China Restricts Nvidia H200s, Meta’s Huge Compute Bet & Apple’s Google Deal | Jan 13, 2026

0:13Welcome everyone to the information's TI TV. My name is Akash Pastricha. It is Tuesday, January 13th. We have a breaking story out of our Asia Bureau. The information has exclusive reporting that China is restricting NVIDIA chip purchases to special circumstances. We'll have more on that story in just a second. We're also breaking down Meta's big new Meta Compute Initiative aimed at scaling its data center footprint. We have a data center expert coming on the show to talk about that. We'll then shift to a conversation with our Google and Apple reporters about the landmark partnership between those two companies that will bring Google's Gemini into new versions of Siri.

0:54I'm also bringing on a deal-making expert, the chief business officer at Brex, to get his take on all of these collaborations. And finally, we will wrap with another story that we published today about how Google is going to great lengths to secure enough power for its rapidly expanding data center buildout. It is a big show, so let's get right on into it. The information published exclusive reporting that China is restricting NVIDIA chip purchases to special circumstances. I want to bring on our co-executive editor, Martin Pierce, to break it all down. Martin, welcome back to the show. It's great to have you here.

1:30Hey, good morning, Akash. So we saw this report coming out of our Asia Bureau. What do we know? What's the news? Well, what my colleagues in Asia have reported is that the Chinese government met with some tech companies this week and has told them that they will only approve the purchases of the H200 chips under special circumstances and that the companies should only consider buying the chips if it's absolutely necessary. Now, what that means is not very clear. They did indicate that they would approve purchases by universities, which suggests that they will allow them, but they really don't want the companies to be buying them as a matter of course.

2:19And that's so bad news for NVIDIA, which had been hoping that the fact that Trump had said he would approve NVIDIA to export these chips, that gave a sign that NVIDIA might be able to get back into China, which is a very important market for them. And I guess it's also not that great for customers in China, the businesses that are trying to get the latest and greatest NVIDIA chips, right? That's right. The companies like Ali Baba and ByteDance need these chips because the locally made chips that the government wants them to buy are not as good in training the AI models as NVIDIA's chips are. So that is the problem.

3:07And, you know, I wonder when you have a story like this that is changing so quickly and the details, look, I mean, they are what they are today. A week from now, things might change. What are the broader questions around the US-China AI race that come to mind for you leading a newsroom here? Well, you've hit the nail on the head. I mean, we reported last week that the government was, you know, they were sort of, they were indicating that they were going to be very careful in whether they would approve the purchases and they've begun to harden their position. Our story says that because the government was a little vague, it sort of opened the door to the idea that they might relax these restrictions over time.

3:59It really does depend on, I think, relations between the governments. You know, we had reported last week there was some speculation that Trump's moves in Venezuela might have annoyed China, which buys much of its oil from there, and that China might have been responding to that by clamping down on NVIDIA chips. So the questions that we look for is obviously we're trying to track what is influencing the government and what could happen next. I want to ask you about another story that has been making headlines today and this week, The New York Times reported that Meta is going to cut about 10 % of its staff in its Reality Labs group.

4:47They said some of those people might go to the wearables group. And then we also had this report from Bloomberg that Meta could double its Ray-Ban glasses output. So, you know, we're seeing sort of a shifting focus here, maybe away from Reality Labs, more towards wearables. If we just stick with Reality Labs here for a second, we've written about this extensively at the information. I mean, is this just a minor restructuring in your view? Is it the beginning of the end for Reality Labs? No, no, no. I think that the wearables is in Reality Labs. Oh, okay. I think that the Ray-Ban increase is a sign that they are focusing more on the product that has actually sold well.

5:30And they're trying to increase capacity. They had to recently restrict sales of those glasses in Britain because they didn't have enough production of them. So I think what they're saying is we're not going to invest as much in these sort of very futuristic metaverse technologies, which they've spent a fortune on over the last five years. And instead, we're going to ramp up the glasses, which have a real role in AI. I mean, I think Zuckerberg believes, and he's not the only one, that AI in the future, people will interact with it using wearables. So I think that they're just refining the focus.

