah. Be it coming in with that mentality. I think it takes some of the edge off. Like, I have to learn this whole new thing. You know, it's actually like, oh, I'm gonna try to have a good time with this.

[C]

Oh, the caricature thing is hilarious. On ChatGPT, if you want to make fun of your friends, you could do that really well. So I think this is. So I hear you say, giddy up, get using this. Have fun with it. Don't be so intimidated. You don't use it. Which brings us. And if I'm wrong on that, you should stop me because I'm going to go to our traditional last question.

[D]

If I say maybe the one other thing is don't ignore the role of foundational knowledge. I think there's this really nice report that was written by Anthropic about you kind of total output, you know, the people who get the most output from the LLMs. And yeah, I'd say that the TLDR on that was that people who had very strong foundational knowledge in the area, they were getting the most out of it because they knew all of the right questions to ask, they could probe it in the right way. When it gave wrong answers, they could be able to kind of like detect it and kind of steer it back in the right direction. And so it's not that there's like an additive relationship between what I can get from AI. It's not like I can't become a

[C]

brain surgeon by studying AI. It'd be better for me to work on some stuff I might know. That's what you're saying, right?

[D]

Yeah, you just won't be able to get the most out of it. So you need to continue to invest in yourself, too. And it's not just investing in AI. You need to learn all the base level stuff too. This is just something that kind of. It's like a multi, multi, multiplier that it kind of makes you that much more effective over. It's like a percentage increase of what it is that you already know kind of in your random access memory.

[C]

All right, well, I think this is a great segue into our traditional last question, which, Dan, you know, practical advice for our audience we haven't yet discussed, or the funniest news story you can share on the air. You can pick one or both, but you have to pick at least one

[D]

practical recommendation. All my students are going to hear this, and they're going to be like, oh, you're doing it again. But I have to say it. I have to say it.

[C]

You heard it here first.

[D]

Talk to your phone. Talk. Talk to your phone. And what I mean by that is people will kind of type in their queries. You know, they go to, you know, ChatGPT or Claude, and they're there typing it out. And I respect that, but typing is effortful, it's slow. And you know what the LLMs crave if you want to get a really good answer. What they crave is context. The right context. And so what I'll be constantly doing, you know, here in this basement office of mine is I'll just kind of go with my phone in my hand, and I'm just walking around talking for. For a minute, you know, so if there's something I want to ask it, I end up with this humongous amount of words, you know, that really kind of fully elaborate my goal. What I'm going for, why am I even asking this question? And. And the results are just so much better. So. So resist the temptation to treat it like, you know, I need to go on my computer and type everything out, speak to it.

[C]

And I hear you saying, talk to it like a coworker or an agency or a consultant. Don't talk it to it. Like a piece of technology.

[D]

But, yeah, but not advanced voice assistant use it just to get the words down and then hit Enter on a good model, and I think you'll be pretty surprised how much better answers you'll get from it.

[C]

All right, so giddy up, talk to your phone, have fun. Thank you Dan and thanks to everyone for listening to CMO Confidential. If you're enjoying the show, hit the like button and subscribe. New shows drop every Tuesday and you can find our more than 160 shows on Spotify, Apple and YouTube, which include the Warby Parker case. I can see clearly now through my CLTV glasses. Colonel Mustard in the study with the Job Spec Marketing at Meta, the View from the Eye of the Storm, and the Truth behind the curtain in B2B marketing. Hey all you marketers, stay safe out there. This is Mike Linton signing off for CMO Confidential. Typeface is changing the way to think about brand marketing at scale. Their marketing orchestration engine is the first of its kind and built specifically for the enterprise. The orchestration engine uses shared brand intelligence designed to turn brand guidelines into personalized voice, visuals and messaging delivered in a way that fits the context of your audience. It's how brands like Asics and Post holdings scale what works without sacrificing quality. Start orchestrating your brand at Typeface AICMO.