| ▲ | captainbland a day ago | |
I'm not sure, it seems like everyone and their mother has a decently capable TPU design at this point or is in the process of designing one. I think the moat isn't as big as it was for GPUs. | ||
| ▲ | eecc a day ago | parent [-] | |
Exactly, and that explains Jensen’s strategy. Nvidia’s got a time-limited (crumbling) moat and it needs to capitalize it to death before it goes away, replaced by competing stacks. If the US datacenter industry can’t absorb their max capacity, its only option is to throw its momentum towards building clouds on their own stack, wherever they can. Asia. Next step would’ve been boostrap their own AI lab on OSS models, and fork. All this, without getting caught “entering their clients’ businesses”, American clouds and American ai labs | ||