They spent a long time trying to convince you that AI wasn't a bubble, but it is. And finally, many of them are admitting it. But now they're trying to convince you that it's like the dot-com bubble. And that when the smoke clears, it will be a whole new and much more prosperous world. But this bubble's not like the dot-com bubble. And not just because it's way bigger, and not just because it's a higher percentage of the economy is being propped up by it. But because generative AI, unlike the Internet, is not a big step down a path that could lead to prosperity and abundance. In many ways, this bubble is much more akin to the cryptocurrency bubble than the Internet one. The reason this bubble seems so much more pervasive than the crypto one is because, unlike with cryptocurrency, the myth underlying this bubble shifts control toward the biggest stylish companies instead of away from them, which greatly increases the hype and the propaganda. And it seems to me that, like with cryptocurrency, generative AI has been hyped far beyond what the tech can actually deliver. And when this bubble pops and people see what the tech is really worth, we're all going to be F------

This is Internet of Bugs. My name is Carl. I've been a software professional since a decade before the dot-com bubble. And for the last couple of years, I've been here on YouTube trying to make the Internet a safer and less buggy place. I also have a Patreon community where I post exclusive videos, have community text chat and weekly live calls. I'd love to have you participate. If you're so inclined, you can go to the InternetOfBugs.com or follow the links in the description below.

Okay, so unlike most of the people comparing AI to the dot-com bubble, I actually lived and worked through the dot-com bubble. So let me set the scene for you. The promise of the dot-com bubble was that the Internet would disrupt the way that people bought and sold things. And it would open up the opportunity for lots of services to be provided that would not otherwise have been possible, allowing for both providers and consumers to prosper. Turned out eventually that turned out to be true, but it didn't happen the way or in the timeframe that the investors hoped it would, because there was a hard problem that still needed to be solved, what was called the "last mile" problem. In order for the Internet to reach the potential of its promise, we had to figure out how to get high-speed connections into people's homes, and we had to figure out how to get packages moved from sellers to buyers in a much more cost-effective way. That took an extra decade or two. But in the investment mania of the late 90s and early 2000s, a lot of infrastructure was built. A lot of interconnection points were set up, a lot of fiber was pulled between population centers. That infrastructure was put together a lot faster, with a lot more capacity than was needed at the time. And so it didn't generate the return on the investment that was necessary to keep everything going. So the bubble popped, and a lot of companies went out of business, and a lot of investors lost a lot of money. But eventually those fiber cables and interconnection points were put to good use to make the Internet that you're currently watching this video on. So the four things to remember are: one, the promise, a huge expansion of commerce. It was there and it was correct. Two, the winners. The people who wanted to buy or sell things to people not near them and people that moved things between buyers and sellers. They were far more numerous and far more powerful than the losers, the brick and mortar retailers, who wanted to maintain their position in the value chain. Three, the roadblock between the promise and the fulfillment, the last mile problem, was solvable eventually. And four, the things purchased by all the speculative investment money, the fiber cables and interchanges, were still usable and valuable when the roadblock was finally solved.

Okay, now I'll real quickly run down the cryptocurrency bubble. The promise there was that information, financial or otherwise, could be stored and retrieved and verified without the need for the gatekeepers using a technology called the "blockchain." That would enable commerce and services that would not otherwise be possible. That's true, I guess, as far as it goes. There's quite a bit of debate about exactly how valuable that newly enabled commerce would have been, but there would be some value at least. The roadblock, well, the biggest roadblock, there may be others too, but the main roadblock is that in order to participate in this new ecosystem, everyone needed to acquire the technology to connect to it. Electronic wallets, new points of sale system, that kind of thing. It's pretty much the same as the kind of last mile problem that the internet had. It's not insurmountable, but it's time consuming. The winners and losers, though, that's a big difference, because the losers would be the people and the companies that are currently the gatekeepers of the old systems, banks, credit card companies, financial institutions, that kind of thing. And they are far more powerful than the brick and mortar retailers who would lost their business back to the internet in 2000. And that means that they were much better positioned to throw a wrench in things. And then finally, the things purchased by all the investment where computers used to manage and work with the blockchain. And it turns out that the kind of operations those computers were made to perform are very similar to the kinds of operations that are needed to run Generative AI, which means there wasn't a huge crash. The resource was just kind of rolled over into the next bubble. I'm oversimplifying, but that's basically the idea.

