Lucas Swisher

Look at Snowflake.

It's got 20, you know?

Margin matters.

But early, it can be a misleading indicator, especially when an architecture shift is happening.

The reason why in AI, and I'll give you the bull case on this, right, is...

The reason why margin might not matter early on in a company's life in AI is the cost curve is coming down so fast.

Say my inference margin is 10% today.

It may have been negative a quarter ago and super negative two quarters ago, but the token costs are coming down so fast.

Maybe I'll be, if I'm an application AI company, I'll probably be able to develop my own model for some of the workloads.

I'll probably want to use frontier models for some of the workloads.

I'll probably want to use really small, cheap models for some of the workloads.

And over time, I'll be able to optimize my margin.

That's what we really believe is going to happen over time.

But listen, these companies are structurally lower margin than the last generation because you pay the cloud and you pay the LLM.

Well, from, I think, gross margin, yes.

But what you might say is, hey, I'm actually substituting a lower gross margin for lower OPEX because my engineering team, maybe it's more efficient.

My sales team is using AI tools now.

Maybe it's more efficient.

My legal team, maybe it's smaller and maybe I'm more efficient.

So your terminal operating margin may actually be higher in this world than the last world.