This is Bloomberg Businessweek Daily, reporting from the magazine that helps global leaders stay ahead with insight on the people, companies, and trends shaping today's complex economy, plus global business, finance, and tech news as it happens. The Bloomberg Businessweek Daily podcast with Carol Massar and Tim Stenovec.
This rebound in chipmakers is booming. The stock market is back. Buyers emerged on bets that the AI investment boom will continue to support solid earnings following a selloff in one of Wall Street's most crowded bets. We did see a gauge of chip giants climbing. As we said, the Nasdaq 100 is higher. Keep in mind we'll also get some insight into that AI investment boom, thanks to Meta Platforms reporting earnings, along with Tesla also reporting.
And then there's more to come. Here to kind of set us off for the week is Mandeep Singh. He's global head of tech research at Bloomberg Intelligence. He's here in our Bloomberg Interactive Brokers studio. Last week was tough for the AI corps. We've, just in general, it just felt like such a pushback to the semiconductors. We get Meta, SpaceX, Tesla. How important are these to you in terms of getting another read on what's going on in the AI boom?
I mean, I consider SpaceX to be a hyperscaler along with Meta. And so SpaceX to me is really interesting given the backdrop of, you know, the Starlink situation and what's going on with them. And obviously, you know, they won't comment much, but I expect, you know, on the analyst call, you would see questions around Starlink and then what it means for the cloud segment growth, because in the end, you know, they need more customers. They need that cloud segment to ramp up. And the expectations are quite high because the company guided $200 billion in revenue by the end of the year.
And they're nowhere close, are they?
Right now, they are closer to a $25 billion, $30 billion revenue run rate. So even if you extrapolate, they are nowhere close to that $100 billion mark. And so that's where the expectations are high. And you want to see them talk about customer growth, which they have — I mean, Anthropic and Google. But they need more than two customers.
Yeah. And I mean, this is just their second public quarterly earnings report. So do you feel like there's going to be any sense of understanding, I should say, or maybe do you think investors may be a little bit easier on them as they're just kind of getting out? Or you mentioned the bar being high. Talk about that.
I mean, I think overall the sentiment going into the print is, I've been hearing a lot about Starlink. It's not very clear what happened. And look, that is the more established part of the business. And if they can't win business outside the US, that's going to become a problem, whether it's Starlink or the cloud side of the equation. So something to keep an eye on in terms of how they guide. And I feel they've been quite aggressive with their guidance. So in the end, it comes down to the commentary around the expectations for 2027. And then how does the market react around CapEx? Because we have seen what's gone on with CoreWeave. I mean, SpaceX is incredibly levered as well. So if the market is not liking, you know, the high leverage for these neoclouds, I don't think SpaceX gets a pass either.
Yeah. I mean, debt is debt. And it's, you start to look at the — I'm looking at the equity side. SpaceX shares are down about 27% from that high just after the IPO. All right. So we're watching that one very closely. What about Meta and how that fits into the AI trade? Because it feels like all of a sudden, like, wow, look at what Llama is doing and everything that they're doing. And they've gotten a lot of attention.
I do, but with that being said, now that they have a frontier model to show for all the CapEx that they have done so far, the monetization question still looms. And, you know, if they're talking about raising CapEx for 2027, which I think they will, given they feel emboldened by the fact that they have a frontier model now, then the debt question starts to come, even for someone like Meta, because now it's not about financing everything through your operating cash flow. You have to ensure that. And yes, Meta is investment grade and they don't have any debt concerns like CoreWeave. But at the same time, you know, the bar is getting higher in terms of showing that monetization.
Why is it all of a sudden getting higher? Like, it just does feel like we've been kind of building to this, but it does feel like we're kind of at a — cost of capital has gone up. Look at where we were at the start of the year and where we are now with Treasury yields. And so everything has gone up in terms of, you know, what do you have to pay to get any debt? You know, it's interesting because I feel like if you flash back just about six months ago, it was a lot of enthusiasm about the AI spend. So do you think it really is that cost of capital that's really changed the narrative in a sense? Or is it also a combination of something else?
And, you know, the revenue that these companies have to show. So we know the frontier labs have grown their revenue, you know, fivefold from the start of the year to where we are now. But even within Anthropic, they had stellar growth for the first nine months. Well, now we are starting to see maybe that monthly ARR growth is slowing down. Yeah. So when you are in that sort of situation where, again, the growth numbers are great for the overall year, but then people are parsing through what's going on recently. And why is it slowing down? Is it because of open-weight models or token pricing coming down? So it impacts the ROI. And, you know, what these companies have to show for all the CapEx that they are doing.
Dan Ives is going to join us after you, after a break. He does write this week, a validation moment for AI, kind of some of what you said. He says through the names at the center of the AI revolution reporting, he says we're going to know whether or not Meta can begin to monetize AI on top of the ad engine, whether SpaceX can show Starlink's cash funding an AI business that is starting to pay for itself, and whether Tesla can move from autonomy, from optionality to a fleet. Do you kind of agree with most of this?
I mean, I feel this earnings season, you will see the cloud hyperscalers. And I want to put Meta in that basket. So it's really your Amazon, Microsoft, and Google, right? They will shine because their cloud business will show a sequential acceleration. You know, the earnings will reflect that because it's not just AI that's part of that business. It's that part of it. So they are benefiting more from the infrastructure demand. So they are running the workloads that are needed for OpenAI and Anthropic. So even though Anthropic is facing pressure from token pricing going down, they still need to run their workloads on Amazon and Google infrastructure, and they're not cutting their prices. So in the end, I feel because AI demand and, you know, usage has gone up, that will show up in the hyperscale cloud earnings. And so they will have a nice acceleration. I expect all the clouds to do well. But then I think the lab side seems to be incrementally slowing down because of open-source pressure.
That's — these guys are not the way out from that. That's real. Yes. Yeah. I'm curious, you know, which of the hyperscalers do you think the earnings report is most consequential this season? Do you think it's going to be a bit of a digest of all of them together, or is there one that you think might be the most sensitive?
I mean, with Google, there is always pressure on them because everyone feels their search business is under threat from, you know, OpenAI and now Meta and all that. So there's always that, you know, risk associated with Google. But they come out with great numbers. So I feel from that perspective, Google is always the one which is under the radar, but they've come out and beat their numbers, you know, every time.
You know, it's funny that you say other open source. Like, it just feels like we had the CEO and co-founder of Perplexity on and talking about doing just that in terms of open source and thinking that's the best way to go in terms of users and possibly people who might want to do something wrong. Just have it open to everybody. But increasingly there's more and more newer names that are coming into this, and it changes the dynamics.
It does. Look, I think you still need compute. If there's one thing we know about AI is I need a lot of compute. And companies, even the labs, haven't been able to optimize that. But it's still not a guarantee for the neoclouds, right? Not a guarantee for the neoclouds because of their financing structure. So, you know, that's surprising. Part of me has shown that negative sentiment despite the GPU rental prices going up. So you would think that should be a favorable dynamic for them. Right. But it's just the financing structure and all the commitments they have. And that's why, you know, they had that pressure last week.
Stay with us. More from Bloomberg Businessweek Daily coming up after this.
As we start to get a gut check on the AI spend, the AI narrative, the valuations, a lot of big tech players starting to report earnings this week and then in the next few weeks. Our next guest is out with a research note calling this week a validation moment for AI. Let's get to it. Good to have back with us on Bloomberg, Dan Ives. He's partner and senior managing director at Yorkville Partners and Company. There's also the Dan Ives Wedbush AI Evolution ETF, which has some $885 million in assets, up more than 20% year to date. And then there is the just launched $200 million publicly listed closed-end investment fund called AI's Ultra AI Opportunities. It invests in private VC-backed AI companies. He's always a busy man, but he always finds time for us. Dan, good to have you back. You're out there?
