Chamath Palihapitiya

We took slightly different technical approaches.

The big difference that what Cerebra said initially from the start, and this is my friend Andrew Feldman, so I'll give him a shout out.

Very smart guy.

If you see a chip, what you actually see is a wafer.

And a wafer is full of little chips, right?

And you literally tear them off.

That's why they're called chips.

Andrew was like, no, I'm just going to make one ginormous chip the size of a wafer.

And so if you look at it, Nick, maybe you can find a picture of it.

These things are huge.

And at the time, people were like, I don't think they really understood what the power of what Cerebras was building and what Andrew was building.

And then the first folks that really got on board were the Emiratis, so MGX and G42.

And you started to run these models and what you started to see was, man, their inference is blazing fast.

Again, back to what we talked about last week, because when you have the compute and the memory in the same place, physical distances are now minimized and the complexity is minimized.

So you just have incredible speed.

What I can tell you is that

OpenAI is aggressively trying to diversify so that they have multiple paths of inference available.

They have a huge deal with AMD.

They have a huge deal with NVIDIA.

They now have a huge deal with Cerebrus.