Is AI a Bubble?

Announcer (00:04): Welcome to Raw Data with Rob Collie, real talk about AI and data for business impact. And now, CEO and founder of P3 Adaptive, your host, Rob Collie.

Rob Collie (00:20): Hello, friends. There's been a lot of recent news suggesting that AI is essentially the latest bubble, and a bubble that's going to pop sooner or later. I've been getting a lot of questions about what I think about that, so I figured I'd do a short episode on the topic. Now, there have been two flavors of these articles and headlines. The first flavor to come along was, essentially, there isn't enough revenue to justify all of the spending. This was primarily aimed at OpenAI and the gargantuan amount of money that they've either been spending or getting their hands on. The questions being asked in that first wave of news all essentially boiled down to how can they ever pay that all back and get into the black?

(01:00): The second, and more recent wave of news, has highlighted the amusing snake eating its own tale of nature of all these companies investing in each other. NVIDIA, for example, investing a jillion dollars into OpenAI so that OpenAI can afford to buy the next round of GPUs from guess who? NVIDIA. NVIDIA gets to say, "See, demand for our chips is going to remain robust into the future," even though in some very real sense it's just NVIDIA buying their own chips from themselves. There are other versions of this too, involving Microsoft, Amazon, Oracle, and other big players in the data center space. The flavor of that one is, "We give money to AI companies who then give it back to us when they rent our data center capacity." So, is this behavior fishy? Well, yeah, I think it's definitely fishy, or at least highly unusual. We don't see this sort of financial creativity very often.

(01:52): Now, I'm sure it takes place quite frequently at individual companies kind of in the dark. You know that creative financial engineering designed to juice a company's stock price? Oftentimes those situations do end quite poorly like Enron. But to see this sort of thing happening out in the open, at an industry-wide scale, with basically every major player involved in the same shell game... well, that doesn't happen often at all. In fact, it might be the first time we've ever seen anything like that, and it's definitely worth noting.

(02:19): You can view this second wave of news, the "I lend you billions of dollars so that you can use it to buy billions of dollars of stuff from me" as fitting very nicely together with that first wave of articles, which we're asking, where's all the revenue going to come from? See, if the revenue isn't going to be there in time to offset all of this borrowing and spending, well, that would create the conditions where these big tech players needed to do something creative in order to keep the game going. So it's kind of like, the shell game wave of news is basically our proof that the first wave of news, the "where's the revenue going to come from," was 100% on target.

(02:54): Where is the revenue going to come from? Well, it's going to come from each other. LOL. So let's just call it a bubble. At the same time, we should note that not all bubbles actually pop, and even when they do, they often take a lot longer to pop than what you would expect. But let's just admit that there's at least being a bubble inflated here. And then let's ask, what does that mean for us? On the one hand, as private investors whose retirement accounts might be overweight in these companies, it might mean quite a bit, but as people who are trying to improve the businesses we work in, using AI, I actually think it means very little.

(03:30): I'd like to make a big distinction here. The bubble isn't AI itself; the bubble is the behavior of the tech companies. It's a financial bubble, not a technology bubble. In other words, if this bubble is real and it pops, what happens? Well, companies involved in it will get hurt, and the more those companies are all in on AI, the greater the risk. So for example, in the most extreme case, OpenAI could go bankrupt. NVIDIA's stock price could plummet. More diversified firms like Microsoft, Google, Oracle, and Amazon, well, they'd also take haircuts, but not nearly as sharp as the more pure AI plays like OpenAI. That said, even if OpenAI did go bankrupt like pets.com 2001 bankrupt, ChatGPT would live on. The LLM itself, with all of its fabulous training and intelligence, would still be sitting there in data centers ready to work, just like the villagers in the classic late '90s video game Warcraft II.

Announcer (04:31): Ready to work.

Rob Collie (04:32): The productivity-enhancing usefulness of the LLMs themselves, in other words, would be 100% the same the day after an OpenAI bankruptcy as they were the day before such an event. The usefulness of these tools do not care about the stock market.