Yeah. How are you? Good. No. It's good. It's obviously been, you know, it's been exciting but turbulent last few weeks. We were in Korea last week trying to get a better sense of numbers and memory and what demand looks like there. But look, it all comes down to this week and next week in terms of earnings. I mean, that's, I think, altering what the street needs in terms of the hyperscalers and big tech.
What did you glean from your time in Korea? Obviously an important point. And we've seen their markets under extreme pressure. And there's a lot of names obviously that play into the semiconductor space that seem to be pretty vulnerable last week, as were US semiconductors. But what did you glean from what you saw and heard?
Yeah. No doubt, nervousness. But demand supply 13 to 1 for memory chips. So I just look, as I've said on the show, you know, over the years, like, we're going to go through these narratives. Go back to SaaS apocalypse earlier this year. You look what's happened here. I am not saying that there's no validity. Of course there's validity in terms of some nervousness. What we saw with CoreWeave, even the bonds and CDS spreads. But the reality, this acceleration is happening in terms of demand on chips, hyperscalers, use cases. And that's why I think earnings season can be sharp.
And yeah, just how important is this week in terms of just restoring confidence really in the AI trade?
I think it's very important because in the vacuum, when you don't have numbers, you could have different narratives, different reports, you know, and I think that sort of could dictate sometimes. And I think it gets investors off their game because there is nervousness. We're going into the midterms. There's always a lot of volatility around this period of time. But at the end of the day, it comes down to fundamentals. What does demand look like on chips? What are enterprises doing relative to the hyperscalers? And what are the use cases? Palantir, Snowflake, you know, ServiceNow, and so many others. And that is what would dictate if stocks go higher. And I still believe we're in year three of a 10 to 12 year buildout.
What do you then — you know, what does it mean if we're in year three of a ten-year buildout, Dan? And you've got, you know, in scale, we had a headline cross, and Scale AI and Anthropic in talks for a $45 billion cloud computing deal. So basically, Scale AI would be selling computing power to Anthropic. And this deal could be worth $45 billion. What's interesting is, and it could be finalized in the coming days, according to our daily live show, Stena Bass and Brody Ford. But it comes after we just had a story about how Scale AI's valuation is being questioned. This $30 billion valuation, investors against the backdrop and concerns about AI and especially the neoclouds, that maybe it's worth about half of that. So do you have a feeling on neoclouds versus maybe the hyperscalers in terms of their security and their role longer term?
Yeah, I think hyperscalers, they're in a totally different world relative to not just their balance sheet. And then you look at the enterprises in terms of where they sit, how they're positioned in terms of the strategic buildout as well as on the data center buildout. And that's why I think Microsoft, Amazon, Google continue to be in just a different stratosphere. Oracle is a good example of trying to be a hyperscaler. But there is pain with that, right? When you take on debt and they've had a cost when you could say, I'm a CoreWeave, CoreWeave's dues are going to be sort of champions relative to what we see on neoclouds. But I think you have to separate it out and look in the game of musical chairs. Will there be some without a chair? Of course. But I think relative to where we are today, it's so early. And if you look at the demand and you could ultimately capacity where it's going to go, even if you're haircutting from the data centers that don't ultimately get built out, I still think the street's underestimating numbers massively over the next one, two years.
Can I just ask you, are you invested in Scale AI in your — Yes. We just raised — so that $200 million we just raised, we haven't made an investment yet from the IPO. Do you want to? Oh, yeah. Can you tell us where exactly? They always give you the whole — But look, the whole reason we did that, yeah, it was to give investors around the world the ability to have a public equity that invests in private AI, private, you know, private AI, wider companies that are out there because so many are private. And it shouldn't just be a handful of people in Silicon Valley that could own those names. That's the whole goal of the idea. So a name like Scale AI is what you would want in that fund. I mean, that's a good example of one's — whether it's, and they're all, whether it's OpenAI, you know, whether it's Figure, I mean, there's so many great private companies.
