- 41 minutes ago
Category
🗞
NewsTranscript
00:00I'm joined by Bloomberg Tech and Industrial Policy reporter Maggie Eastland,
00:03Jess Menten, our Bloomberg Senior Equities reporter, is here with us in New York.
00:05Maggie, let me turn to you first, and I look at this statement from OpenAI on that incident,
00:10a new kind of security incident, unprecedented cyber incident.
00:14Talk a bit about what happened here, explain it to me as best you can,
00:18and how unforeseen was this, and how worried should we be about it?
00:22And why has no one watched the historical documentary, The Terminator?
00:26But that's fine, all right, it's a very Skynet moment, but please continue.
00:30Yeah, this does seem a little Skynet, but it's not quite there yet, thankfully.
00:35So what happened was OpenAI disclosed that two of its models, both the sole model,
00:42which is its most recent one, and actually a model that hasn't been released at all yet,
00:45that's still in development, kind of like got out of the box, so to speak.
00:49So they were testing their new model kind of in an air-gapped environment,
00:53and they asked the model to do something, to complete a security test.
00:57So instead of completing that test in the way they had envisioned,
01:01this AI model went a little rogue, and it actually kind of like hacked into the internet,
01:05which it wasn't supposed to have access to, and then it went to HuggingFace,
01:09which is a platform that a lot of software developers and AI researchers use regularly.
01:14So it was trying to pull some things from HuggingFace in order to do the test that OpenAI had asked
01:21it to do.
01:21So this is clearly an out-of-the-box situation, though it was responding to a directive,
01:27but clearly the AI models are starting to act in ways that are a bit unpredictable,
01:32and according to directions, they haven't exactly been given.
01:37Don't sleep on that innocent-looking emoji logo for HuggingFace.
01:43Here's the other question.
01:44Some people are saying, yes, they have an obligation to disclose this,
01:48but it's a little bit of a show-off as well.
01:51It functions a bit as an ad for their AI to say, look, we've built this thing.
01:55It's so powerful.
01:57Look what we can do.
01:58Is there a bit of braggadocia here,
02:00or is it mostly just they need to disclose it to tell people what's going on?
02:03Yeah, I think that the part that maybe doesn't get as much attention
02:10is that OpenAI did ask the model to complete the security test,
02:14and it was trying to do that when it hacked into HuggingFace.
02:19It was just doing it in a way that wasn't really what the company had intended.
02:24So I think it's sort of a mixed picture here, right?
02:27You could see models acting and trying to act on the directions they've been given,
02:32but just doing it in a way that's not what the user intended.
02:36So it's a bit nuanced here, and that's more the concern, at least at this point,
02:39rather than a full-blown kind of Skynet situation.
02:42All right, fine.
02:43All right, Jess Menden, all of this happening in the crucible of tech earnings season.
02:47So we heard from Alphabet.
02:49That company talked about its spend on AI infrastructure.
02:52I think if you look at those four major hyperscalers,
02:54it's something like $725 billion is what they're forecasted to be spending on this.
02:58How did our sense of that, our sense of the level of investment change
03:01as a result of what Alphabet reported,
03:03and what are we going to learn this week from these other companies?
03:05Well, the playbook has changed for these companies and specifically their stocks.
03:09Normally, your stock would be rewarded as a hyperscaler
03:11if you're spending more money on capital expenditures related to AI.
03:15But then what changed here?
03:16Well, a lot of that has to do with the macro.
03:18I mean, you all talk so much about oil prices spiking.
03:20Just saw that this past week, briefly above $100 for Brent prices.
03:24And then also, what does that mean for inflation and the path of rates?
03:28Because if these companies want to issue more stock issuance,
03:31but then if their stock is falling, that's a problem.
03:33So if you have to tap the debt markets, that's even more expensive
03:37than what their earnings outlook would have been at the start of this year.
03:40So then something has to reprice.
03:42But with that said, if you look at overall S&P 500 earning growth
03:44for the first half of the year,
03:46it's expected to track around 28% based on Bloomberg intelligence data.
03:49So for perspective, outside of the bounce backs from COVID
03:53or the great financial crisis, that would be the best growth since 2004.
03:57So normally, you wouldn't see that at this point in the economic cycle.
03:59So it's not to say that it's doom and gloom because earnings growth is one of the key pillars
04:03for corporate earnings and obviously how that ties to GDP growth.
04:07But now we're supposed to get an update for the first time on second quarter GDP growth
04:11at the end of this next week after the Fed meeting.
04:13Obviously, that's more backward looking, but the Atlanta Fed GDP now models a little bit below 2%.
04:17So what does that mean for corporate earnings in general
04:19if these are the stocks that have really powered that growth over the past few years?
04:23I was also reading that, you know, the tech selloffs look pretty scary,
04:26but in the end, it might be a better thing to stabilize parts of the sector.
04:30Can you explain to us why that might be true?
04:32Well, the big thing the past few years and something that's switched recently
04:35is typically if you talk to whatever moniker you want to use for these big tech giants,
04:39typically because of the concentration and record concentration they have
04:42in the S&P 500 as well as the NASDAQ,
04:45you would see it just completely bring down a lot of the indexes.
04:48But what's interesting, whether it's this past week or in recent weeks,
04:51you've seen a divergence between the MAG7 stocks
04:53and see what's happening in chip makers, specifically memory.
04:56So, of course, Apple's going to be reporting next week after the bell on Thursday.
05:00That's a different type of AI spender.
05:02It doesn't spend as much money related to AI in the same ways as the other companies.
05:06And it's also the best-performing stock out of the MAG7 this year.
05:09It's up over 20%.
05:10Tesla's the worst down.
