Jordi Visser

If the bottom 60% are making up for the down five, that means these things are all up over 20%.

That is a change, and that's the point that you go through.

I think people have to adjust to the fact that we have reached

The constraints, the models will get better every six months as they have because the chip efficiency is growing rapidly.

Rubin was announced at CES this week by NVIDIA.

We have Blackwell rolling out this year.

Only about 20% of the chips inside the data centers are Blackwell.

They're far more efficient than the hopper.

All right.

So when you get to the point, Blackwell, let's say it goes from a 10 gigawatt, so 10 gigawatts of power to do the AI we're doing now.

That's not the actual number, but let's just say that's what it was.

So you're now going with Blackwell down to only five gigawatts necessary.

Cuts it in half.

Okay.

Rubin, which was announced, will take it down to about one.

So it's a 90% reduction.

2027, yeah.

So the reality is, even if we never build another data center other than the Blackwells that are rolling out this year, that percentage of the overall power usage will go from 20% to 30% to 40%.

Colossus 2, Musk's...

know next part of colossus is supposed to be done in the first quarter he talked about it on the on the interview that's a massive gain so that's why the models and you use the car let's let's go further back than a honda and let's say when you left horses and you started with cars and the first car is the hopper then you get to a porsche which is the rubin well the first car is going a max of 15 miles an hour and now the porsche is going at 150 miles an hour that's what you're talking about