AI Bubble Pop or Not?

[JLC]

Not the link to the original article I read (behind a paywall). They argued that the reason Meta is selling its computing capabilities is because it over built. Trying to save/recoup costs by starting a cloud service. So if Meta overbuilt, how many other AI hyperscalers overbuilt? Have we reached the initial AI buildout saturation?

Meta pops 9% as company makes cloud push to sell excess AI compute power...

[mostlylong]

@JLC

So if Meta overbuilt, how many other AI hyperscalers overbuilt?

SpaceX/xAI, you know the AI company with a $22.7 trillion TAM in AI, that doesn’t actually do much AI.

Just like Meta, they are leasing out their data center capacity to actual AI companies with actual paying customers, Anthropic and Alphabet. (nothing wrong with being a data center REIT, it’s just not AI per se)

Similar takes:

@GDavenport

if true, imply the over capacity is ALREADY here at the (BLEEDING) capability edge. When overcapacity arrives at the scaling edge (users x average consumption), the AI build out craze will be well and truly over.

@mostlylong

• Again because xAI can’t do very useful actual AI, they are leasing their data center capacity to more capable AI companies Anthropic and Google. (nothing wrong with that, it’s just not AI, it’s leasing real estate and hardware, like a data center REIT).

@GDavenport

There is excess compute for all inferior solutions. I suspect we will continue to see bare capacity (like the hyperscaler wannabe’s -NBIS and for those who have no way to leverage that infrastructure for profit - META, SPCX, etc.) sold to the winners to monetize.

[pauleckler]

Building excess capacity before you need it is common in a growing business. Often it’s easiest to build in standard sizes. Renting out surplus capacity is a good idea. It maybe develops a new business opportunity. Not much risk unless it drives down competitive pricing. Many players continue to say they need more capacity.

[Goofyhoofy]

@pauleckler

Not much risk unless it drives down competitive pricing. Many players continue to say they need more capacity.

Actually this is two of the biggest saying “not only do we not need more capacity, we have more than we need.” That signals to me that they are going minimally into the “AI Business” and more into the “picks and shovels for AI” business. Nothing wrong with that, except it doesn’t scale the same way. Or get paid the same way. Or have any of the attributes of a high-leverage market. It’s more a “we buy hardware and rent it to people” market. Like AWS or Google Cloud, but also like Hertz or Xerox in the old days. They’re buying depreciating hardware and hoping to recoup (+) costs and expenses before the hardware goes bad. Has worked fabulously for AWS, and others have watched and are now piling in for the Cloud customers: Oracle, Google, Microsoft, etc. Will it also be great, great, really great for AI companies which do not yet have profitable businesses? Time will tell.

[gcr2016]

@JLC

They argued that the reason Meta is selling its computing capabilities is because it over built. Trying to save/recoup costs by starting a cloud service. So if Meta overbuilt, how many other AI hyperscalers overbuilt? Have we reached the initial AI buildout saturation?

Jamin Ball’s newsletter discussed this. Here’s a snippet of what he said. You can access his newsletter and get the full commentary.

I think there are two big questions hitting AI markets right now:

1. If SpaceX (via xAI) and Meta are all of a sudden turning around and renting out their compute capacity, are we really in a compute crunch or is the system starting to fill up with excess?

   I don’t think this bear thesis has legs for a few reasons. First, let’s look at some of the details of these deals. Meta hasn’t actually sold anything yet. On SpaceX (xAI), all of their deals are VERY short term in nature. Each side (both seller and buyer) have a 90 day “out” where at any point in the contract they can tap out and stop paying (or providing the compute) within 90 days. The price SpaceX was able to charge was also extraordinarily high. I believe SpaceX structured the deals this way because they plan on using the compute themselves in the future. And the buyers (Anthropic, Google, Reflection) were willing to pay such high prices because they couldn’t get the capacity anywhere else.

[Goofyhoofy]

@gcr2016

1. I believe SpaceX structured the deals this way because they plan on using the compute themselves in the future.

In fairness, every AI company is thinking “I will sell more of this in the future.” I would be willing to bet that every AI company has enough capacity to sell to the customers they already have at the moment and perhaps a bit more. (I am reminded of the telecoms who said “we’re gonna need a ton of new fiber in the future” and who overbuilt to such a degree that several went upside down before that capacity-need appeared more than a decade later.)

This “holding action” could be a harbinger of “too much capacity too soon”, if things don’t go as rapidly as the evangelists think.

That’s why I say this “rental to another company” business is low leverage. (High leverage: a radio or TV station, once you have built it your costs are essentially the same whether you have 10 viewers or 10 million. Low leverage: a car plant where every car requires more metal, rubber, glass, assembly time, etc.) It seems to me the ‘rental of spare capacity” is pretty low on that scale, but it’s possible I don’t understand the internals of that business. It seems to be: buy equipment, sell for one price to another outfit, have no moat, be in a price race to the bottom with anyone else in that arena, chips wear out in 5 years, rinse, repeat.

Amazon created AWS not because of “excess capacity” but because they saw a business opportunity to let others use/rent tools they had developed. They turned their infrastructure/knowledge into something sellable/scalable. Maybe it’s me, but I don’t see that in the SpaceX or other “rental” deal.

[Leap1]

@JLC

Not the link to the original article I read (behind a paywall). They argued that the reason Meta is selling its computing capabilities is because it over built. Trying to save/recoup costs by starting a cloud service. So if Meta overbuilt, how many other AI hyperscalers overbuilt? Have we reached the initial AI buildout saturation?

There are logistical geniuses, and there are reporters. I remember the programmers in the 1990s discussing Bill Gates. It was his fault they were millionaires. Now it is Bezos who is bad to the little guys. Then it is Zuckerberg. You’d think these guys had no one working on things.

The thing the little guys hate are made up stories, but that does not stop them from making up stories.

@Goofyhoofy

I would be willing to bet that every AI company has enough capacity to sell to the customers they already have at the moment and perhaps a bit more.

Regardless of potentially crazy IPOs, OpenAI and Anthropic do not. They are the little guys who can barely keep up. Claude just added a “Quick Answer” button to try and respond to Chat’s speed. ChatGPT’s speed is a huge overkill, but only on one side of the dynamic.