Fred Thiel

And as you start seeing an ability to abstract the model from the hardware layer,

So you can orchestrate AI workloads across the cheapest operating hardware wherever it's needed so that you get the quality of service on the delivery of your task within the timeframe you need.

you'll do that and that will drive costs down because this industry, the AI industry has already commoditized around this thing called cost of a token.

And more importantly, how much inference do you get per megawatt?

And so it's already commoditizing.

And so there's this whole price curve that people are gonna have to chase down.

And one word of warning from an investment perspective.

There is a huge obsolescence curve in AI that is a three-pronged or a three-line curve.

In Bitcoin mining, we have one curve and that's ASICs.

And the fact that we have a halving that drives us to use more and more efficient ASICs.

Those ASICs all plug into the same racks.

They use the same power and they use the same cooling.

Biggest complaint that NVIDIA customers have is, I put in H100s, now I'm going to put in Blackwells.

I have to retrofit all my racking, all my power, all my cooling.

So now think about a company that just signed a lease with a hyperscaler, where they are putting in the equipment, not the hyperscaler.

And in three years, that hyperscaler is going to say, you know what?

You need to rip that out and put the next generation of stuff in there.

The company hasn't even paid off this equipment.

So I think you're going to see a lot of blowups amongst people who have gone out and kind of are renting bare metal to the hyperscalers versus people who are building powered land and then leasing it to the hyperscaler and the hyperscaler takes all the capex risk.

Those are the long-term, the people who are in the real estate business are going to be the long-term winners because the people who are putting bare metal and renting bare metal have to bear this obsolescence cost and nobody can.