6:14They're not, you know, I mean, Reality Labs has been a huge money pit and they're trying to rein that in because they're also spending a fortune on AI development. Right. So, I mean, this is really them doubling down on what is selling, essentially, and going to where the market is taking them. Got it. Well, thank you for terrifying that. These are the nuances that sometimes we miss here on the show, but we are grateful to have you on to help us make sense of it all. That is Martin Peers, our co-executive editor here at The Information. Okay, Mark Zuckerberg made big news yesterday in another sphere as well, announcing a new initiative called MetaCompute.

6:55He said the company would build tens of thousands of gigawatts of data center compute capacity over the next 10 years, and eventually will build hundreds of gigawatts. Dina Powell-McCormick, who Meta just tapped to be president and vice chairman, will help oversee the initiative along with other Meta executives. I want to bring on Raul Martinique, CEO of data center management company, DataBank, to help us make sense of Meta's new compute initiative. Raul, welcome to the show. It's great to have you here. Thanks for having me. It's great to be here. So I've been dying to talk to someone who is deeply integrated into the data center story to help us make sense of what these big announcements from hyperscaler tech companies and Meta mean.

7:41How are people in your orbit reacting to the announcement yesterday from Meta declaring, hey, we're on a 10-year journey now and, you know, hundreds of gigawatts are coming? Yeah, well, look, it's one of many announcements like that. Obviously, earlier last year, we had the Stargate announcement, right, which at the time seemed to be, you know, astronomical, you know,$500 billion. And now we have, you know, announcements like this and, you know, others have also said that. So, I mean, fundamentally, you know, there is obviously a big boom in terms of building data center capacity to support, you know, these artificial intelligence efforts, right?

8:20A couple of things to remember. I mean, this announcement really was, as you know, it was a thread post. So there wasn't a lot of detail. It was really an indication that, hey, we are going to invest in capacity over the long term. And long term is really the operative word here because data centers, they just don't get built in a quarter or a year. And especially at this scale, it actually takes years to build these things, right? So earlier this week, Meta made an announcement on the nuclear power side where they made a number of commitments to some of these startup nuclear power companies, Oclo and TerraPower, plus some existing nuclear power plants with Vistra.

9:00And that, again, is a long term indication. So, look, my view is that it's an announcement. You know, Meta's obviously made some big investments most recently in Louisiana, where they're building a 2.2 gigawatt data center campus there in partnership with a group called Blue Owl. You know, something like that. It's not going to see capacity till, you know, two years from now, right? So these are kind of long - This is a long time away. I mean, this is really just the start. And I mean, you hinted at it here, but how does the scale of what Zuckerberg is talking about yesterday, how does that compare to their existing footprint in terms of data centers?

9:41Oh, it would be a 10X, a 15X, right? So to give you some perspective, right, before this AI craze started in 2023, the world had about, call it 35 gigawatts of data center capacity with half of it in the US. Over the last couple of years, we've seen record, as we call it, absorption in the space. Last year, the estimate's about 10 gigawatts. So we're probably in the, you know, 50 to 60 gigawatt on a global basis. It's hard to keep track because there's so many announcements. So, and with, again, about half of that being in the U.S. So, you know, obviously, if you're saying you're going to develop hundreds of gigawatts, well, that's, you know, orders of magnitude more than what we have today.

10:25But I think we have to, like, step back and say, okay, you know, what is actually this capacity going to be used for, right? I mean, ultimately, we know the state of AI, right? We have a bunch of companies like OpenAI, Anthropic, and Mistral generating revenue. But these revenue numbers are modest, especially when you compare it to the size of the investments required. So I think we need to take a deep breath and see how this plays out over the next 12 to 18 months, see how the monetization of this technology emerges. And then I think that monetization will ultimately have to merge with these investments.

11:04Because obviously what we don't want is to have too much investment go into an area where the revenue can't keep up with it. Because we've seen that film before in 2000. Well, you hit on a good point here, which is that what are these data centers going to get used for? one of the things I've been thinking about is Meta has, of course, its ads business. And we've talked a little bit on this show about Meta can use its AI for personal super intelligence and some of its more longer term bets. But then it also has this ad business that can make a lot of use of AI in what is very much a profitable business for them.