Now, let's talk about Generative AI. Not hypothetical AI. Just the current LLM transformer-based chat GPT kind of Generative AI. Lots of promises have been made about what this generation of AI will be able to do from "replace 50% of all white-collar workforce" to "cure cancer," "increase human life spans," "kill the entire human race," et cetera, et cetera. Is the promise real? I don't think so. But let's assume for the sake of argument that it might be, and let's see how the rest of this would play out. Who wins in this scenario? Well, the people that win when companies make more money, the investors, executives, key employees, those kinds of people. Who loses? Everybody else, especially the people whose jobs would get replaced by AI. So you can see why the investor class and AI companies are pushing this so hard. If this actually happens, they stand again immensely. So what's the roadblock? Well, it depends on who you ask. What we do know is that the promised productivity gains are not being realized. The AI companies are losing a ton of money. Customers are complaining about how much their AI token use is costing them, and the only people making more money are the chip manufacturers. If the productivity gains were real, then the companies that are customers of AI would be making so much more money that the cost of AI would be worth it, and they wouldn't be complaining, and this wouldn't really be a bubble, or it'd be a very small bubble. It may just be that it takes a while for the gains to show up. It may be that the companies have to figure out how to use AI to be more productive. It may be that with more scale and more data centers, the gains will just appear, or it might be, because the promise isn't real. But it does not appear to be a last-mile sort of problem. Adoption has not been slowed, nor limited, because customers needed better access or more of some resource in order to make use of the service. We're seeing record levels of customers flocking to use these products, but despite that, we're just not seeing productivity gains. Generative AI is not a situation where the promise is clear, but the adoption is lagging. In fact, the opposite is true. But the most important question is, what is the investment buying, and will it still be valuable when the roadblocks are solved, or can they be used for something else if the roadblocks don't get solved? And the answer is pretty much no. All that investment is going to build data centers and fill them up with computer chips to run more AI. How long are those data centers and chips still going to be valuable if there's a long wait for the roadblocks to get solved? Probably not all that long. The limiting factor on how much work can be done in a data center is power. That's electricity. And it takes years for new power plants to be added to the electric grid. And the number of operations that can be done for a given amount of power is doubling about every two years. So by the time new power is finally ready to be used in a data center, any chips that were purchased at the time of the power plant started building would be only a quarter or less as efficient as whatever the current generation of chips is. That's very, very different than the case with old fiber runs. Because of increases in the efficiency of electronics and the protocols used to move data over fiber cables, the amount of work that can be done over a fiber connection increases over time, or at least it has. So the fiber that was run before the doc-com bubble popped were even more useful by the time the roadblocks were cleared than they were when they were first put down.

The most important thing to consider, though, is the promise. The underlying advancements underlying the bubbles that have proved worth it over time are the ones that enabled far reaching increases in productivity and commerce. The internet, the railroads, the cargo ships, the airplane, the telephone, the things that really lifted all boats, as they say. Those technologies all distributed economic opportunity. Yes, the railroads and the telephone created big monopolistic players, the Central Pacific Railroad and AT&T. But at the same time, they created opportunities for a lot of other businesses. It opened up huge, continent-wide markets for goods to be sold and delivered that make thousands, if not hundreds of thousands of companies successful, in ways that wouldn't have been possible beforehand. Crypto also didn't lift all the boats. In theory, blockchain technology should be democratizing. But in practice, the vast majority of all the winners were either in the exchanges like Coinbase or FTX, a facility that transfers between cryptocurrency and currencies like Dollars and Yen. Or they were the crypto miners that devoted a huge amount of sophisticated hardware to mining make it virtually impossible for masses to compete. So the question seems to be, does generative AI allow for opportunities to be scattered across hundreds of thousands or millions of companies, or does it benefit only a few? Well, not exactly. Generative AI might distribute economic benefit, or it might consolidate it, but that's not really the relevant question. The important question is, do the resources being funded by the investment bubble benefit the fewer than many? And I think it's clear that the current capital spending is going to build data centers and buy high-end AI processor devices. And I think it's clear that giant data centers, filled with the largest multi-tenant AI processors that money can buy, are not going to benefit just everyone. Huge data centers create the conditions for large companies to charge large rents to use their AI's. And rent seeking extracts money from the economy, and it doesn't increase the overall size of the economy. Taking the inner AI technology and running on local processors could maybe end up facilitating, making the economic pie-bigger or rising tide raises all boats kind of thing, although the real productivity games that would allow that to happen haven't been measured to date. But the bulk of the current investment isn't like that. What's driving this bubble isn't like that. These data centers being built are such that each of them will draw more electricity than almost any US city, any except for the ten largest. That's not something that creates new markets or new opportunities for commerce. It doesn't connect businesses to new consumers, it centralizes the economic benefit to those who can afford to build and run giant data centers. And that's why generative AI bubble won't have the long-term benefits to the economy like the internet bubble, or the railroad bubble did. It's not funding technology that helps the larger economy increase its overall value. This bubble is funding technology that helps the large companies keep an even bigger share of the profits, which leaves less money and opportunity for the smaller companies where the innovation typically happens. It's not expansive, it's extractive, and it makes the rest of us worse off. And most important for the people that watch this channel, it makes the internet worse off. And that stinks because the internet already has far too many problems. And anyone who says differently is probably selling something like a fairy tale about how the data center being built next to your neighborhood is good, actually. Thanks for watching. As a quick reminder, there are links below if you'd like to be part of the Internet of Bugs Patreon community. And let's be careful out there.