Episode Transcript
[00:00:00] Good morning. Today I want to talk about economics in the world of AI and HR tech. And there's a lot of stories and articles and analysts talking about the AI bubble and the massive amount of capital that's being thrown into this. I'll explain what that is in a minute, but the implication is that this stuff is going to get very expensive for you and me as a user. So let me talk you through what's happening. There's no question that in order to fuel the demand for AI, whether it be generative AI or images, graphics, videos, audios, et cetera, we need lots and lots of power in data centers. So over the last 12 months, roughly 1.5 to $1.7 trillion of capital investment has been put into the construction of data centers, which includes power, electricity, Nvidia chips and human capital. Based on the research I've looked at, about a third of it goes into the chips. About 50 to 60% of it goes into the physical infrastructure, air conditioning, power, et cetera, and then the rest into labor. So it's a small amount of labor, but it is labor too.
[00:01:07] And that investment is coming out of private equity and all sorts of private deals. For example, Meta issued a new off balance sheet debt product to fund its large data center, which is getting a lot of criticism. Oracle now has apparently $100 billion of debt just to fund the data centers that they've promised to build. With OpenAI and Oracle's stock plunged below what it was before they announced because Oracle is turning into a real estate company, not a software company. And so they're leveraging themselves up like a large real estate company. And you know what happens to them. And then the other companies, Microsoft has invested in this, of course, and Anthropic has put billions of dollars into infrastructure as well. Now, OpenAI and Anthropic are small companies. They're not that big financially. Their revenues are rather high. We believe OpenAI is going to finish around 10 or 11 billion, but that's questionable because we don't really know where that's coming from. OpenAI gets most of its revenue from consumers. Anthropic gets most of its revenue from business customers. And then along comes Google with Gemini 3. And all of a sudden everybody says, whoa, why am I spending so much money on OpenAI and anthropic when I can use Gemini, which is now better and beats everybody on the benchmarks. So you've got these big companies building infrastructure, worried that their demand is about to suddenly shift. And this is not like buying enterprise software, where it takes you five years to shift platforms. You can shift from Anthropic to Google in a. Well, first of all, if you're a user, you can shift in an hour or in a few minutes. If you're an enterprise or a software company, you can reprogram your interfaces to use Gemini in a few weeks and then turn Anthropic off. So if you're anthropic or OpenAI and you just spent billions of dollars on infrastructure, you're stuck. And the analysts are beginning to say that this 1.7 trillion or so that's been invested could end up somewhat stranded by the way, in the middle of all this. Of these big high tech companies, there's a bunch of other small companies, Cora Weave is one of them, that have suddenly jumped into this with no resources at all, but a lot of ambition and tried to create data center businesses of their own. So the analysts that I follow basically say that within a fairly short period of time, a lot of these smaller owners are going to go bankrupt and then the bigger guys will buy this stuff. Nevertheless, whatever happens, and let's hope, and it happens in a somewhat orderly way, we don't have a huge crash, but sort of feels like that could happen. The cost is high. If you look at the cost of consuming AI and it's usually priced as cents or dollars, but really cents per thousand tokens. A thousand tokens is a thousand, essentially small words. It's 2, 3, 4, 5 cents at the high level and then lower as you go up. So if you start multiplying out every employee in your company using AI two hours a day, creating thousands of words and pages and graphics and images, it's going to cost you some money. This is not going to be free, which means, as I'm going to talk about a lot in the predictions that you're going to want to focus your AI investments on high return areas. Helping people read their emails may not be cost justified. If it costs you a lot of money to consume the AI. And most of the vendors haven't started charging for consumption, but SAP does and workday is about to start doing that. So workday to have to pay for this. Now, in addition to the cost that we're going to be dealing with, there's the psychological effect of paying for something that feels like it should be free. Google search has been free. Now you're actually paying a very high price for it because of the ads. But the companies placing ads are paying for the search. You're not. So you don't see It. And of course, that business model has been so spectacularly effective that Meta uses it, doordash uses it, and everybody else uses it. But I think the better analogy, and I just heard this in another analyst discussing this for where pricing is going to go, is Uber. Now, if you remember the beginning of Uber during the pandemic and right before it, it was really inexpensive to get an Uber. I mean, I could go from Oakland to San Francisco for $6 or $7 and a cab would have been $75 or something like that. So we all got very used to using it. We were very got familiar with it, the convenience, the different features with the app on our phones. And it really turned out to be a money saver, timesaver and convenience. But of course, what Uber did is very shrewdly is over time, as they dominated more and more of the transportation space and cabs sort of disappeared from the face of the earth or became just kind of grungy, they started raising the price.