11:46to what extent do you think these data centers that the company is building, you know, does it have to be all moonshot bets that the demand is sort of accounted for? Or is the ads business and the growth of an ads business, I mean, does that warrant a data center built out at all the way they're doing it? Yeah, that's a great point. Because, you know, when we talk about AI, we tend to not explain too much to go to market strategy. And that these large players, especially the hyperscalers and Meta, they all have different strategies around AI. You're absolutely right. In the case of Meta, I see their investments happening in two areas, and Zuckerberg's given a couple interviews where he's talked about this.

12:24I mean, they have an incredible ad business, right? There's no doubt machine learning before that. Now, LLMs, generative AI is helping them generate more revenue on the advertising side. So they're going to make investments there. Your previous guest was talking about reality labs. And obviously, we know Meta has made a huge bet on the metaverse, which really hasn't panned out there. So there's other of these kind of long-term investments where they want to apply this technology towards agentic AI or trying to monetize what they have with WhatsApp or through reality labs and the metaverse and things like that.

13:02So those are the ones that are more speculative, but there's no doubt that a big part of this is about just growing their organic business. And that, as proven by their track record, that's a good investment. I want to ask you about the risks that come with a data center build out of this scale. And here, I'm talking not just about Meta, I'm talking about all of these tech companies, big and small, profitable or not profitable yet, who are building these data centers so quickly. I'm reminded of a project that we had at The Information Here. It was the – we had a database of gigafactories that were involved in building everything you need for electric vehicles and the materials and the batteries and stuff like that.

13:49And look, a lot of those facilities, I mean, they were announced. They maybe started construction. They were scaled back. Some of them were shuttered entirely. Do you think that there's a chance here that some of these data centers that are working towards satisfying this AI demand three years from now, could we be seeing a similar scenario where these projects are scaled back or even shuttered? And I say that knowing that at DataBank, your business is operating these data centers. Surely that is a risk that you must think about as you seek to advise people on what data centers they should build and not.

14:26Yeah, that's the multi-trillion dollar question at this point, right? And look, here's the way I think about it. Long term, there is no doubt in my mind you're going to need more data center capacity. The question is, over a short term period, it's three to five years, how does that supply demand curve look, right? And that, to my earlier comment, it really, you've got to follow the revenue here. You've got to follow, you know, what is happening with these labs in terms of, you know, their revenue generation potential with this technology. And yes, you have a significant amount of capacity being brought online, presumably to support these new revenue streams.

15:05If those new revenue streams take longer to develop, well, guess what? There will be a quote-unquote oversupply and we'll end up in a different situation. Again, I think long-term, these investments will make sense, but it's hard to predict, right? In 1995 and 1997, I lived the dot-com era. Everyone was making the same investments on Fiverr. Well, guess what? It turned out it took a little longer to absorb all that Fiverr. It's still very useful, but a lot of people lost a lot of money. So that's kind of the way we think about it. And at DataBank, we're developing about a gigawatt of data center capacity across our footprint between 2026 and 2030.

15:46We think that's a reasonable amount of capacity to bring online where regardless of kind of that macro supply, demand that we'll find customers for it. And last question for you, as you think about the risks or you talk about the demand risk, and that is something we'll see how it plays out. There's also what I sort of maybe call construction risk or innovation risk, which is can you get the power? Can you get the economics of the power to work? We had a founder on the show yesterday who is innovating in the cooling space and finding new ways to cool things. As you think about the risks and where the rate limiting steps are, why these projects could take longer and longer, are power and cooling, are those the two main areas?

16:30Are there other areas that we should be thinking about in terms of bottlenecks or rate limiting steps? Yeah, there's a lot of bottlenecks. I mean, obviously, power, access to power is a big one. But, you know, look, look what's happening, right? Some people are, you know, to get around that, they're going into, you know, 400 miles west of Texas, you know, to be able to access power, right? You know, location, this is still a real estate business. So location is important for a lot of customers. So that's a constraint. You know, the technical constraints really, from a data center perspective, they don't really exist.