05:11It's so funny because I feel like it wasn't the shiny new kid for a while,
05:14and now people are going back to it.
05:15But then also it had those issues with price hikes for those devices we have.
05:20Mark Gurman of Bloomberg reported about that a month ago.
05:22Its stock briefly got hit, but it's not spending the same amount on AI
05:27and related to CAPSACs as these other companies.
05:29But it's more unique when it comes to the memory components that's more expensive for it.
05:33So you see the counterparts in the chip space, whether it's a Micron, SanDisk.
05:36Those were the two best performers in the S&P 500 the first half of this year.
05:39But noted, since the stocks peaked last month,
05:42I mean, those have been hit by double digits and pretty hard.
05:44So even though you're seeing the hyperscalers pull back,
05:48you're still seeing the overall market, whether it's the S&P 500,
05:51still about 2% of it is all-time high.
05:53So you're seeing a divergence there in the concentration of the market.
05:57Maggie Eastland, we had our moonshot moment in recent days,
06:00and that was presaged, of course, by what happened with DeepSeek a while back.
06:04Bloomberg reporting now that DeepSeek is pausing this second fundraising round
06:07after some comments purportedly made by its chief executive.
06:10Explain what's going on here and what this says about the dynamic
06:13between the innovation and work that's happening in China
06:16vis-a-vis what's happening here in the United States when it comes to AI.
06:20So as you said, we're reporting that DeepSeek is delaying.
06:24There are plans for initial public offering a bit.
06:26I think bigger picture here, what we're seeing is even though this is delayed,
06:30we also saw with the release of Kimi K3, another Chinese model just a few weeks ago,
06:34and they're also looking to IPO potentially in the next year,
06:37is the Chinese companies are moving kind of at a pace that's similar to the U.S. companies.
06:43Obviously, OpenAI and Anthropic also have IPOs upcoming,
06:48but actually these Chinese competitors are proving to be a real risk for them
06:52just because the AI that they offer is very good, and it's also very cheap.
06:57So that's been a huge point of debate and kind of sending shockwaves through Washington
07:02over the last two weeks is just, you know, how good the Chinese competition is
07:07and sort of the question of whether those models were created in a way that was fair.
07:14Policymakers here are weighing.
07:16Yeah. Speaking of policymakers, I'm also wondering,
07:18as we see some of these jailbreaks from these systems
07:21and people start to talk a little bit more seriously about some AI regulation,
07:24what that means for the development as you get more competitive with the Chinese sectors
07:29and if these AI companies are concerned about regulations that could, you know,
07:33hem where they want to go with this technology.
07:35Yeah. I mean, some technologists have said point blank that, you know,
07:41this new regulatory framework that the U.S. companies have to follow
07:45might be hampering their development
07:47and putting the U.S. at disadvantage in the race with China.
07:51On the other hand, you have safety advocates who have been saying
07:53that we need to be having these tests for a long time.
07:55And when you have, you know, AI models kind of like hacking other websites,
07:59maybe they have a point.
08:01So right now there is some testing that is essentially required.
08:06It's nominally voluntary, but we've seen, you know,
08:09companies that don't comply be export controlled.
08:12So, you know, Sam's going to be here this week in Washington
08:15and he's talking about the next generation of models.
08:17No doubt he'll be getting questions about the hugging face hack.
08:20But also from his perspective, it's going to be important to make the case
08:24for, you know, releasing that next wave of models
08:27because OpenAI will try to get Washington's blessing for that.
08:30We'll see what the conversation is like
08:32around that kind of FINRA-esque regulatory model that's being proposed here.
08:35Jess, Matt, maybe wrapping up with you,
08:38we were talking there about the prospect of IPOs, yes, in China,
08:41but the prospect of them happening here in the U.S. as well.
08:44How does this environment that we're in right now,
08:46maybe a little more heightened skepticism about this investment,
08:49color the plans that these major companies have for IPOs in this space?
08:52We've seen some recent ones that's passed.
08:54Actually, in June, we obviously had the SpaceX IPO,
08:56and then there's been smaller ones where you're starting to see those stocks
08:59priced below, actually after they were already trading below,
09:02where they initially were supposed to price in that midpoint range.
09:05So that's something that's another issue
09:06when you think about the growth of the economy,
09:08but also Oracle will be reporting next week.
09:10And a big thing is the credit default swaps on a five-year basis.
09:13That's trading around multi-year highs.
09:15And that, according to Barclays, is more,
09:17what does that mean for the AI type of debt fears?
09:20So obviously not housed in the MAG-7,
09:22but another important key because that's also becoming more expensive
09:25for whether it's Amazon as well as Meta.
09:28And then Meta actually has that $12 billion data center in Texas
09:32that needs to be financed soon.
09:34So that's going to be another issue
09:35when you're thinking about the financing with that.
09:37So that's going to be all key of how all that ties together
09:39when you're looking at the credit side of things as well.
09:41I mean, I am starting to hear a bit more people say the quiet part out loud,
09:45and that's the word bubble.
09:46I mean, what's your take on that?
09:48Or is this as Mark Twain said, you know,
09:49the reports of my death have been greatly exaggerated.
09:52The reports of a bubble have been greatly exaggerated.
09:53I feel like because of the earnings that have backed up growth,
09:57whether you're talking about from a profit perspective,
10:00it's very different than the dot-com bubble fears
10:02because pets.com is always the famous example people want to point to.
10:05I mean, these companies have the earnings to back it up.
10:07So it's a little bit different than a bubble,
10:09but bubbles can go on for years too.
10:11But a lot of my sources tend to push back on things like that.
10:14So it's a little bit different.
Comments