NVIDIA is the American open source champion. ... They can release GLM 5.2 or better whenever they want.

everything that cuts against training works for inference. ... I do think distributed inference, distributed inference clouds are coming.

CXMT is going public in China. They're going to— they may be the cure for Apple's ills. They will flood the market with, to some degree, cheap consumer-grade DRAM.

I think Anthropic is worth $3 trillion today.

DRAM is the most important bottleneck. There's a whole segment of people on X who are very focused on bottleneck— bottlenecks. I call them the bottleneck bros. ... The bottleneck that matters is DRAM, and DRAM and HBM DRAM. This is the most important bottleneck simply because memory capacity and bandwidth are foundational to the performance of every AI model.

And I think they're on this, like, you know, they have this crazy belief that, oh, you know, our own internal chips are good enough. They're not.

Yeah, I do think the bottleneck grows, as they are called, are this like bottleneck trade is kind of nearing its end.

And if you have all these points of presence and you can deliver really low latency tokens to high value users, I think what's interesting for all these CDNs is they're getting a massive premium for that latency. It turns out one of the lessons from Cerebris is people are willing to pay for speed.