17:05It's about how you get that more efficiently. But I'll give you another risk, which is maybe we don't need so much power to power all this AI. If you look at NVIDIA's recent acquihire of Grok, the Grok platform developed by Jonathan Ross, who obviously is one of the founders of the TPU, that is an inference-only type of platform. Right. That could bring it down entirely. It operates at 10, 15, 20x less power consumption per token, right? Right. There's a lot of new startups occurring in Silicon Valley that are trying to do 1000x more efficient than the NVIDIA GPU platform, right? So I think we have a long way to go before we say that, you know, power is the constraint on a long-term basis.

17:54It certainly is one of the constraints now. Great. Raul, I want to thank you for coming on. That is Raul Martinique, the CEO of DataBank here on TITB. Okay, Apple and Google's deal yesterday was a landmark announcement for iPhone owners and the AI ecosystem at large. The companies announced that new versions of Siri will use Google's Gemini under the hood. Our Apple and Google reporters were both working the phones all day yesterday trying to figure out how exactly the deal will work behind the scenes. I want to bring them on to talk about it. This is Aaron Tilly, our Apple reporter, and Aaron Wu, our Google reporter.

18:31Welcome to the show. It's great to have you both here. Thanks for having us. The Aaron and Aaron Show. Once again, we're missing Aaron Holmes, as usual. I'm waiting for the trio segment. Aaron Tilly, let's start with you. How is this deal going to work behind the scenes from what we know? Yeah, so basically, Apple is allowed to take these Google models, the Gemini models, and adapt it for its own purposes so it can tweak it and do what it needs to do behind the scenes to get it working on Apple products, Apple features like Siri. So it's really incorporating Gemini as if it will be an internal model.

19:14So it's not going to be routing out to Google servers. It's going to be incorporated like as if it's an internal sort of model that Apple's using for its features. and remind us timeline wise when are we expecting the new siri to come out again um so apple's continually delayed it over the past uh you know almost two years now and they've said this year is the year they'll finally kind of come out with what they initially promised in summer of 2024 and we understand there'll be some um releases in the spring and that they will then announce kind of a more full-fledged AI in the summer at their Worldwide Developers Conference.

20:00Erin Wu, do we know how much Google is paying Apple here at all? Yeah, actually, in this situation, Apple is paying Google. We don't totally know. Oh, sorry. Right. My bad. Apple is paying. That's what I meant. How much is Apple paying Google? Apple is paying Google. So yesterday, the two companies released a joint statement that did not get into any of the terms of the deal. So we don't 100 % know. Bloomberg reported last fall that Apple is paying Google$1 billion annually. A source with knowledge of the agreement yesterday told me that the number could be way bigger than that. So this is definitely a win for Google from a monetary perspective.

20:37And I wonder how you think, Aaron Wu, about this from the context of Google now has this deal with Apple. Google obviously has the Samsung ecosystem and its whole Android operating platform as well. I mean, it just strikes me that they obviously would integrate Gemini into that ecosystem. And so now they really have the market cornered in terms of what I think are the top two distribution ways for hardware, right? Yeah, I mean, they're really quartering the market essentially on mobile devices. So they've got this deal with Apple. They've also got a deal with Samsung. Last week, the Samsung co-CEO said that they plan to increase the number of devices running Gemini to 800 million to double the number of devices running Gemini.

21:25So Gemini there, there's the Gemini consumer app and there's also Gemini powers Galaxy AI, which is like a lot of AI features on Samsung phones. So essentially, Google in a lot of ways has the kind of distribution that OpenAI is really trying to get via creating these mobile devices. Like Google already has essentially the mobile devices. Aaron Tilley, one of the interesting parts of what you were reporting on was the privacy element of the data here. I know that when Apple Intelligence was announced, I mean, privacy was a big thing that they highlighted and they had their whole private cloud server business.

22:02I don't understand how it works, but I remember them saying something to the effect of, don't worry, the data is still sort of private to your device or maybe you have a private cloud. Is any of that changing with these models here? How does it work with Siri? Yeah, so none of those plans are changing at all because the way they're incorporating these Gemini models, like incorporating them as if they're Apple AI models, They will look completely, like they said earlier, where all the AI processing will happen either on device or in their private cloud system. And with their private cloud system, they promise they're not using that data to train their algorithms further.

