Ed Ludlow

I do think that there are enormous investments in accelerators, particularly at DeepMind, but there are other legitimate efforts as well.

And I think it all speaks to the overall demand in this market.

The reason people are making these really strong investments in other chips is because they're going to go spend billions and tens of billions and hundreds of billions of dollars, trillions if you listen to Sam,

against this compute over the next decade, right?

And so I think the question of winners is the problem today is just meeting the demand.

And all of these players are trying to come up with alternatives, less about cost and more about creating more supply.

I enjoyed reading about and learning about base 10.

There's a section on your website where you break down pricing.

And I don't know, that might seem really mundane to both of you, but you basically have pay as you go at the low end right through to enterprise level deals.

Explain that kind of strategy.

You know, we've discussed with Cursor the idea of per desk versus pay as you go.

It's actually a very interesting model.

And Sarah, what gives you the confidence that this doesn't get commoditized or unleashed?

How sticky are the customers when it comes to inference?

I think training is a job-by-job sort of buy, right?

Inference has the stickiest characteristics of any business I've ever seen, right?

And so, you know, without giving specifics, the NDR on base 10 is through the roof.

And we see customers, they come with us, they scale with us.

They don't want to take on this enormous engineering burden of, you know, these different specializations I described.

Yeah.