(04:48): Going back to the episode a couple of weeks ago with Bill Krolicki, their AI-powered vendor bot, which has cut their manufacturing outage rate in half while also improving their margin and effectiveness even in the cases where materials still don't arrive in time, well, that vendor bot would just keep humming right along doing its thing in the event of an OpenAI going bust type event. To be super clear, even if their vendor bot is calling OpenAI APIs, which I think it likely is, they can be pretty sure that those APIs, those calls, would continue to function even on a day where OpenAI "failed."

(05:24): That's how bankruptcies work. The failing entity, if it's still collecting money, will continue to collect money while its new owners figure out what to do with it, and there would be new owners because all of those LLMs developed by OpenAI are incredibly valuable IP. It's in everyone's best interest, the new owners of OpenAI, in that case, the existing employees of OpenAI and all of their customers, to keep that money flowing. But even if there was some 48-hour outage in the API because of some unforeseen behind-the-scenes power struggle or something, International Packaging, Bill Krolicki and crew, they could just switch their API call over to Anthropic or Google and not really miss a beat. The real benefits of AI for business value are completely separate from the fortunes of specific tech companies in the market that are making big investments in it.

(06:16): If it turns out that these companies have just not been charging enough to build sustainable businesses, well, maybe we'll have to pay more in the future. But even if the prices we have to pay did something dramatic like doubling or tripling, I think we'd still all be paying it. These things are still very inexpensive compared to what they actually do for you. Now, we might be a bit more careful with our experimentation if prices dramatically jumped like that, but in terms of deployed solutions, in most cases, the LLM costs aren't really that big of a deal. The ROI on vendor bot at International Packaging, for example, is insane. While the costs of the calls that they make to those OpenAI APIs, that cost is miniscule. In that particular case, I bet they wouldn't blink if the API costs went to like 50X what they are today.

(07:03): Now, there are plenty of deployment scenarios, plenty of use cases where AI would no longer be cost-effective if it were 50 times as expensive. But again, I think it would be dramatic to see a doubling or tripling in price. Long story short though, as far as those of us who are out here using AI to improve businesses, the primary impact we could expect to see if this was all a bubble is that they weren't collecting enough money and then they need to collect more. So worst case would be some sort of significant price increase.

(07:30): I think one of the reason these bubble articles get so much attention is because of how much we, as human beings, secretly want AI to be a passing fad. Many of you listening are in that boat, and that's okay. It's perfectly normal. AI is disruptive to our pictures of our careers. It introduces uncertainty into our value as professionals that none of us expected or wanted. When something like that comes along, of course it generates a conscious or unconscious rooting interest against that thing. So if you're in that boat today, don't feel bad about it. It's just good to self-reflect a little bit on why you might have been unconsciously happy to read those articles.

(08:10): To give you an alternate picture of happy, I think there's basically zero chance that the LLMs, the AI technology on its current trajectory is going to replace us as professionals. We're going to need to see something completely different, a complete change of trajectory in order to put anyone listening to this truly out of a job. Now, maybe that's coming. There are people working on it. There are AI researchers working on it for sure right now. How successful will they be at inventing that next big thing, and how long will it take them? Unknown. But after several years of just seemingly breathtaking breakthroughs, breathtaking expansion and capabilities, we're on kind of an LLM plateau right now.

(08:50): These things aren't getting exponentially anymore. Most of the progress we're seeing right now is progress in terms of plugging LLMs, plugging AIs into other non-AI regular software systems. The innovation right now is coming more from traditional, like just plugging Lego bricks together type of innovation, type of advancements, type of work. So the current AI trajectory isn't going to put any of us out of a job. It is good enough though that it will force us to adapt. Our jobs will not be the same, and we can't cling to the idea that tomorrow is going to be like yesterday. There are adjustments you need to make. The good news is that you can do it. You can make those adjustments. But even better, these changes will also open new doors, oftentimes better doors, even as some of our comfortable, familiar doors slowly close. And all of that, for better and worse, is independent of NVIDIA's stock price.