[00:06:00] And the price of an Uber now is almost as high as the price of a carry limo. In fact, it's higher. I take limos to the airport periodically and for me to go from Oakland to San Francisco is about 75 to $100 in an Uber black. It's about the same in a carry limo. The carry limo drives up and picks up my bags and zips them in the car and puts me in a beautiful car. The Uber driver doesn't do any of that. And I don't know what kind of car I'm going to get. So it's now a pricey, maybe you could argue, slightly overpriced product, but it's so convenient. We're still using it and Uber's very profitable and doing quite well. So good for them. That might be what happens to AI if you get so used to asking questions and not typing in search queries and asking the system, the computer to generate content for you or generate graphics or generate a video or analyze a spreadsheet. You're going to kind of forget how to open Excel or open Word and sit around and think about how to write something because it's just going to become second nature if the price starts to go up. You're just going to be say, okay, it's worth it, I'll pay for it. And so as we use these things more and the utility goes up, we're more willing to pay the price as the price goes up. And that's the expectation of all of these tech companies for this huge valuations of the AI. Vendors and these massive debt undertakings they're placing to build more and more and more data centers. So we as HR people, as tech people, as professional business people have to think through the high return use cases of AI. Now of course, the vendors are happy to help with that, but they're not really there yet. There aren't a lot of off the shelf AI systems that work extremely well as applications. Yet there's lots of applications within SAP, Oracle, Workday and AD ukg. All of them are building really cool AI apps within their HCM systems. And there's great recruiting tools and there's great learning and development tools and so forth. But we have to cost justify them. And so what I'm going to be telling you all about in the predictions is how to do that. And it's very, very interesting what you can do in hr. You can do some massive things. Let me give you the best example we have because we've been doing this for a couple of years in learning and development. The learning and development industry is close to 400 billion today. If you just include training, forget about knowledge management, and that includes onboarding, leadership development, technical training, sales training, operations training, compliance training, all sorts of things that are sprinkled around in different topics and courses and books and magazines and subject matter experts. And about 60 to 70 to 80% of it is internally developed. It's not off the shelf. Your training in leadership is different from another company's training in leadership. So you might buy a vendor product or offering, but you're still going to change it. And that market is about $1,400 per employee per year in consumption. So the average employee on Average is around 13 to $1,400 of training per year. High level consultants receive $3,000, $4,000 of training per year. Executives receive 10,000 to 15,000 to $20,000 of training per year. Frontline workers in a McDonald's might get a couple hundred dollars of training per year. And they get a lot of training from their managers who get training at a medium level too. So we've got all this training consumption and behind that we have instructional designers, we have performance consultants, we have LMS administrators, content developers, graphics developers, assessment specialists, many, many really interesting domains. You go to a training conference and you'll be fascinated at the amount of innovation and training, including virtual reality and simulations and all sorts of cool things. But that's expensive. And we've now done three projects for large companies and when they're in the middle of some others where we've done Learning and development transformation. And what you find is that the amount of money spent in corporate L and D might only be a third of the total L and D spend because the sales and the manufacturing groups and other people have their own training budgets. And so the company's spending 1 to 1 and a half to 2% of its revenue on this whole area of training people. Now, you know it's not a small number, but it's well worth it because if you're keeping your company productive, it's a small price to pay. But that number, as we now know, could be cut in half at the corporate level easily with a new platform like Galileo. Learn, because you won't need the content development resources, you won't need the performance consulting. The LMS administration is enormously easier. You don't need to do any manual skills tagging because the system does that for you. You don't need an LXP, etc.
[00:10:46] So you're talking about half the corporate budget on L and D possibly going away, maybe more. And those resources now working in the business to do more knowledge management, real time learning, interactions and support so you don't have to lay people off or you can. And that justifies a high price for an AI platform. And I know this because I know what people are paying for products like Sana that we use. And it is not inexpensive. And of course it means that the LMS part of the market is going to come down in price and the AI is going to go up in recruiting the same thing. You might think it's cost effective to just buy a new ATS and create a bunch of chatbots for your candidates. Well, no, it's expensive and you have to cost justify it. So one of the innovations from vendors like ams, Paradox and others is they do business cases before they sell you anything to figure out what the return on investment would be. Because for some companies the quality of hire more than pays for the AI. And other companies the speed of hire and the time of retention more than pays for the AI. And for other companies it's the cost of the HR talent acquisition operations. So each of these use cases has a financial return. And I think in 2026 we're going to have to sort of sharpen our pencils here because we're going to be talking about bigger, more super apps, which I talked about last time, and the price is going to go up and just like Uber, you're going to suddenly say, wow, this is kind of expensive, maybe I should drive, maybe I should take the Carrie, Limo. Maybe I don't need to go.
