Many are wondering if there is an AI bubble. They see the valuations of certain companies and are baffled. Certainly, this is not sustainable. Revenues do not match the hype.
Then we have the money invested. Hundreds of billions are flowing in, coming from a number of different sources. We see governments, private equity, and public corporations dropping money in 10 figure denominations.
When will it pop is the second question. If we are seeing a bubble unlike any other, then surely there will come a time when the air is let out.
Of course, the correlation that people like to make is to the DotCom bubble. That was something that set the standard for the technological hyper versus reality.
The presumption is we are witnessing the second coming of this. Hype simply does not match the result.
Or do they?
In this article we will dive into this discussion.
The TL;DR is there is no AI bubble. We are looking at something much different from the DotCom era.
To start, most of the money is being spent on compute. This means we are looking at infrastructure. Here is where we are dealing with a different animal completely.
Jensen Huang has often discussed the idea of the AI factory. When we look at the most basic component, the output is units of cognition. In other words, the ability to think.
Naturally, we can debate the definition of think and what is meant by it. There is also the dispute of whether we are actually seeing something that "thinks". Regardless, few attributed the same barometer to calculation, something humans have to think about yet computers seem to do naturally.
Basically, we are looking at trillions of dollars that will be spent, globally, to build the infrastructure for the new economy. This is unlike what took place with the DotCom era where that term was used yet nothing really changed, at least for the first couple decades.
That is no longer the case.
If we break down the infrastructure, we can see how this money is being utilized. Here are the specifics for a data center that OpenAI is building in Abilene, Texas.
Enough electricity to serve the population of Seattle1
More than 250× the computing power of the supercomputer that trained GPT-42
A plot of land larger than 450 soccer fields3
$32 billion in construction and IT equipment costs
A few thousand construction workers4
Around two years for construction5
As we can see, these are massive investments. If we consider what happens each time we prompt ChatGPT (or any other chatbot) we can see how massive amounts of GPUs are needed.
Why are companies doing this? They know there is a need. Consider the fact that it is some of the largest Big Tech names behind this push. We know they are well aware of their compute capacity (limitation).
How can something be a bubble when the growth rate is expected to explode?
Here is the GPU forecast for the next two years:
On current trends, the largest AI training run in two years will need around 2.5 million H100 equivalent GPUs.23 At the same time, we estimate that OpenAI/Microsoft’s Fairwater data center will have double this compute capacity. So there’s enough compute to keep scaling training in a single data center.
The largest data centers will have 2.5 million H100 equivalents. To provide a contrast, xAI Colossus is roughly 230K. That is a 10x from its present size.
Consider the same multiple going across the entire industry.
A part of the bubble talk is the fact that we see circular financing happening. To many, this would seem like a Ponzi scheme.
On the surface this makes some sense. However, it loses its validity when we dig deeper into the numbers.
Nvidia made news with its announcement that it was investing $100 billion into OpenAI. It was a deal which locked the company into the purchase of Nvidia chips. Of course, there aren't too many options out there so this wasn't groundbreaking.
What we have to keep in mind is that Nvidia is going to generate $100 billion in profit this year alone. That is at the present levels of revenues. As we stated, the demand for GPUs is only increasing. Nvidia is the leading player in this field. In other words, that $100 billion profit will have a multiple increase over the next couple years if the company can scale.
There is no demand issues for Nvidia.
Essentially, we have big numbers coming from some huge companies. Meta and Google are two other corporations that are placing huge bets on AI. The key is to realize they have users that enter the billions. AI features are being added on a weekly basis to the different platforms. Each of these uses compute.
The question becomes will people use more compute or less going forward. OpenAI reportedly has 700 million users. Do you think the number of prompts is going to decrease, especially as the company brings out more products?
We all know the answer to this. Whereas one company might fall out of favor, the totality of the industry is only going to increase. For this reason, there is no bubble. We are going to see a massive shift in the economy in the next 24-36 months.
