Welcome, everyone, to The Information's TI TV. My name is Akash Masritcha. It is Tuesday, April 28th. Today on the show, The Information has new reporting on behind-the-scenes negotiations that have happened between OpenAI and Microsoft to avoid a legal battle over OpenAI's work with AWS. This is a follow-up story to the news that we had yesterday of the revised partnership. We're going to have our Microsoft reporter on in just a second to walk us through what he knows. We also have an exclusive story about Google's deal with the Department of Defense. We'll then hear from our reporter covering the Elon Musk-Sam Altman trial. He'll bring us inside the courtroom as opening statements begin. We also have some highlights and reflections from our flagship finance conference at the New York Stock Exchange yesterday. And we will wrap with an expert reaction to reports this week of OpenAI working with Qualcomm. It's going to be a fun show, so let's get right on into it. The information has new details about the internal conversations that led to another shakeup of the Microsoft and OpenAI partnership. My colleagues Aaron Holmes and Tremapiti published the details of that story today. And Aaron joins me now to share with us what he knows. Aaron, talk to us a little bit about what you found since 24 hours ago when we heard the news about this revised deal. Yeah, so to recap, we essentially learned that the deal was the result of several weeks of negotiations between OpenAI CEO Sam Altman and Microsoft CEO Satya Nadella. And those negotiations started over concerns that this planned product that OpenAI was building with Amazon would have potentially violated its contract with Microsoft, in Microsoft's opinion. And, you know, we had heard about a month ago that Microsoft was actually looking at, you know, exploring potential legal action to potentially prevent what it saw as those contractual infractions. And, you know, after those concerns were raised, OpenAI actually approached Microsoft and said, you know, let's settle this outside of the courtroom and renegotiate our deal, which is what led to this new agreement yesterday that essentially let OpenAI out of the exclusive Azure selling arrangement and also resulted in OpenAI continuing to share revenue with Microsoft, including the revenue that it will make from its new Amazon product. So I just want to take a step back here for a second and just acknowledge this. I mean, if there was a legal battle between Microsoft and OpenAI at any point, if that ever came to pass, that would have been kind of epic, right? I mean, you know, here you have Sam Altman and Satya Nadella who, you know, they're just the best of friends, you know, on every podcast they go to. They cannot stop singing each other's praises. I don't know. It would have been kind of fun, right? It would have been great for us in the press, too, because, you know, lawsuits mean that everything goes to discovery and we get to actually read in detail what's been going on between these companies. But, yeah, I think that both companies decided that it was probably not worth the time and money that that would have involved. I get it. I get it. OK, fine. And so they've averted the legal battle. But my question for you now is, I mean, let's go back to that AWS contract that has sort of become central to these negotiations. So did Microsoft not know that OpenAI was doing this work with AWS? I mean, did you hear about any sort of conversation that happened while that contract was getting sorted out in the first place? So Microsoft knew that OpenAI was doing a deal with Amazon. You know, the companies back in February were in talks to have Amazon invest what ended up being $50 billion into OpenAI. And they were also working on a product together. But what Microsoft did not realize is that that product was essentially going to be a substitute for buying OpenAI's models themselves. and additionally you know Microsoft's understanding of their contract was that any other product OpenAI were to build would essentially be a wrapper around the models running on Azure which you know would mean that Microsoft's Azure is still capturing most of the compute costs of running those but what surprised Microsoft as we learned was that OpenAI and Amazon announced that their product would actually run entirely on Amazon's cloud. And it would be, you know, this thing called a stateful runtime environment that would not need to make any calls back to the models on Azure. Stateful meaning what? Stateful essentially means that, you know, it's a product that remembers more about an individual user's session. And that it was essentially called a stateful product specifically because the contract between Microsoft and OpenAI gave Microsoft the rights for stateless models, which is the generic form. And I'm told that executives at Microsoft essentially believed it would be impossible to build a product that had a stateful environment without relying on those stateless Azure models. But once OpenAI and Amazon announced their stateful product and started saying it would run entirely on Amazon's cloud, that's where it became a problem for Microsoft. now i asked you yesterday when we were recapping this deal if you thought that this was sort of the end of these negotiations given that this this partnership has gone through many different iterations and what you mentioned to me was that because the agi clause has sort of been removed now from the equation this sort of feels like the final form or as close we're going to get to a final form. I wonder, based on the reporting that you've done now in the last 24 hours, are there still questions you think that will need to get sorted out between these two? Yeah, so you're right. I mean, the AGI clause was kind of the ticking time bomb that has now been diffused. And, you know, the companies don't have to worry about any potential fight over that anymore. But there are still some sticking points that could continue to come up. For example, we learned that under the new deal as of yesterday, Microsoft still is entitled to all of the intellectual property that OpenAI would get from doing an acquisition. And we know that that's actually been a sticking point in the past. For example, last year, OpenAI tried to acquire the AI coding startup, Windsurf, and it tried to ask Microsoft not to get the IP from that acquisition because Microsoft has its own AI coding tool, GitHub Copilot. Microsoft said, no, we're going to take the IP that is contractually ours. And that deal ultimately fell through. So it's possible that there are still going to be some future