AI Spending Delivers Mixed Results to Stocks

0:00Hello, I'm Stephen Carroll. I'm in Brussels where many of Europe's biggest decisions get made. And I'm Caroline Hepker in London with the hosts of the Bloomberg Daybreak Europe podcast. We're up early every weekday keeping an eye on what's happening across Europe and around the world. We do it early so the news is fresh, not recycled and so you know what actually matters as the day gets going. From Brussels, I'm following the politics, policy and the people shaping the European Union right now. And from London, I'm looking at what all that means for markets, money and the wider economy. We've got reporters across Europe and around the globe feeding in as stories break.

0:37So whether it's geopolitics, energy, tech or markets, you're hearing it while it happens. It's smart, calm and to the point. And it fits into your morning. You can find new episodes of the Bloomberg Daybreak Europe podcast by 7am in Dublin or 8am in Brussels, Berlin and Paris. On Apple, Spotify, YouTube or wherever you get your podcasts.

1:02Bloomberg Audio Studios. Podcasts, radio, news.

1:11Bloomberg Tech is live from coast to coast with Caroline Hyde in New York and Ed Lovellow in San Francisco.

1:22This is Bloomberg Tech coming up. AI CapEx in focus after Meta, Microsoft and Tesla all report earnings with mixed fortunes. Plus, Apple is also set to report earnings later today with holiday season results and the impact of memory prices top of mind for investors. And Amazon reported hundreds of thousands of pieces of content it believed included child sexual abuse material, which it found in data gathered to improve its AI models. We have the Bloomberg reporting. And it is significant in reporting at that. But let's turn our attention to what's happening on the markets right now, Ed, because it is a tough day to say the least.

2:02It is all about CapEx in AI and you're going to go there with the individual names. But I point to a Nasdaq 100 that's having its worst day since November of last year. Two months now, we're down by two percentage points. Some huge moves on individual names, but crypto also languishing. We're off by 5%. We're at 84 ,000 at the moment. So much for being in digital gold in this apparently debasement environment. it ain't working out for this particular area of gold, because now it's just working as a risk asset. What are you looking at? Look, since we've come on air, there is a bit of a vibe shift or a deepening of concern around AI capital expenditures.

2:37Two very similar stories, but stocks going in different directions. Meta told us this year$135 billion of CapEx, but showed in the current period and beyond significant growth, particularly in the ad business, that justifies it. Microsoft, CapEx,$1.5 billion beyond consensus in the quarter gone. But 38 % growth in Azure, the cloud unit, that's the one we look at for the AI progress. On a constant currency basis or currency adjusted basis, it's basically in line with estimates. And people are worried that this growth in CapEx is outpacing what is now decelerating growth in Azure for the time being, Caro.

3:13Much to dig on here. There is. We've got the perfect person to do that with. Let's bring in Gabriela Borgias. analysts covering the emerging software sector within Goldman Sachs' global investment research unit, maintaining a buy rating on Microsoft, citing long-term strategic AI positioning. So do we have to swallow the 66 % increase in CapEx in the quarter for long-term rewards with Azure that just aren't showing up quite yet? Yes. Hi. Good morning. That's exactly the question that investors are asking us. What's really interesting is this is the second quarter in a row where Microsoft is actually telling you there is a more nuanced way to think about their CapEx allocations.

3:50Typically, what you would expect to see is CapEx goes up, Azure revenue goes up. There's a really direct correlation. What's happening in reality is Microsoft is telling you there are two important strategic priorities in addition to Azure revenue. The first is Microsoft Copilot and what they're doing in the first-party apps ecosystem. As Copilot adoption goes up, so does the CapEx allocation that has to go to support that business. The second is what they're doing with internal R &D. So think Microsoft AI, for example, very strategic imperative because if they get some of these new markets right, like medical diagnostics, for example, that could be a really interesting long-term product cycle for them as well.

4:26Amy Hood tried to get that nuance across in the earnings calls saying, look, we could have had markedly bigger percentage growth for Azure if we'd served external clients, but we're serving ourselves with the compute. Why do we see the market overreacting like this, even with that nuance? Yeah, it's a trade-off between short-term and long-term. Short-term, you see that in the Azure revenue number every quarter. Longer-term, though, something like Copilot, that's a gross margin accretion cycle and a long-term monetization story that happens over multiple years. Microsoft AI, that's something that you're going to get maybe a couple of blog posts this year, a couple of really exciting data points, but it's hard to predict when exactly you'll get those data points.

5:08So the way we see this is there's a little bit of a short-term negative reaction here because it's hard to see into the future. Ultimately, we do still believe the data points will accrue positive on those two other strategic initiatives. Let's go back to basics a bit, Gabriella. And Caro, we talked about this before the show. This is a drop of 12%. So this is the biggest drop in the stock since March of 2020. $429 billion of market value wiped off. That suggests serious concern, Gabriella. what do you think is the counter argument to that concern in the market? Yeah, I do think it's going to be a more balanced evolution on what enterprise AI adoption looks like this year.

5:52We know for two years now it's been a real struggle to see adoption tick up on the enterprise AI side. There's a lot of heavy lifting that has to happen on the data architectures, the data infrastructure, marrying some of these really cool LLMs with the richness of an enterprise context, and that's taken time. So as we go through 2026, one of the data points we'll be looking for is, is that enterprise AI adoption starting to go higher? And what Microsoft told you last night is actually they have an increasing uptick on adoption on the copilot side. We've seen data points on Claude Cowork as well just in the last couple of weeks.

6:25You put all of that together with what ServiceNow is saying last night, and there is starting to be enough critical amassed to suggest that enterprise AI adoption will go up from here. Net income was boosted by a gain in Microsoft's investment in OpenAI, lifted per share earnings by$1.02, right? But at issue here is still how Microsoft manages the capacity issue that OpenAI presents them, as well as the circular financing issue, right? That's the arrangement. Here is some capital, and here are the services that we'll charge you for as well. How do you think about that, Gabriella, and whether we now shift OpenAI a bit more into the risk column for Microsoft or not?

7:11So it's really interesting because Microsoft is taking steps to diversify their exposure to the AI ecosystem. Last quarter, we saw Anthropic come on to Microsoft Azure for the first time. We're also seeing Microsoft invest in their own, what they call, off-frontier Microsoft AI models. Each of these steps diversifies the ways that Microsoft can win, independent of whether OpenAI is doing well or not. With OpenAI in particular, what we did see last night is that$250 billion contract that was announced about three months ago now. You're now starting to see that in the backlog, and that will roll on over time.

7:46The other positive catalyst to look for is the Fairwater Data Centers. They're coming online over the course of a multi-year time frame, but we'll start to get more visibility into that as we go through the next two quarters. We've had a bit of visibility about the silicon strategy. We've had visibility about sort of the hardware side, but ultimately this company is getting caught up with software negativity. What about Anthropic coming out? And yes, we see relationships, partnerships, but a co-work offering that could just not co-pilot out the water. How are you thinking about competition right now?

8:16Yeah, the burden of proof will be on Microsoft being able to show that the quality of outputs in a knowledge worker desktop and a knowledge work environment are better than what the competitors can provide. One of the key reasons they may be able to do that is because they're already entrenched in the Microsoft ecosystem. We already use Microsoft Excel, Outlook, PowerPoint. There's a whole suite of apps. And so if the intelligence that's embedded in those apps can be married with the intelligence of the LLMs, you get to a better point from a productivity standpoint than if you bring it at their party.

8:49Exactly. I'm so pleased we're going to this conversation. This has been so focused on earnings and like classic Bloomberg stuff. Like I grew up in the generation of people that learned computers using Microsoft Word and Excel and PowerPoint, et cetera, et cetera. Irrespective of what I use today, I don't use that. I use ChatGPT, Gemini, Claude. The totality and functionality of those platforms makes that kind of suite of software obsolete. And I'm certainly not paying for it. How big a sort of existential question is that for Microsoft? And how reliant is it, sorry to have a long question, how reliant is it for Copilot being a success, therefore?

9:29Yeah, I think there are multiple ways to tackle this problem. I do think there is a disconnect between the consumer-facing market and the enterprise-facing market, where we see this at our own organization. In order to get really good AI tools in the hands of people who work at enterprises, you need enterprise-grade security, reliability, depth of functionality. There are a number of systems that have to ultimately connect to build you something that is sustainable and durable over time. So I do think Copilot is one of the key ways that Microsoft is addressing this problem. But there's also something like Foundry.

10:02Microsoft Foundry allows you to actually benchmark LLMs against each other and pick where you are on the frontier of cost and performance in a way that's quite differentiated today. And so there are multiple ways, whereas the infrastructure layer, the platform layer, the application layer over time that will come together. And one of the things that we wanted to highlight today is just the strategic totalness and completeness of how Microsoft is thinking about this. Yeah. Gabriella Borges of Goldman Sachs has been brilliant to have you on Bloomberg Tech. It's my pleasure. Thank you so much. Thank you for having me.

10:30Let's get over to Meta. Thank you. Shares of Meta right now on track for their biggest jump since July. The social media giant has eased some of Wall Street's concerns about its massive AI spending after beating expectations in the holiday quarter. But it's all in the outlook. Joining us is Shweta Kajuri, a managing director of global internet at Wolf Research, an outperformed rating on the stock price target$850,$135 billion of capital expenditures at the top end, but also enough growth in the core business aided by AI to justify it. Is it as simple as that? I think so. Well, first of all, thanks for having me.

11:07I think the key question getting into the print was, well, Meta is one of those companies that is spending a lot of money not only on AI headcount, but also on CapEx. And they have yet to show something for it. Well, this is a quarter where they're showing it, where the first quarter guide of over 30 percent year over year on a reported basis is the highest we have seen since the third quarter of 2021 in over four years. And so and at that time, Meta's business was about half the size as it is now. So this is best in class, remarkable growth rate. So the point is, if it can accelerate to this level, even if the growth decelerates going forward modestly, they have enough monetization levers that seem durable because they're making so many changes to their own models that are driving impressions growth and ROI for advertisers.

12:00in addition to potentially new monetization beyond advertising revenue that will come on because of their AI models over near to midterm. And that is giving comfort on, OK, well, maybe they are spending to sustain a pretty healthy level of top line growth. Meta is not a hyperscaler, but something interesting Caroline brought up with our last guest, Gabriela from Goldman, is Microsoft tried to talk up all the benefits it's feeling internally. from allocating compute and investing. Susan Lee talks about exactly the same thing. Engineering output internally is up 30%. Is that a data point that you look at and say, okay, I believe you that this is worth it because internally you're doing better as well?

12:46Yeah, absolutely. One of the metrics that we do track is headcount growth. And so what we are seeing is, and what we expect is headcount growth to moderate growing forward, even though it will be in the positive territory, but more moderate, as well as compensation skew to shift more towards engineering and higher compensated employees. In return, what we expect to see is perhaps greater level of productivity or shipments of products and new features that we think will then drive further ROI for advertisers and in exchange drive impressions, growth, engagement growth, and then monetization. That's how we see this play out.

13:25What a difference Accorta makes and how much we beat up over CapEx last time. And this time we see the optimism and the driving forward. The promise of AI that understands us personally is how Mark Zuckerberg was talking about it. But how crucial is therefore the next iteration of their large language model? How important is Avocado, what the code name is and what they're building, that it lands unlike Larmaphor? Well, it is incredibly important. Going into the print, that was one of the key concerns for Meta, that we haven't heard or seen a model yet when Gemini and arguably GPT-5 as well as Cloud's models are way ahead.

14:01The expectation, however, is not for Meta's next model to really break the top three in the LM Arena rankings. It's more about Meta's models to be in the ranking of in conversation of being a competitive model. If that happens, then that is a sign that they are on a good trajectory for upcoming updates to frontier models that they can leverage then for better impressions or monetization. And that should be enough to give investors some comfort. Shrestha, what excites you the most? Is it the commerce applications, the ways in which we're going to find it easier and easier to purchase using their own AI offerings?

14:40Or is it actually that we're purchasing hardware through them? The fact that billions of people can be wearing glasses and might be therefore wearing Meta glasses? I think it's the former. They have shown a lot of competency in technology, in product development, and then also in distribution. And so this is not a new thing for Meta. When we go back and Snapchat was picking up in their product, well, then we saw Reels and it's over$50 billion of revenue run rate. There's OpenAI Sora and guess what? There is Meta AI's Vibes as well. And they benefit from having a massive distribution, great resources, as well as the tech knowledge and product knowledge to scale these types of products and compete.

15:26So I'm more optimistic on their ability to drive monetization, not only within advertising, because they are upgrading their models, but also beyond advertising, whether to your point, it's agentic commerce, because guess what? A lot of Gen Zers start their shopping and they start their search on social media. So there's a huge untapped opportunity there, as well as business AI, where they don't necessarily compete with anybody but themselves in getting a lot of those businesses on WhatsApp on board to spend on their business AI product that we think is a $40 billion opportunity for them. Shredo Kajuria, always so great to have you on the show.

16:03Thank you from Wolf Research. Coming up, more earnings, more CapEx to delve into. Tesla announces a new TerraFab build out as it looks to make its own chips. We're going to speak with Tasha Keeney of ARK Invest. This is Bloomberg Tech.

16:21I think in order to remove the constraint, the probable constraint in three or four years, we're going to have to build a Tesla Terra fab, a very big fab that includes logic, memory, and packaging domestically. That was Tesla CEO Elon Musk outlining ambitious plans for a TerraFab, a chip factory for Tesla in an effort to alleviate chips and parts constraints. Shares now down around 2%. Actually, at one point in the session, we were up 2%. And then as low as down 4%, the company's fourth quarter earnings topped estimates. Everyone looked past them. Tasha Keeney, ARK Invest Director of Research, Autonomous Technology, Robotics, joins us now.

17:09The kind of big headline for me was the capital expenditures commitment,$20 billion this year. You know full well it's way beyond what they might do in a typical year. So let's start with that. It's clearly a signal that these things that Elon Musk has kept you guys looking to the horizon on for some time, now it's on paper. It's time to act. Yes. Well, I think the capital expenditures are not too surprising to us. We had fairly aggressive investment estimations in our public valuation model. But that's all to say, you know, I think the next five years for Tesla will be dominated by the robotaxi story.

17:51You know, we've previously published we think this could be attributable to over 90 percent of the enterprise value for the company in that time period. And, you know, in order to put many vehicles on the road, which is their plan, they need to invest. Of course, a lot of the money is going into factories, AI infrastructure build out. So, you know, I think that the story here is that Tesla, while they were, you know, second or behind Waymo in terms of a commercial launch for a robotaxi platform, they have an incredible scale that is really not matched by anyone in terms of the public announcements that they've made of putting actual vehicles on the road.

18:29Right. On TerraFab specifically, you and the team at ARK know full well, a fab at that scale is a fixed cost of$20 to$30 billion. And the CFO was clear this is not reflected in their CapEx forecast. But there's a reason that the industry doesn't put memory, logic and packaging on one site. That doesn't exist, but it is Musk's proposal. How seriously are you taking that? Well, I think, you know, he's always been a very ambitious CEO. I would interpret this as, look, he's concerned Tesla needs a lot of chips. They have grand ambitions for AI data centers in space. We think that could be a really interesting play, especially as launch costs continue to decline.

19:14It could actually be cheaper than data centers on Earth. So I think the way to interpret it is that he knows that he needs a lot of chips. We've seen him vertically integrate in the past in the electric vehicle business, for instance, and parts of the supply chain that he did not see popping up in the U.S. that they needed in the U.S. So, you know, I think this sort of fits in with what he's done in the past, albeit, yes, it's a major announcement for Tesla. So I'd say broadly that, like, I would interpret some of it as a call to action that, you know, he is looking out there at the supply chain.

19:49But, yes, for now, they have this very ambitious plan. And again, I would focus on the, at least from a valuation perspective, on the robo-taxi picture here. Of course, they'll also use those chips for optimists. But I would focus on robo-taxis. And let's focus just on what you have told the market you anticipate. If I'm right, by 2029, what is it, the stock will hit 2 ,688 % of the company's value will be robo-taxis. And you've talked about the millions of robo-taxis that really could be produced by then. To that end, how integral is the XAI integration, if we're going to think about integration?

20:25Is that important for the story and the$2 billion, well, basically capital allocation that was made towards it? So, you know, of course, we hadn't previously accounted for that. The way that I would think of it as, you know, on the call they mentioned, of course, they're making Grok available in the vehicles. I think the more interesting parts of the story are, of course, you know, they could potentially share AI infrastructure investment and mutually benefit from each other. He did mention that Grok could sort of act as this like maestro for Optimus and for the robotaxis in terms of, you know, like networking them together and sort of being the AI control center, if you will, for robotaxis and any sort of like large scale Optimus project.

21:08So that's how I would think about that investment. Let's talk about this robo-taxi unlock then. What needs to fall into place for the 2029 moment to occur from your perspective? Because at the moment, we've got just a handful in Austin that don't have someone in the driver's seat. How quickly does that scale, do you think, in realistic nature? Yeah, so you're correct that, you know, there really are only a handful that currently have no one in the driver's seat. They did say, though, that they have 500 robo taxis and that they plan to double that every month. I mean, if that's true, if you look at Waymo's fleet, the last reported number we saw was about 3 ,000 cars.

21:45So in other words, Tesla could surpass Waymo's robo taxi fleet within three months. I mean, I think that's incredible if you look at the fact that Waymo launched the service back in 2019. Right. So, again, I would focus on the fact that this is like unmatched scale. They're a vertically integrated player. They do not have to partner with other auto manufacturers. This also gives them a cost advantage. So in the big ideas presentation that we just published, you know, we estimate that they have a 50 percent cost advantage over Waymo on the next generation vehicle if you compare it to the Cyber Cab.

22:18So this is really what matters for Tesla scaling robotaxis that competitors do not have. On top of that, they have a data advantage. You know, if you look at Waymo's miles, we estimate they're doing around 400 ,000 per day. Tesla gets 17 million miles of FSD data per day that they can use to train the fleet. So that is something that competitors just do not have in the robotaxi front. And it's important, that scale is important, because we think that what will drive robotaxi adoption is a reduction in the price per mile. So today, human-driven ride hail costs over$2 per mile on average. We think at scale that could be profitably priced at$0.25, but the reality is there's a lot of headway in between that$2.25 to make profit.

23:03You think it's eventually cheaper than driving your personal car. Tasha Keeney of ARK, we thank you very much indeed. And a correction to myself, the safety drivers are in the passenger seat, as Ed points out. Meanwhile, let's talk Invidia. Still hasn't received any orders from Chinese customers. We'll talk that next. This is Bloomberg Tech.

23:26Welcome back to Bloomberg Tech. We're in it. This is it. Mag7 earnings. And the story, you know it. Capital expenditures, how do they compare to growth directly from AI? Well, in Meta's case, it was enough for the market to believe up. By the most since July of last year, about 8%. Microsoft having a very hard day. Biggest drop since March of 2020. $400 billion of market cap shed. And the reverse equation, Tesla now lower, but it had been higher in the session. A big CapEx number, the likes of which we've not seen before from that name. Plenty of other earnings as well. IBM posting revenue, topping estimates, surviving, doing good in this software environment that's difficult.