Water Cooler Small Talk, Ep. 10: So, What About the AI Bubble?

is a special kind of small talk, typically observed in office spaces around a water cooler. There, employees frequently share all kinds of corporate gossip, myths, legends, inaccurate scientific opinions, indiscreet personal anecdotes, or outright lies. Anything goes. In my Water Cooler Small Talk posts, I discuss strange and usually scientifically invalid opinions that I, my friends, or some acquaintance of mine have overheard in the office that have literally left us speechless.

So, here’s the water cooler opinion of today’s post:

AI is a bubble, just like the dot-com bubble of the turn of the millennium. It has no real value and is going to eventually collapse and fade away.

Phew! 😅 There’s a lot to unpack here. Unlike the opinions I usually discuss in other Water Cooler Small Talk posts, for this one, there is no easy, straightforward answer. AI is clearly an impressive technology, for sure, the breakthrough of our time, and a few years ago, we could only imagine such things within the realm of sci-fi.

Having said that, we cannot deny that there is a certain bubbliness to it — at least in its current state. For instance, Mira Murati‘s (OpenAI’s former CTO) new company, Thinking Machines Lab, is currently valued at $12 billion, basically, with no product. On top of this, NVIDIA hit an astronomical valuation of $5 trillion a few weeks ago (roughly equal to the GDP of Germany, for scale).

Undeniably, operating AI models is a rather expensive hobby, with no real returns up to now. Companies like Anthropic and OpenAI pay billions to run their fancy models (operating ChatGPT alone costs $700,000 per day), but at least for now, they get back pennies and operate on a net loss. So what’s going on? Is there a hidden piece of tech that will suddenly be revealed and render this entire mechanism sustainable? Are all these tech billionaires just fools? Are they scammers? Have we all been tricked into believing in an impossible, extremely expensive future? Or are we heading towards an AI winter, and that will be it?

💹 What about the dot-com crash?

The most popular argument of people calling AI a bubble is the dot-com crash of the late ’90s. But in reality, there have been plenty of other bubbles long before the dot-com bubble, sparked from some technological breakthrough. For instance, we can think of the railway mania of the 1840s or the radio boom of the 1920s.

Anyways! Something that people seem to forget when comparing the current state of AI with the dot-com bubble is that most of the dot-com companies that failed back then basically had no proven business model. In particular, in the late 90s, there were a lot of startups operating on funding and the promise of an Internet-fueled future, but they had no real business model of how to actually make money. Online ads weren’t really a thing yet, and a lot of those companies essentially had no solid business plan.

By contrast, most of today’s AI companies have a very structured business plans. And unlike in the ’90s, people — both individually and within organizations — are very familiar with the idea of paying for stuff online. Unlike the 90s and the 2000s, the idea of free stuff on the internet — like, for example, torrenting, or free value in general — is gone for good. We are now well aware that if you want something valuable to come out of any screen, you need to pay; otherwise, you are only going to get ads, noise, and a bad experience.

Even for the most famous dot-com flops (think Pets.com or WebVan), the premise of the idea wasn’t wrong. Pets.com delivered pet supplies to retailers, something that widely exists today and is a valid, no-nonsense business idea. Similarly, WebVan made home deliveries for stuff within 30-minute windows, a market that bloomed 20 years later during the COVID-19 lockdown, with services like Uber Eats or Amazon Prime. Thus, it is not the business idea or the technology itself that leads to a ‘bubble’, but rather its execution and how mature the market is for such an idea. So, my take on tech-driven bubbles would generally be that the technology itself usually has value, but it is most probably too early, and the world is immature for all this potential to be implemented.

💸 Are all those tech billionaires just fools?

So, if the market is rather immature for AI, why do all those companies keep investing so heavily in it? Don’t they know any better? What do they expect to happen other than an AI market crash?

The short answer is that they are all just expecting the computing power to get cheaper.

In the past, faster and cheaper computer chips have solved a lot of tech riddles that were thought to be unsolvable. Looking back to the first thoughts of humanity about AI, people were wondering if they could build a program to beat humans in chess. Funny how chess was considered an activity requiring very high human intelligence, so naturally, it would serve as a good measure of an intelligent machine. A machine that is good at chess couldn’t be but an intelligent machine. Nevertheless, programs beating humans at chess were eventually made possible, not by discovering some fancy algorithm, but by implementing the same computing algorithms with increased computing power. And this is exactly the type of thing everyone is awaiting to happen with AI.

