Hacker Newsnew | past | comments | ask | show | jobs | submit | wmf's commentslogin

The next step would be forcing them to pay out FCF as dividends. When that doesn't work Bernie will come up with an even worse idea.

28nm chips is just about suitable to run frontier models

I doubt it. 28 nm is 4-5 generations back so inferencing would need a large number of chips with very high power consumption. Maybe you're thinking more of 7 nm which is what Chinese fabs have; it seems to be OK for companies like Huawei.

And so if I was a mid-level State would it be worth while to take my nascent chip industry and push it out to build a 28nm foundry and supporting eco-system.

It never reaches breakeven so you'd have to provide billions in subsidies per year forever. The sovereign chip stuff only makes sense for the US and China; even the EU probably isn't large enough to make it work. A single country definitely couldn't.


If GPUs have 75% margin then 50% cheaper is no surprise.

Operational costs far outweight hardware cost.

Do they? Genuinely ansking.

Yep, I was surprised to learn that too.

Don't forget Stargate.

Update: Somebody on Twitter said it's going to be hosted 50/50 at Microsoft and Oracle.


I forgot Stargate


I don't see any evidence of this. Open source software thrived on proprietary hardware for decades.

“People who are really serious about software should make their own hardware.” ― Alan Kay

Good models will require multiple Taalas chips but Groq and Cerebras also require a lot of chips and that hasn't stopped them.

I assume OpenAI has been buying memory and "giving" it to Nvidia in exchange for a discount.

It's probably the same speed but cheaper.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: