$NVDA · InvestAnswers · 2026-10-30

Jensen saying, you guys are amazing doing it in a year for Colossus one and Colossus 2, right?

So then he's very confident that he will have his allocation from Jensen where Jensen said, "Oh yeah, 1.1 1 million chips are coming in for sure and then his ultras."

Second of all, Jensen knows, you know, he wants his products to go into operation as quickly as possible to showcase them. He knows there's no better person to turn them on than Elon,

I think we're in the early Nvidia after chat uh chat GPT phase where there you're think I'm going to almost double every quarter and then you you know actually do double and then you're constantly beating and it's very predictable because I'm pumping in this building I'm dumping in the chips I'm doing the deals two months ahead of time.

Nvidia margins right then because the demand stays high then you know the profit grows faster you get a quick two

you know the um Nvidia starts doing the um the modular DSX which installed in 14 days that still have to scale up but they start doing it and in Q4 things start speeding up, right?

let's talk about Reuben's for a second because we have dark chips. We have a power bottleneck, power constraint.

talk to us about the GB300's versus the Reubins and how much more power efficient they are and how much more token productive they are and how that actually multiplies with a view to understanding that Elon Musk has all the supply locked up.

So um the the Reuben ships use 1.5 times like one Reuben ship will use 1.5 times more power than a GB300. So then the whole data center would use 1.5 times more power but you're getting depending upon your use case three to six as much as 10 times but you know like more conservatively three to six times more inference than a GP300.

So that means instead of $26 billion the $50 billion per gigawatt you could be making $150 billion per gigawatt from the inference of of the the Reuben um if you can fully monetize um and then Ruben Ultra even more so it improves the economics a lot

I've been more so each data center would be um smaller like it be going to about 600 billion so I can only make 1.5 5 mini hard data centers for a gigawatt versus 2.5 for that.

So it's it's I think a tripling of revenue is a reasonable conservative number um when I go to the higher inference for for Ruben.

So going to Ruben though is more money. Every time you go to a better chip you get more out of it. Yeah. So going to Ruben ultra be even better.

And the thing is because you build faster they will be the first ones at scale with Rubin. So then if if I'm if I have um three or four Ruben data centers like the move from um H100 hoppers to Blackwell the Blackwell thing besides more inference it also enabled the training of Fable and the um the better models the Astro models.

So it's not optional for the frontier guys not to move to the the Reuben side.

So, Reuben is valuable for more inference, but also it's more valuable because it will enable better models to be made.