00:00There's a lot of news when it comes to chips.
00:02Cerebris shares are lower.
00:04Amy just sharing that with us.
00:05SK Hynek seeking 45.45 trillion won.
00:09That's about $29.4 billion.
00:11In a U.S. listing, it's tapping investor demand
00:13for high-flying memory chip stocks,
00:15even after a major sell-off shook the group this week.
00:18Shares of Qualcomm are down more than 5%.
00:22The company says a data center will deliver billions in revenue
00:25in fiscal year 2027.
00:28It says two.
00:29Yeah, go ahead.
00:30No, go ahead, sorry.
00:31Oh, no, there's more.
00:33Drum roll, right?
00:34Yeah, go ahead.
00:34No, Micron.
00:35We're like coming out after the closing bell.
00:37And you have been reminding us that every guest
00:39that we've had, market guests, was it yesterday?
00:41All brought Micron up.
00:43Well, Cameron Christ writing today that by his calculation,
00:46it's now the third most important stock in the world,
00:48trailing just NVIDIA and Alphabet.
00:50Let's see what Mandeep Singh has to say about that.
00:52He's Bloomberg Intelligence Global Head
00:53of Technology Research here
00:54in our Bloomberg Interactive Broker Studio.
00:57Do you agree?
00:58I mean, right now, I think we are in an environment
01:02where memory shortage will continue to elevate these companies.
01:08They have an oligopoly.
01:09They haven't really increased the capacity.
01:13And to me, the parallels with the oil oligopoly
01:17and memory oligopoly are, you know, quite similar in that sense.
01:21Like, you know, oil prices go up.
01:24It leads to demand destruction.
01:27Memory prices going up.
01:28Well, at some point, if you don't expand capacity,
01:31will lead to demand destruction.
01:33And that's why these company executives have to make a call.
01:36If they see the road to, you know, five to ten times more demand,
01:40why are they not increasing their CapEx and their capacity?
01:44And I think tonight we'll hear from Micron if they're actually raising their own CapEx
01:49to increase capacity.
01:50And that'll be crucial for me.
01:53As I said, Cameron Kreiss wrote, by his calculation,
01:56it's the third most important stock in the world.
01:58There was a time when we talked about NVIDIA earnings being like a macroeconomic event
02:04on par with a Fed decision or a Fed meeting.
02:07I mean, that was a year ago that analysts were talking about that.
02:12But I asked Mike Cantopoulos over at Janice Henderson the same question earlier.
02:16Can what we hear from Micron today make or break this cycle?
02:21No.
02:22He said the same thing.
02:23At the end of the day, it's your LLM companies that control the frontier.
02:28I mean, yes, NVIDIA.
02:29But they're the ones who are driving the demand.
02:32They are the ones who are driving the demand.
02:34And, you know, they are looking to raise as much capital as they can in terms of adding
02:42power, adding data center capacity.
02:44And really, you know, with Mithos and some of the other models that have been talked about,
02:51what they're saying is we have reached a frontier where if we had, let's say, 10 times more compute,
02:58we'll be able to scale our revenue by 10x without even doing anything else.
03:03And that's the sort of confidence these companies have.
03:06Who drives that revenue?
03:07Who is buying those services, especially at a time when we're increasingly hearing from companies
03:12that they're not necessarily seeing the ROI?
03:15Well, I think the biggest data point is Anthropics ARR growing from a billion-dollar run rate
03:22to close to a $50 billion run rate over a span of last 18 months.
03:26I didn't even get my head around that.
03:27We have never seen.
03:27You may argue the margins are, you know, low, and yes, they are burning a lot of cash.
03:34But within the technology realm, we have never come across a company go from a billion to
03:40$50 billion in 18 months.
03:42It's never happened.
03:44Help me understand, though.
03:45So Micron's about a $1.1 billion market cap company.
03:48I look at NVIDIA, which is about $4.7 trillion.
03:52$1.1 trillion.
03:53Oh, did I say $1.1?
03:55No.
03:56Yeah, $1.1 trillion.
03:57Oh, $1.1 trillion.
03:58Forgive me.
03:58Sorry.
03:59NVIDIA is about $4.7 trillion.
