AI Bubble Crashing Its Way into AI Abundance?

Posted by hmichaelson24 10 hours ago

As we move into a world of maximum engagement over substance and hype over reality, it becomes more important than ever to sift through the noise and find the truth. This is especially true in today's AI industry. If you read Twitter, it sometimes seems like everyone believes AI is going to take over the world and everybody's jobs next year, or that the whole thing is just a giant bubble.

At this point, most people have probably seen some version of the basic AI bubble theory: AI spending is a massive bubble, frontier labs are spending aggressively ahead of IPOs, and eventually the market will crash and everyone will return to reality.

I'm exploring a different AI bubble scenario centered on the massive GPU and data center buildout of the past few years. I think the closest analogy may actually be the 2022 crypto crash, rather than the dot-com bubble or 2008.

The basic idea is a deflationary spiral that starts as a financing and liquidity problem near the top of the stack and propagates downward through the physical compute market. If GPU-backed infrastructure becomes distressed, forced selling could push GPU prices and rental rates down sharply, which could then create more distress elsewhere in the stack.

The more interesting question to me is what happens after that. What do the second- and third-order effects look like if GPUs temporarily become absurdly cheap?

Has anyone considered this possibility or know of good writing on the topic? I'd especially love to hear from people who have thought about GPU financing, data center economics, or the secondary GPU market.

Comments

Comment by WheelsAtLarge 8 hours ago

I think the goal is to reduce the cost of tokens by orders of magnitude similar to the reduction of price on the transistor. So this means that the token processor, a GPU or a replacement, will get much cheaper on a per token basis. So the cycle will continue, more powerful new workhorses and less costly older models. A GPU crash? No, a continued GPU cycle of more usefulness at a lower cost per output is what's coming.

Additionally, the other side effct that people are not talking about is the reduction of cost on a watt of power. We will use more energy per person but the cost per watt will be so much cheaper. The datacenter build out is making electricity more expensive now but over the longer term capitalism has a tendency to make things we use, specially things we use a lot, very cheap on a per unit basis.

Comment by hmichaelson24 4 hours ago

[flagged]

Comment by rvz 10 hours ago

Abundance for who and what of?

Mostly likely an abundance of security breaches, exploits and other terrible things.

Does that 'abundance' benefit humanity? We will get the negatives first before the positives unfortunately.

Comment by hmichaelson24 4 hours ago

[dead]