00:00It is remarkable of how much this company is growing.
00:02Is it just this idea that we continue to invest in the bottlenecks and chips and semis is a clear
00:07one?
00:07Yeah, and when you look at the gross margins of a company like Micron,
00:12I mean, we've gone from negative gross margins to 81% gross margins in a span of few quarters.
00:19That just goes to show how tight that supply is.
00:22And even though, you know, the CEO was talking about adding more capacity,
00:26none of it will be enough to really, you know, supply enough for all these LLMs and inferencing.
00:33And so from that perspective, I think the agentic AI shift that we have seen over the past few months
00:40has been a real driver for memory.
00:42That wasn't the case when, you know, we were in that training of LLM phase.
00:46So that shift from training to now agentic AI has been a huge driver.
00:52And everyone is trying to model the price increases.
00:55But, you know, they keep exceeding estimates.
00:58And which is why you're seeing, you know, price target revisions and just the margins continue to expand here.
01:04I was looking at the EE page.
01:07So you can look at a nice summary of earnings estimates on the Bloomberg terminal.
01:11It shows, for one thing, that the forward 12-month price earnings estimate is still less than 8.
01:18But for another thing, it shows, you know, on a list of, here it is, on a list of different
01:25lines.
01:26So EPS, adjusted revenue, net income, adjusted operating profit, EBITDA, how many analysts got it right.
01:33Almost none of them were above what was published.
01:37Are they going to do, are the analysts going to do, Mandip, are your colleagues going to do a better
01:42job this time?
01:43I mean, just think of, you know, what we are talking about here.
01:47These companies used to have 15%, 16% operating margin.
01:51Right now, they are at four times, almost, you know, 65% to 70% operating margin.
01:57And I think even though you try to, you know, extrapolate the trends and you don't try to box yourself
02:04into linear thinking, it's very hard to see that, you know, huge jump, 4X jump in margins, especially when you're
02:11trying to model ahead.
02:13And so from that perspective, it is sort of a gradual buildup in terms of, okay, they have pricing power.
02:21We know they've been raising prices, but of this magnitude and for this long of a cycle, which is why
02:28I think this will exceed everyone's expectations in terms of the length of the cycle and how much room there
02:35is to grow.
02:37And look, the capacity additions will catch up eventually, but for now, we know it's a very tight market.
02:44When does the cost of inference become a real problem?
02:47You've seen it selectively for a few companies like Uber.
02:49I was also looking at a story this morning that DeepSeek has cut their pricing by something like 75%
02:54for their current model.
02:55When is that going to be the route that these U.S. frontier models try to do, that they realize
03:00that companies can't keep paying up or they won't have as much pricing power?
03:04Maybe they will forever have pricing power and they can command whatever they want for inference costs.
03:08No, so that's a function of that token demand.
03:10So right now, we are seeing tokens growing 7X and Google actually gave that number at their event.
03:16If that starts to come down and 7X is a 700% increase, so you really want to see that
03:22come down.
03:23And then it's a question of how good are the open source LLMs versus the proprietary LLMs?
03:29Do proprietary LLMs continue to command a pricing advantage?
03:33And so from that perspective, we've all these questions that will be answered over time.
03:38But for now, we know the demand when it comes to tokens is like off the charts.
03:44And that's why everyone is paying up.
Comments