04:01Sorry, Ms. Red.
04:02And then you've got Cerebrus, which is, what, about $40.7 billion.
04:07Like, help me understand Micron's role in this, in the AI build, in terms of what they do.
04:13So there has been a big shift from training to inferencing over the past 12 months.
04:19And I would say when these LLMs were in training mode, they needed memory, but it was more about
04:27setting up the largest cluster of NVIDIA GPUs with adequate memory.
04:32With the shift to inferencing, the multiplier effect for memory has been, for every GPU,
04:39I need 10 times more memory than I needed for training phase.
04:43So suddenly everyone has realized, gosh, if we need more context for these LLMs,
04:49what is the one thing you need?
04:51It's memory and more storage, which is why all these names have run up so much.
04:55Including Micron.
04:56Including Micron.
04:58And even the storage names.
04:59I mean, look at Seagate, Westrindage.
05:01All these names are up, I think, a good 3,000, 4,000%.
05:06It's kind of crazy.
05:07So the correction that we have been seeing as of late in them makes sense?
05:10Is it just the nervousness about the AI trade?
05:14Does that make sense to you?
05:16Or do you think it's just...
05:17Well, I go back to the point that I made earlier that, you know, when you don't increase capacity,
05:22there comes a point when people are like, I know this technology is great, this tool is great,
05:28but the cost is too high for me to adopt it.
05:31And we may be hitting that point because the CFOs are getting concerned about the token budgets,
05:36and all of that has to do with the fact that typically when technology scales up, prices come down.
05:43That doesn't seem to be happening, and it's nowhere on the horizon.
05:46I don't think I can make a forecast for the next 6 to 12 months where prices for memory or
05:52storage will come down.
05:53These used to be commodities.
05:54So that's where the assumptions have to be reset.
05:58Can I just ask you, we talk about in these semi-cycles that when you have such demand,
06:03prices go up, people ramp up and start to build out and increase the supply.
06:08Is it not happening?
06:09It's not happening.
06:09Are they going slow for a reason?
06:11Just because they have scars from the prior cycle.
06:14So you ask the CEO of Micron and all these guys, they will tell you it's a cyclical industry.
06:20So we want to be measured with our supply expansion.
06:23It's just like the energy market, like you said.
06:25Yeah, before we run out of time, I want to just mention Qualcomm, down 5.6%.
06:29The company said Meta Platforms has agreed to use the chip maker's new data center processor and its infrastructure.
06:35It suggests that the company is gaining a foothold in the booming market for AI gear.
06:40Meta is going to use Qualcomm's Dragonfly C1000 chip and succeeding generations in its facilities.
06:45Qualcomm said the microprocessor will be available in 2028.
06:49What's your read on this?
06:49Why are shares lower, too, by 5.5%?
06:51I mean, look, everyone is looking to tap into the data center opportunity.
06:56We had this news from OpenAI doing their own chips.
07:00So that chip market is getting crowded in terms of announcements.
07:03And then you have a stock like Cerebra is dropping because, you know, they have OpenAI as a customer.
07:10And that OpenAI is doing their own chips.
07:12So it's not very clear a new entrant like Qualcomm or OpenAI can really extract a lot of revenue on
07:21the chip side.
07:22I mean, you still have NVIDIA guiding to a trillion dollar run rate, you know, over the next 24 months.
07:29So from that perspective, I think the established players still have a huge lead.
07:36And yes, it's a large, attractive market, which is why you have so many announcements.
07:40But Meta doesn't have the leading LLM.
07:43If Qualcomm had added Anthropic as a customer, I bet you the stock would have been up.
07:47So that's the other aspect of it.
07:49Qualcomm buy modular.
07:50Is this a good thing?
07:51The confirmation just got about 20 seconds.
07:53I mean, look, all of these announcements, I think the large addressable segment for me is still the data center
08:01opportunity.
08:01And that's where the established players have such a huge advantage, the likes of NVIDIA and Google TPUs, that everyone
08:08else may get a small pie.
08:10But to ramp up to that level that NVIDIA and Google TPUs are at, it's going to take a few
08:15versions of their chips to get there.
08:17That's so easy, right?
08:19Not so easy.
Comments