Beth Kindig

There are many bottlenecks, Caroline.

Networking is another one that comes to mind.

Of course, energy we just spoke about.

Just keep in mind that it's not just GPUs, but there's almost an equal amount of money being spent on all of the components, all of the networking, all of the power, and we are no longer compute-constrained.

We're no longer GPU-constrained.

It really is becoming a power problem and a networking problem that needs to be solved, especially as we move into the inference market, which is only going to require more memory, more bandwidth, less latency, and more and more energy.

No, I do not believe that will be the outcome.

I think AMD is a strong contender, a strong number two, especially in the second half of this year.

They've really been preparing to bring the rack scale architecture to market, which is called Helios.

And Ed, what was...

I've covered AMD very thoroughly and we could get into the specs all day long, but OpenAI giving them that six gigawatt deal is a massive nod.

I mean, that is a big nod that AMD is a contender from the leading R&D firm.

Bloomberg Daybreak is your best way to get informed first thing in the morning, right in your podcast feed.

Hi, I'm Karen Moscow.