In particular, AI at its current state is just too expensive. It doesn’t make sense financially because it costs a lot of money to run such models, and it remains to be seen who would be willing to pay that much. For instance, for OpenAI just to break even for its ChatGPT Plus plan, it would need to charge about $50 per month, which is more than double its current price. Thus, creating faster and cheaper chips would allow them to render the entire process of developing and operating such models more affordable, allowing those companies to finally break even. In other words, hardware progress is the lever that could make AI sustainable.

I can’t help but mention Moore’s law at this point, cliché as it may be. Moore’s law is the observation that the number of transistors on an integrated circuit tends to double approximately every two years, leading to exponential improvements in computing power, efficiency, and cost-effectiveness over time. Even though classic Moore’s law is now kind of over (progress has plateaued in the last decade), the underlying mindset it has created (the expectation that computation will keep getting dramatically cheaper) still shapes the expectations of today’s tech giants.

Replacing CPUs with GPUs allowed for parallel processing and removed the physical limitations that existed previously. With the rise of ever-larger GPU clusters, computing power is expected to keep growing for years to come. And this is why everyone is so bullish about this.

🤔 So, where is the bubble?

A few weeks ago, Michael Burry placed a $1 billion bearish bet on NVIDIA and Palantir by buying put options on both stocks. Michael Burry was the one who famously shorted the housing market back in 2007, and essentially, predicted the entire 2008 housing bubble and subsequent financial crisis. So, naturally, when he called out the alleged AI bubble and market crash, everyone held their breath for a minute. Later in November, he mysteriously closed his investment firm. What a bold way to declare that we are in a market top!

But again, the thing is that AI is fundamentally different than the housing bubble. There is only so much we can do with houses, but the potential upside of AI is truly unlimited. It is rather a question of when and not if, because AI is going to completely transform the world eventually. And unlike housing, its value is not arbitrary; AI is a sophisticated technology, a miracle of engineering, having inherent value tied to the labour of the great minds that worked to put it together. On the flipside, houses have been built in the same way for hundreds of years, with minor technological progress. It’s easy to dismiss AI as a speculative bubble, just because we are not equipped to understand it.

I think that it is more appropriate to compare AI with other technological booms related to laying some kind of infrastructure, like electricity, telephone, or the internet, even rail. Oh, but rail mania was a speculative bubble, wasn’t it? Well, was it? The technology itself wasn’t a bubble; laying a rail infrastructure was not speculative and had real, no-nonsense value. The thing is that other alternatives apart from rail eventually emerged, like automobiles, airplanes, or piping, condemning the rail to never fulfill its potential. In other words, the transportation need was ultimately shared among several different markets apart from rail. The bubbliness originated from assuming that the rail is going to be the only means of transport ever. Thus, a meaningful AI bubble would require some other, magical technology to substitute for part of its potential.

Anyways! Just like the internet, the main value of AI originates from whatever comes next and sits on top of it. Laying the electric grid allowed for a bunch of electric appliances sitting on top of it to be created later on; building the internet allowed for a bunch of apps like Google or Facebook to be created later on. Naturally, whoever lays the infrastructure for anything is going to be paying lots of money, because they essentially pay for whoever is coming next. Nevertheless, this doesn’t mean that the technology itself is not valuable, and does not have the potential to lead to further substantial growth.

That leads us to an important insight: for AI to make sense financially, a bunch of successful and meaningful things sitting on top of it need to be created. The investment in the infrastructure itself doesn’t make sense if it isn’t ultimately used as an infrastructure. So, this is the real bet, the real question one needs to answer. The technology is there, the investment is there, the infrastructure is there; now it remains to see if enough truly useful stuff is built on top of it.

On my mind

I know this post might age like milk. 🤷‍♀️ Maybe in a couple of weeks or months, I will be reading it and wondering what on earth I was talking about. Still, AI is genuinely impressive: a sophisticated technology with the potential to transform the world permanently, much like electricity or the internet, and to set the stage for an entirely new era of technological growth.

Loved this post? Let’s be friends! Join me on:

📰Substack 💌 Medium 💼LinkedIn ☕Buy me a coffee!

What about pialgorithms?

Looking to bring the power of RAG into your organization?

pialgorithms can do it for you 👉 book a demo today