But what more do you need beyond that?

Yeah.

See, high bandwidth memory solves a very critical bottleneck in the AI, uh, ecosystem, right? Which is how do you, um, provide, um, maximum bandwidth, uh, to the compute layer so you could actually respond to queries very quickly, get, uh, memory, uh, capacity as well, uh, supplied to the compute layer.

But the reality is that, um, HBM cannot be, uh, um, a cure-all for all the issues that we are seeing, uh, in the AI systems, uh, space, uh, primarily because of a few bottlenecks.

One is the cost.

I mean, HBM is very expensive.

Um, if you look at, uh, the HBM cost all in including advanced packaging, CoWoS, uh, and, and what it takes for testing, uh, and compare that to something like DDR5, there is about a three to four x, uh, cost difference between HBM and DDR.

Was there, uh, an aha moment for you that kind of pushed you to thinking this was gonna be a really, really big thing?