22:45It's all just processed anonymously, not attached to a user. So they say through how they're incorporating Gemini, it'll continue to promise that privacy. And so sticking with you, Aaron Tilly, what happens to the deal with OpenAI? That's still in place. The deals were pretty different. The deal with OpenAI, it sort of links out to OpenAI. When you ask Siri a certain way, if you ask Siri a question a certain way, it will sometimes link out to OpenAI, including putting a prompt window saying, do you want to go to OpenAI? So that still exists. It's just like it'll become even less and less relevant as, you know, Siri can answer these questions itself more and more with Gemini helping it.

23:43Aaron Wu, I'm sort of thinking back to all of these antitrust cases that Google was fighting. And I seem to remember the judge having said something to the effect of the AI has changed the game entirely. And, you know, when we were thinking about whether or not Google needed to sell Chrome, The judge said, hey, it's a totally different world than a couple years ago when these antitrust lawsuits may have originally been filed. That was at a time when ChatGPT was ahead in the race against Gemini. And it still is ahead, I think, based on the numbers. But here you have Google again doing big deals with Apple.

24:29Do people in your orbit talk about how quickly the game has changed and how somewhat ironic it is that that was one of the reasons that the judge ruled the way that it did? Yeah, I mean, I think there's definitely been an insane vibe shift in the past few months where everyone's really bullish on Google now. A lot of people are saying, oh, a lot of the doubt was misplaced and we've seen them really come from behind in the past few months, in the past year, to either have models that are at least on par, if not better, than what OpenAI is doing. And Google is still behind when it comes to chatbot distribution, according to all of the numbers that we have.

25:10But Google also has all of these other distribution channels, like Google has Search, Google has all of these other products that it's increasingly putting Gemini in. And to your point with these distribution deals, in some ways, they're following a very similar playbook with search, which is getting searched on Apple, getting searched on Android. Again, what came out in the remedies ruling was that Google isn't allowed to essentially tie the installment of Gemini to other apps. So they can't say like, oh, if you want to have Maps or if you want to have YouTube, you also have to have Gemini. And so there's some stuff in the Remedies ruling that's hopefully going to prevent Google from using similar illegal monopolistic techniques to distribute Gemini.

25:50But it's obvious that they have these relationships with Samsung. They have these relationships with Apple. And so they're also continuing to leverage them to get the AI on these devices. And again, like the models are good now. So it's not as if these companies are taking like an inferior model because of their existing relationship. Just simplify what you were saying here for us for a second. The distribution deals. So remind us here. So the map stuff, what was the deal then? I just want to understand it. Okay, so basically what was ruled to be monopolistic was that Google was making exclusive distribution deals with partners like phone manufacturers, mobile carriers, etc.

26:33and they would condition having other apps on having search. So if you want to have these other apps, for example, the Android Play Store without which these games are useless, you also need to have search. And so right now in these deals, as far as we know, they're not exclusive. They don't condition having any other apps on having Gemini. So as far as we know, Google is complying with the term. Right. In other words, you can have an app that doesn't use Gemini. It's not like it's the only way that that app is going to work type of thing. Got it. Great. Well, it is an exciting deal. And I want to thank you both for coming on to help us make sense of it.

27:16That is Aaron Wu, our Google reporter, and Aaron Tilly, our Apple reporter, here at The Information. Okay, Apple's deal with Google this week is just the latest partnership that has seen even the biggest tech companies look outside their own businesses for ways to offer more to their customers. We, of course, have also seen a string of AI-related acquisitions lately as big companies try to grow their teams and their capabilities even faster. I want to bring on Art Levy, Chief Business Officer at Brex, to help us break down this current moment for these types of deals and transactions and partnerships.

27:52Art, welcome back to the show. It's great to have you here. Great to be here again. So what is your assessment of this Apple-Google partnership that we saw this week? Yeah, so I think the name of the game on everything you touched on, whether it's the M &As, the partnerships that we're seeing, is the need to move quickly. All companies from a two-person startup to the largest businesses in the world are realizing they need to move as fast as they can. And for a long time, Apple's Siri product has not been great. And so this is them admitting, okay, we will partner with Google Gemini and immediately, because it's going to ship in 26, make the feature set in Siri much, much better inside of Apple intelligence.