A lot to unpack here, but first let me stress the difference between the AI investment bubble bursting, the AI hype train derailing, and any chance that AI is going to disappear from your life. All three are big questions and each has a very different answer. Working backwards:
AI is not going to disappear from your life. In fact, you’re gonna have a difficult time avoiding it. Software companies are going to keep jamming it into everything, and eventually it will make your life easier and better. But it’ll probably annoy you and mess things up along the way. Electricity was a lot like this. Intermittent and prone to setting things on fire, but the benefits were great enough that we pushed forward, refined the tools, and accepted it as a necessity. The internet went through similar growing pains.
So for those who hope AI will just go away, it won’t. And you might not believe this now, but in ten years you’ll wonder how we got by without it. Automated customer service without ever talking to a human and getting immediate and better results is just one thing worth looking forward to. Personalized travel agents who know what to book and handle all the details for you. Only ever seeing emails that need your personal attention (yes, your AI and another person’s AI will sort things out without your involvement). Guess what? Most executives have staff who do the same thing, and soon everyone will be able to afford that luxury. Your real estate agent needs a smoke detector form signed and notarized? Your AI will just make it happen. Anything electronic you don’t enjoy doing can be automated.
That’s what we have to look forward to. But right now, the tools are not living up to expectations. In fact…
… the AI hype train is derailing as I write this. AI will eventually lead to gains in productivity, but those gains are going to be marginal. That’s the optimistic take. Right now, there are practically zero gains. A study published just this month found that 80% of 6,000 firms polled showed no gains in productivity from their integration of AI. Forecasted gains of 1.4% over the next three years are expected, but that’s not the employment apocalypse many fear. And those are just guesses. We are just as likely to see a productivity paradox, where distraction from these tools and time spent double-checking their work takes as much time as doing it via human labor.
There are already cases of supposed AI turning out to be remote operations from cheap offshore labor mills. It even has a name that I absolutely loathe and because it annoys me, you’re going to have to live with it as well: Fauxtimation. The Nate example is one of the more egregious and fraudulent. But Google has done this with its Duplex service, Amazon was doing it with their cashier-less stores, and Tesla does it with their driverless taxis. There’s a lot more fake-it-until-you-make-it happening than actually making it. My personal experience with this was being at a tech conference where a robot was making and serving drinks, and me and a few friends followed a power cord to a curtained area, behind which was a human in VR controlling the robot manually. The magic of AI is currently a bunch of Wizards of Oz.
A few issues I foresee while we’re on the hype train. Two of the biggest problems are both summed up in the Pixar movie The Incredibles. Helen tells her son, “Everyone is special, Dash.” To which he replies: “That’s just another way of saying no one is.” The benefit of AI is the democratization of digital labor. The profit problem AI has is the democratization of digital labor. When everyone can make an app, there’s no profit in making apps.
This is a problem both on the macro and micro scales. On the micro scale, someone will tell you they vibe-coded an app to handle splitting dinner receipts among friends in a single day and launched it on the app store. The problem will be getting people to pay money for this app when their AI can handle this task for them already without the need for an app. It’s not just programming jobs that will get replaced, it’s the programs they were making that will no longer be needed. Instead of purpose built apps, you’ll have agents who can handle most tasks. Two historical examples that are similar: the smart phone replaced dozens of purpose-built devices all at once, and the internet browser replaced dozens of purpose-built apps that used to live on your desktop.
On the macro scale, the browser provides another good historical lesson. There was a similar level of panic and investment in silicon valley back during the “browser wars.” Microsoft’s antitrust problems stemmed from this era, as they tried to limit Windows customers from using rival browsers. The idea back then was whoever won the browser war would profit just like the companies who won the OS war. The problem is that customers weren’t willing to pay for a browser when free options worked just fine. It’s worth diving deeper into this to see why AI will follow the browser route and not the OS route.
Operating systems might have been free as well, with UNIX and Linux giving it their best shots. Power users back then (I was one of them), loved installing these distros and fiddling with all the dials. But most users just want something that works, without needing to troubleshoot anything. Simplicity always wins out over quality (see MP3 vs FLAC in the music space). Most of what a customer pays for when buying a computer is the hardware. The software is a fraction of the cost. So why pay $1,000 for a LINUX laptop that requires a computer engineering degree to operate when you can get a Windows for Mac laptop for $1,050 and it uses a uniform OS that you are already familiar with from the computers at work, the library, school, etc? Uniformity, familiarity, and simplicity win.
Once you were familiar with the OS that came with your PC or Mac, you were trapped in a upgrade cycle that cost money over time, upgrading from Windows 3.11 to Windows 95, to Windows ME (omfg), etc. Microsoft made lots of money without needing to make a product much better than the best Linux distribution (which got better and easier to use, but by then it was too late). The lesson here isn’t that Microsoft made a great tool that increased our productivity, the lesson is that Microsoft got lucky by being first and humans are lazy and set in their ways.
The first mover advantage was strong in those days. It’s less powerful now. OpenAI is not likely to survive the AI wars, because they need to copy someone’s profit engine and business model. But all other companies need to do is copy OpenAI’s training techniques. Which they already have. Chinese companies then copied those techniques, leading to models like DeepSeek that do most tasks for free just as well as paid-for models. And here’s where the browser wars have something to teach us: customers will not pay extra when free is good enough. A company like Google, which makes plenty of money, is going to let you use their AI for free, which will prevent a competitor like OpenAI from ever making enough profits to vie for Google’s ad marketshare. This is why Microsoft and Apple don’t charge for the use of Safari and Edge (Internet Explorer). Charging you money immediately creates a market for their competitors. If they ask $19.99 for Safari, Apple creates a market for Firefox to charge $9.99 for a much superior product. With the browser now the true user Operating System, that’s not worth the risk.
