Jason Friedberg

And to your point, there is a ramification of it, which is it created a headwind against the already AI bubble, which was deflating after Brad Gerstner popped it.

Oh, there you go.

Conspiracy corner.

Let's get our tinfoil hats on.

The individual.

He was accusing also, he's got the Palantir short, but on the AI side, he was saying, hey, Meta and Oracle have been cooking the books with 176 billion in hidden depreciation to inflate earnings by over 20% in 2028.

And so here's the chart.

Basically, we've talked about what is the reasonable life of an H100 of

an NVIDIA chip or really an NVIDIA server, if you think of it that way.

Do these things have a three-year lifespan, a four, five, six-year lifespan?

CoreWeave might be putting them at six-year.

And that dramatically changes your expenses, which then, of course, given the scale of the data build-out, the data center build-out could radically change your earnings because

Dividing the cost of a data center that costs $100 billion by six or three is, you know, 2x.

So your thoughts, Freeburn?

Shamath, what do you think?

You're involved in, obviously, inference.

How long will the Grok inferences be out there doing jobs?

And it does seem like, to Freiburg's point and the gentleman in the video, that there are other jobs to do, right?

So as these things move down the life cycle, there'll be something for them to do.

When would you actually unplug them, I guess, is the question, and stop using them?