[00:12:25] Right. So what seemed like it was virtually free and frivolous and fun and exciting is not going to stay that way for a lot longer. It's going to stay that way for a while. But you can just see the pent up costs and expenses of these vendors. Now the other economic issue that is definitely worth thinking about for those of you that are buying and building and dealing with all this stuff is the vendors. My personal opinion, and I am not that informed, but I know a fair amount about this, is that the two vendors that are the most at risk are Anthropic and OpenAI. They're both doing great, but they're small compared to Microsoft, Google, Amazon. And they have decided because of their investors to become vertically, vertically integrated companies. Rather than being research labs, they're building their own data centers, they're selling enterprise, they're selling corporate, they're selling consumer, they're, they're probably going to deal with ads, they're generating their own data, they're buying data cleanup companies and so forth. So they're trying to be big companies, but they're not that big. So if you're building something or buying something, you're, you need to make sure you can deal with multiple LLMs. Now the product that we sell, Galileo, has multiple LLMs in it and a lot of the vendors do this. But if you're, you know, a big vendor and you're selecting a platform like this, you gotta be a little careful because it would not surprise me if one or two of these guys disappear. And my history with this goes back to the database industry in the 1990s, 2000s, believe it or not, the relational database industry was exactly like this. It was completely the same. There was Sybase, where I worked, there was Oracle, there was Ingress, there was Informix, there were a bunch of others building relational databases because relational databases were the big, big innovation in corporate applications.
[00:14:18] And they became easy to buy. They ran on unix, they later ran on nt and it was a war to see which one was going to win. They were advancing new features. There were benchmarks, they were called TPC transaction benchmarks. Just like we have these benchmarks on AI and everybody was, you know, making claims that their benchmarks were higher than somebody else's. And Oracle was taking out ads in the Wall Street Journal hyping their benchmarks. But look at it today. The relational database industry is mostly dominated by Oracle, Microsoft SQL Server and the embedded products from SAP that are embedded into hana. So yeah, there's others. There's all sorts of new database vendors and a lot of it's proprietary. Google built their own database, Amazon built their own database, but you don't go out and buy those things. So the same exact scenario happen here. Most of you don't have to think about this at all, but you. It's good to be aware of it. So I wanted to just bring it up because it's such a tumultuous time. And by the way, there's going to be more vendors. Deepseek is an extremely good product. When you go to China or Hong Kong, you can't get to ChatGPT, but you can get to DeepSeek. And Deep Seek works really, really well. So there will likely be other LLMs. And you know, as many of you may know if you're techies, a lot of the LLM scientists are beginning to talk about the fact that the LLM may not take us to the next level. And we need what are called world models, which means AI that understands science and physics and light and weight and mass and momentum, things like that. So there's going to be new technology as well. The bottom line for 2026 in our world is it's time to get really serious about big applications of AI. The desktop productivity part of this is going to keep growing, but that's the part that's hard to measure. It's fairly expensive and it's very difficult to really cost justify it until it just becomes a normal part of your life, which it is. But the big apps are going to be much, much bigger. And I'm getting a whole bunch of information for you guys on what those are. And our HR blueprint will. Our AI HR blueprint will really educate you a lot, and we'll get that out in Q1. One more thing. I'm going to produce a spectacular podcast from iata, the Industry Airline Transport association, about the airline industry. And I really, really recommend you listen to it. Jane Hawkinson, who is their head of Talent and Learning and Diversity, has been building skills and talent models for the entire airline industry using Galileo. And when you listen to that podcast, you not only are going to learn a lot about the airline industry, which is growing at extraordinarily high rates, believe it or not, but you're going to learn about talent management and Galileo in a way that'll really blow your minds, and you'll be able to get your hands on their model and use it. And she's also a huge fan of Galileo and the lesson that I learned from her experience and others, including some clients you're going to hear more about is that because Galileo is so narrowly focused on human capital and HR and leadership and management topics, and it doesn't try to answer questions about, you know, what coat to buy in the winter or what is the cause of autism or anything like that, it is really smart. And what you'll hear her talk about is how she uses Galileo as her thought partner in developing this future skills model for all of the jobs in the airline industry. So I'm going to publish that later, either this week or next week, but I really want you guys to listen to it. So I'll link a couple of really interesting articles here about the economics of AI. You know, everybody is going to be watching this and it's going to affect the stock market and it's going to affect our budgets. And I do think that the money we're spending on AI is going to go up. So get ready for more interesting, fascinating work here on cost specification, on project management, and all the great things that we do. And you'll also read in the predictions that we are not going to be eliminating that many jobs. The size of our HR departments will undoubtedly go down, but there will be plenty of things for people to do as we implement all this new technology. Okay, that's it for today. Have a great week. Bye, everybody.