Now for the AI Investment Bubble. This is the one most people are talking about when they refer to a bubble, and the reason it’s currently popping is related to the comments above. The productivity gains are not evident, and any path toward profitability is currently unclear. The entire bubble is powered by a combination of techno-religious fervor from Silicon Valley and get-rich-quick-don’t-get-left-out insanity from Wall Street. The biggest tell that we’re in a goldrush is that Nvidia is the only company making real profits here, and they’re the ones selling pickaxes and bluejeans. But the cracks are already showing.
Datacenter construction projects are already seeing cancelations. Capex promises are being revised downwards. The comparisons to the tech bubble of the late 90s misses something crucial here. After the tech bubble popped, we still had miles of fiberoptic cables, network infrastructure, and trained technicians and engineers. When the AI investment bubble pops, we’re going to have warehouses full of GPUs which get outdated with every passing year and don’t have any other good use. Cloud computing doesn’t need this much compute. Serviceable LLMs already run just fine on your laptop or smartphone, without needing any cloud compute at all. And the amount of money being thrown around here is staggering, estimated to reach SEVEN TRILLION dollars by 2030. You’d get more productivity gains with that money if you gave people healthcare, high speed trains, and a functioning democracy. And that’s where the techno-religion comes in.
The justification for spending so much money is the race to reach AGI first. This isn’t conspiracy; AI CEOs admit as much proudly. The idea is whomever gets there first will “run the economy.” Winner take all. Destroy the competition. One person or company gets to rule the roost. And every other disgusting way you want to phrase this. Basically, betting the house because the CHANCE of winning is great enough that going bankrupt is a smart calculation. Meanwhile, most AI experts are telling us that existing LLM technology is not a path toward AGI. OpenAI itself released a paper saying hallucinations are an integral property of LLMs and can’t be avoided. So these tools will never be able to automate work without a human double-checking everything. And the cost of mistakes will only go up, especially when human lives are at risk.
The reason for the pullback from investors is pretty simple: ChatGPT 5 was a major disappointment. The gains seen in ChatGPT 3 and then 4 were not replicated. And every other model has seen a similar plateau. Investors were pouring money in thinking the gains would skyrocket, but it turns out that LLMs grabbed low-hanging fruit and can’t seem to reach much higher. This doesn’t make them useless, but it does make it pretty silly to divert a large percentage of the economy toward their further development. We might want to invest in other things as well.
Let’s zoom out for a second. The biggest companies on earth tend to extract value out of an enormous userbase by providing them with some value in return. Google gives great search results, so we spend a lot of time on their website, and our eyeballs and mouseclicks are valuable to advertisers. Amazon lets us shop in one place and delivers to our door, so we do a lot of our shopping there. Microsoft and Apple sell us hardware and software we can’t seem to live without, however much we loathe. Netflix gives us Love is Blind. Facebook lets us know which friends to stop talking to. All of them make money from hundreds of millions of customers.
What customers will a purely AI company have, and what service will they offer? The biggest use right now is creating images and videos, returning search results, and chatting. Existing players are already offering all of these for free or on the cheap, and the number of copycat models is just going to explode going forward. So what are they going to offer? How will they turn a profit? Or are they like Uber and AirBnB, which gave us value while the VC inflows were keeping them cheap but become hellscapes when they need to actually make money?
Microsoft and Amazon make a lot of their money now by offering tools and services to other companies. Is there a possibility for OpenAI and the like to do the same? I don’t see it. Google didn’t need to license OpenAI; it just built Gemini. Amazon is working on their own model. Apple licensed Gemini for now, but the cost of creating their own model is going down by the month (DeepSeek proved it doesn’t require billions to train a model as good as ChatGPT 5). To make back SEVEN TRILLION in capital expenditure, these companies are going to need to offer a compelling product that hundreds of millions of people are willing to pay for or spend their attention and clicks on. What could that be? Because I’ve got no idea. Our attention is already full. Population growth is going to slow and then reverse. Unlimited growth is not an option. Most of the consumers coming out of poverty and participating in the global market did so in the last two decades.
The last zoomed-out thought to consider here is that most consumer spending is accounted for, so your massive profits need to come at someone else’s expense. Google is getting advertising dollars that used to go to classified ads, newspapers, and magazines. Netflix is making money that used to go to Blockbuster, the cinema, and broadcast advertising. Amazon is making money that used to go to smaller retailers and other big-box stores (Circuit City, Toys-R-Us, and Borders are no longer with us). Apple is making money that used to go to Casio, Texas Instruments, music labels, broadcast radio, and others. Facebook is getting ad money from the same sources that Google is stealing it from. We have tech giants because of consolidation and a sudden rise in global affluence. Existing money and low-hanging fruit. So where’s the SEVEN TRILLION gonna come from?
If you think a small company is going to gobble up all the money that Google, Amazon, Facebook, and Apple are currently making, rather than these companies simply offering you AI services that eventually become helpful, I want some of what you’re smoking. When I zoom out, I see something different. I think AI is going to change your life for the better in ways you don’t expect, but it’s not going to change the lives of AI CEOs in the ways that they hope. Things will get a little easier for you, and they are about to get a lot poorer.