Transcript

Why don't we start on that last point. So Alphabet and Meta obviously the most recent to put out bonds to finance data centers. According to Morningstar, those bonds, right they're priced 50 to 80, or yielding, excuse me, 50 to 80 basis points higher than Treasuries. What does that say about investor sentiment, about these bonds, about the fact that these companies are issuing debt? Well I think that the big hyperscalers people believe they're going to get paid back and they should. But I do think the spread widening, I do think the rise in the credit default swap cost to ensure against an Oracle default that has spiked, I think it's just highlighting the hugely capital intensive nature of these businesses. I mean the CapEx numbers that these companies are employing right now are just unbelievable. Oracle is spending 52% of their revenue on CapEx. It was 10% before this build out began. Meta is spending 35% of their revenue versus 15% before this happened. Google, Microsoft, and yeah these companies are, had phenomenal based businesses but these are no longer the asset light hugely free cash flow generated businesses that they once were. And then you mentioned CoreWeave and a lot of the other Neoclouds and other companies that are sort of providing the picks and shovels, and just the cost is just head scratching. And now obviously debt taking on is needed to finance it all. You know Peter, you and I are actually, I think we're thinking the same thing separately, we didn't even talk but you're looking at revenues, you're also looking at cash flow. So bank of America in recent days they came out with a chart looking at Meta, Oracle and the three hyperscalers, Amazon, Microsoft and Google and their cash flow compared to their capex. And if you take buybacks and you take dividends out of there, according to the estimates they're pretty close to spending all their operating cash flow if you do capex dividends and buybacks. So I think the question a lot of people are trying to figure out and you're really talking to is this sustainable? Can they keep spending like this? Well they'll keep spending until they stop. It's really hard to say. I do think though the stock reaction to the, the Meta announcement about what they're spending. Meta had a phenomenal earnings report but the capex number is obviously what concerned investors. Mark is going to spend what he wants to spend. But if a few quarters take place and investors really start to push back on this, that may be enough for CEOs to say, "Okay, we hear you, let's maybe not build that extra building." Because the beautiful thing about technology over the history of the world is that it's gotten more powerful and it's gotten smaller. So our iPhone can do what a mainframe did many years ago. So yeah, Meta is building a 4 million square foot facility, but who's to say in three years that facility can produce even more compute power, but they only need 2 million square feet instead of 4. And I think that is the risk with this build out is you can be sure we're going to build one too many data centers. Well, that's a lot of certainty from you. So very quickly, I don't know if you saw this recent report. According, to JPMorgan, as $5 trillion cost needs are going to kind of hit every debt market. Is there a certain point or is there a certain way that this becomes concerning? Right now you seem to be, you know, in the middle of the road about it. Maybe it's a warning, maybe it's not. But is there a part that becomes concern? Is there a certain type of financing that you, if you see hyperscalers have to tap into that, you're like, wait a minute, this may be too far. When the business impact, when the cash generating returns on this is not enough to finance this, all that's when this thing breaks. When does that happen though? I mean, do you have a point in your mind? Frank, I don't. I do think that we're going to watch Wall Street's response to how these, to these stocks and see how long Wall Street will tolerate all this spend because Wall Street will send a message when they've had enough. And also companies will say, are they generating enough earnings, with respect to selling these models? And one last point, this Chinese are producing open source models too. They're giving away for free. So I'm not so sure that selling these LLM Models to the rest of us, is going to be monetizable if many people are giving it away for free. This is becoming a commoditized product as graded as it is. It's becoming commoditized where the user is going to be the beneficiary, not the builders of this infrastructure.