Oracle Triggered the Implosion of the AI Bubble
medium.com
medium.com
As best as I can understand it, DC construction is handled by REITs and SPVs, who require certain guarantees from their anchor tenants and take GPUs and possibly stock in the tenant as collateral in case things go bad. Oracle is the tenant for Stargate and maybe other DCs, and OpenAI is their customer, and OpenAI would ultimately pay Oracle rent, just like any other Oracle Cloud customer. Since OpenAI doesn't have any real cash right now, Softbank and some others are fronting them some money so that Oracle can seem more "credible" to the DC builders and in doing so command a better deal from the latter.
The assertions seem to be that all this falls apart if the DC doesn't get built on time. Oracle stock takes a beating, they no longer look so attractive to the DC builder, the builder walks away, the end. Is that about it?
It's a really weird analogy.
They use it as an analogy with a supernova:
> And now this massive star is about to blow.
> (...) it runs out of the fuel that keeps the monster from collapsing in on itself under its own insanely powerful gravity.
> An imploding star crushes itself into a tiny ball, and the rebound instantly triggers a colossal explosion that lights up the entire galaxy. A bursting AI bubble threatens the entire American economy.
> The gravity inside the bubble is investor expectations.
> In a physics thought experiment, all it takes is tossing a pinch of iron into the furnace of a massive star teetering on the edge of explosion to trigger an instant implosion.
The last one isn't even wrong, since iron is the last element a star can fuse (iron core collapse)
No, I won’t. End of the story.
Edit: Brave in reading mode also works.
I am glad that I sold all of my Oracle stock.
They plan to pay for most of it through excess cash flow, aka profits.
Follow on question: "they make $100s of billions per year? what did they do with all that previously?"
Answer: share buybacks, nothing (aka put it in the bank), acquisitions.
These companies make a ton of money and have a ton of margin over expenses.
If you doubt me they have openly stated this multiple times in their think tanks.
Is this meant to make it easier to understand or harder?
This seems like a complicated way of saying that the AI bubble is hundreds of times bigger than the economy of one of the Baltic states.
Is the idea of a star hundreds of times bigger than the sun more "normal" than the idea of an economy hundreds of times bigger than one of the Baltic states?
> And now this massive star is about to blow.
> (...) it runs out of the fuel that keeps the monster from collapsing in on itself under its own insanely powerful gravity.
> An imploding star crushes itself into a tiny ball, and the rebound instantly triggers a colossal explosion that lights up the entire galaxy. A bursting AI bubble threatens the entire American economy.
> The gravity inside the bubble is investor expectations.
> In a physics thought experiment, all it takes is tossing a pinch of iron into the furnace of a massive star teetering on the edge of explosion to trigger an instant implosion.
I guess the last one isn't even wrong, since iron is the last element a star can fuse (iron core collapse)
Oracle just made a step towards existing less, reverberating a negative wave through the entire system.
I don't like the state of affairs either, this deluge of crash predictions are probably going to increase the irrationality of the market and make the bubble worse.
But brushing aside the irrationality for a moment, another problem is that there's not a lot of agreement about what constitutes the bubble popping. I think a lot of normal people would generally call it "the day the NASDAQ drops 20% or something like that", or maybe the week, and if that hasn't happened yet then it isn't "popped". There's two problems with that, though, the first of which is that there isn't guaranteed to be a huge down day or week at all, and the second is that such an event is usually a hugely lagging indicator. It is, one might say, the end of the end, but there is usually a beginning of the end before that.
It's kind of funny... sometimes the stock market is living so far in the future it boggles the mind, but sometimes it is the last thing to know, as everyone simultaneously banks on being able to get out before everyone else and grabbing every last penny from the money fountain while they can, even though they may all individually know the party is ending.
Or it could be massive -- it could be the straw that breaks the camel's back in a stock market that seems massively over-inflated given recent interest rate rises.
Until there is more visibility into the numbers, it is hard to predict if or when the bubble pops. He said keep an eye on Anthropic and OpenAI, when one of them gets into trouble, things will get crazy. At the moment Oracle is terrified of being reduced to junk status. They want to avoid that at any cost. Anyone thinking to invest in any of these two companies should realize much of their growth has already has occurred. This won't be AMZN, DELL, etc, etc, type of growth.
He did mention what would constitute an end of the world situation, something like JPM going under. That is an end of the world situation.....
And I feel fine~
So, people stop investing in AI, the bubble bursts. Who knew :)
> First, their margins. They’re nowhere near enough to cover the cost of building models, running operations, and paying back investment commitments.
This is the same kind of thing that has been said about every growth company until they stop investing in growth and turn on profits. Nobody investing in AI companies want AI companies to be profitable right now.
> Second, as it turns out, businesses aren’t exactly eager to switch from older models to newer ones. The old models cover most of their needs, and switching is too complex and too risky to justify.
This just offered without evidence, and there is no reason at all to believe this is true.
> Third, there’s growing competition from open-weight models. The problem is obvious: why pay for expensive proprietary models when you can get by with much cheaper ones? This way, you control not just what you spend on the model but the model itself. You can run it on your own server, and nobody on “the other side” of the API can change a thing.
Two problems with this argument -- First, there are some tasks that cheaper models _cannot do_ and that newer models do easily. Second, you will not be able to run models locally cheaper than the big labs can run them at scale. If open weight models commoditize, OpenAI and Anthropic will _both_ be able to run them cheaper than almost any competitor, or than you can running them in your own data center.