We are in an innovation boom. But the problem is that if you look at it just — and we always talk about with the bears, and I get the debate and so many smart investors that obviously are on the other side of this. But it all comes down to execution, building out the fourth industrial revolution as this happens, that will prove it out in just like the last few years. It will show this is more of a white-knuckle moment than a time to panic.
Dan, what would you actually need to see in these earnings to make you less bullish on the AI trade?
Yeah, it's a great question. If we saw CapEx that starts to take foot off the pedal. If we start to see demand deals that we thought were going to happen on this — on the hyperscale side, not happening. So I think that to me is the key. And I could just tell you from our checks combined, what we see is going to happen. Earnings season, like, for every dollar in CapEx on big tech, there's a five to six dollar multiplier we estimate across the rest of tech. And that speaks to why this is just a return of a generational theme to the three to four trillion that's going to be spent over the coming years.
You know, and I know we've been talking a little bit — you mentioned CoreWeave. We've been talking about the debt. It felt like there was a time, though, where investors — or excuse me, companies could really just take on any amount of debt. We could talk about the AI spend, and it felt as though they were rewarded for it. But now we're seeing a bit of a shift here in the market where you are seeing a lot more scrutiny. Can you talk to us just about the burden of proof that we're now expecting essentially out of these big tech firms?
Yeah, I think part of it is just what we see with bonds and then just with yield. So there's just more pressure as you get to a little more maturity on the trade. Investors are trying to understand, like, okay, what does capacity look like? Like, who is at risk? Who's not at risk? Like, what is GPU capacity? Okay. What is backlog? Okay. So I think that as we go into the next phase, investors start to get more of their colleagues to understand who they're going to bet on and maybe who they're going to bet against. That's also why, like, if you go back to maybe where this was one, two years ago, it's a — we're in a different phase of the market. But I think that's a great thing because the data centers, again, build the use case. You're getting built. Look at Apple. Look at Meta. Look at Tesla. So good space, physical AI. Now we go into the next phases and they will. Not everyone's a winner, but it's — you better be able to discern who the winners are from the losers.
So what has to be a loser? Maybe it's in this earnings season. What company is a loser that you like? Ooh, okay. Maybe this is a sign of something that could go wrong. What company, Dan?
I mean, like on the — like, in large part, I've been more cautious on Oracle. Now, Oracle is not — they're a phenomenal company and they're — and you think about OpenAI's $300 billion deal. Yeah. The reason we're just more cautious on that is that when you have to cut muscle, cut your core to build out data centers, the business gets hurt and it's hard for something to — the stock to work in that period because they're trying to be a hyperscaler space. They're trying to be a hyperscaler. And that's why I think ultimately they're going to need Tesla as part of it. You see where they're doing their video. But who stands in a different league is Microsoft, Amazon, Google, because they're already there.
Stay with us. More from Bloomberg Businessweek Daily coming up after this.
It? Yes, it is. Carol Massar, along with Isabelle Lee in for Tim Stenovec. On this Monday, we're going to go from New York to Vegas because underway in Las Vegas is Money20/20. It's a conference that describes itself as a content, sales, and networking platform for the global money ecosystem. We know there's so much going on right now. They focus on what's next across the world of payments, fintech, and financial services. Something you feel like you talked a lot about?
Definitely. I actually have been to one of these conferences and you meet all sorts of people. It's actually fun. You meet financial advisors, portfolio managers, the intersection of all of this.
Right. And we said go — well, we're going to go there now because right now with us is the CEO and co-founder of the AI search company Perplexity. He is Aravind Srinivasan and he does join us from Money20/20 underway in Vegas. Also with us is the host of Bloomberg Tech on Bloomberg TV, Ed Ludlow. He's out there in our San Francisco bureau. Aravind, I want to kick it off with you. Your panel tomorrow speaks to us in a big way when it gets to what's going on in financial services and how AI is rewriting it. So I am curious, what does it mean in the real world when we talk about this? Or you've got something that's entitled this, and what does it mean for Perplexity or another agent that what it could actually do for the financial world? Because I will say, being at something like FutureProof out in Huntington Beach recently in California, there was a lot of talk about the role of agents when it comes to the financial world.
A lot of people think of agents as just consuming data that exists on your financial services. Like, Plaid is a great connector that aggregates a lot of different data in one single platform. The key difference on Perplexity is Perplexity is an accuracy product. Whether you think of it as a search product or an agent product, doesn't matter. We are super focused on being the most accurate AI out there. And there's this benchmark of search.com built by ByteDance and Columbia Business School on this. Perplexity is the state of the art. Finance is a vertical on which accuracy is table stakes. And Perplexity is the best accuracy because Perplexity focuses on search and focuses a lot on making sure that the agents that power the financial research are grounded in accurate, real-time data, as well as your personal data, with a clear auditability trail to where all that data comes from. So that is the key differentiation to what happens on Perplexity Comet. And we've been working with Plaid for many months now and the usage has been skyrocketing. People who are using the Plaid connector are using it to begin with to track their spend, but over time, they basically use it as their personal CFO. In your own personal life, you're the CEO. Perplexity becomes your CFO for tracking all your personal finance spend, figuring out your investment strategy, how to construct your portfolio, how to make sense of what's happening in the world that affects your portfolio, how to rebalance it. And so it starts from basic things like spending a lot and helping me get to a certain budget of spend per month on what subscriptions I'm wasting my money on, to actually building a portfolio and like, I mean, crafting a retirement strategy all the way there.
I have, and I appreciate the emphasis on accuracy. And there's a distinction, right? In a world where an agent is AI that can take action as opposed to just a recommendation. Could you talk a little bit about the kind of technical challenge of that, how the model can decide when it's right to make an independent, autonomous action decision, and when it still needs to be referred back to the user for a check?
At this point in time, we are only offering read-only agents. So Perplexity Comet can only do read-only actions on data it ingests from Plaid or the data ingested from Morningstar, PitchBook, or FactSet. So we believe that agents are still not quite there to actually go and make investments on your behalf or replace trades on your behalf. And you have to be very precise on what you want them to do. We have, of course, integrated Coinbase and Public and all these APIs for creating. But the way people actually use these agents is not to actually go build a bot that trades on your behalf, but actually make sense of what's happening in the world and help you decide what trades to place. That's where we are today. And we really believe that accuracy is the only way you gain the user trust. You need to first gain the user trust for them to make decisions on what to place, straight down on what investments to buy on using your tool before trusting your tool to actually go do that on your own behalf.
I have — a big news headline in the past week is that you are open-sourcing Space, basically the security system that you sandbox agents in, right? So the agent stays within Space while it runs code, works on files. Why did you take that step?
We want to support the ecosystem. We started building the sandbox platform to serve ourselves. If we relied on a third-party provider to run our sandboxes, we would be spending an additional $50 million a year or something on the scale of Perplexity we serve. So it made all sense for us to build our own platform and to optimize it to be secure. Because you're often reading about how agents are going rogue, escaping sandboxes, doing all sorts of arbitrary things that were not intended to do. Our attackers can gain control into the sandbox on which all your credentials are running, and they could use it in, like, malicious ways. So we wanted to first make sure it saves us a lot of money, and it's very secure for our users and secure for our company. And we decided that it's in our best interest to open-source it because it attracts a lot more eyes on it. And that way, we can make it even more secure and even more robust and even more scalable, as well as optimize it for multiple different chips. So x86 is the default CPU on which all these sandboxes are running right now. But specialized CPUs are being built for hosting sandboxes. So that's another big focus for us.