grappling over parts of the contract like that. And what did you make of Andy Jassy's reaction? He tweeted out yesterday, posted on X, very interesting announcement from OpenAI this morning, which felt like a post that he actually personally wrote. This was not like a comms team post. What did you make of it? Yeah, I mean, it's hard not to read that as him gloating a bit that, you know, Amazon finally got its way and gets to sell OpenAI's models, which, as we know, Amazon has wanted for years. It's also interesting, you know, our colleague Kathy Perloff had a story on Monday just about a lot of customers of Amazon saying that they essentially didn't really care whether OpenAI's models were on Amazon, in part because, you know, in the last few months, Anthropic has gained a bit more mindshare and is in some ways seen as a leading model provider, although they're neck and neck. So I think it will be really interesting to see just how robust that demand is on AWS for OpenAI's models. And also if it can help OpenAI, you know, boost its revenue and make up for some of the hit to its revenue it'll take. from having to pay Microsoft more under the newly negotiated deal. Right. And we, of course, know that AWS and OpenAI have their event today that we will cover tomorrow. Very excited to see what gets announced and discussed there from leaders of both those companies. Aaron, I want to thank you for coming on. That is Aaron Holmes, our Microsoft reporter, here at The Information. Google has signed a deal with the Department of Defense for its AI models. Despite opposition to the work within the company, I want to bring on Erin Wu, our Google reporter, to share with us what she knows. Erin, tell us about what you found in your reporting. So Google has signed a deal with the Pentagon to put its AI on classified systems. And this deal allows the Pentagon to use Google's AI for any lawful government purpose, which is really the key factor in the deal. So is this deal, is it a brand new deal? Is it a revision of an existing deal? Walk us through that. Yeah, so it's a revision of an existing deal. So Google had a deal with the Pentagon to use its AI for any waffle uses. And so what's important about this deal is that it's moving it onto classified systems as well. The previous deal was just for unclassified work. Right. Now, how are Google's employees reacting to this? Yeah, so I mean, Google's employees found out that the deal had been signed at 10 p.m. last night when I broke the story. So a lot of them probably don't know. But Google employees are not happy. So yesterday, like the day that I wrote the story about this, 600 Google employees signed a letter to CEO Sundar Pekai asking him to reject this deal because they were saying that the only way to make sure that Google AI is not misused is to reject classified work because they can know what going on with classified work And how does that response put this into historical context So Project Maven was the other government work that you've come on the show to tell us about. Remind us about what the response was to that. You know, I'm thinking about the 600 employees. Is that more or less than the response to Project Maven? Yeah, so it is fewer employees than the response to Project Maven. So Project Maven is basically the primary example of tech activism. Thousands around, I want to say 3,000 Google employees signed an open letter basically saying that Google should not be in the business of horror. This was in 2018. Google had signed a contract with the Pentagon for Project Maven, which would have involved using AI for drone targeting. Google declined to renew this contract and put these AI principles banning use cases like weapons and surveillance in place after this Project Maven incident. In 2025, after Trump took office for the second time, they actually changed the AI principles and walked a lot of it back, including these two weapons and surveillance clauses. And so, I mean, it's been a huge shift over the past eight years. And a couple of weeks ago, I wrote a feature that kind of dove more into this and like the creation of Google public sector and how Google has really worked to build up this government business. So I wonder, I mean, look, it's been a long time now since Project Maven. A lot has changed, including like you said, the Google public sector team is, it looks very different now. What is, what did the company say in response to you or to the employees around these questions about the contract? Did they give any language there? Yeah, I mean, Google, I don't know if Google has said anything to its employees yet. Like as far as I'm aware, they have not. What they told me in a statement late last night, essentially they said that they are proud to work on national security. And they had an interesting part in their statement that mirrors the language in the contract, which is essentially saying that, yes, they believe that AI should not be used for autonomous weapons and mass domestic surveillance. A lawyer that I talked to said that that's really interesting language, how it is in the contract is not intended for and should not be used for. And basically what this lawyer was saying, though, is this does actually not prevent these use cases. This is kind of saying should not be used for is like saying, yeah, we think this is bad, but it's not a violation of the contract technically. Now, one of the interesting parts of your story that you wrote is you said Google's deal terms appear to be more permissive to the Pentagon than the terms that OpenAI agreed to in February. Yeah. Unpack that a little bit for us. Yeah, and I want to start by saying that this may be kind of a distinction without a real difference. The main thing that lawyers were saying about the OpenAI deal that was signed in February was that the most important part is this any lawful purpose framing, which kind of allows the Pentagon to do anything under the law as currently written. And that's the same thing that Google agreed to in total. couple other points that I want to highlight. One of them is this should not be used for language whereas OpenAI's was more shall not be used for and will not be used for which lawyers told me is stronger. The other point I want to note is that there's this provision in the Google deal that says that Google at the request of the government has to change its safety settings and filters at the government's request. And so this raised alarm with some people that I was talking to. I do want to note that Google's line on this is kind of saying like, well, this is actually standard. Like the filters are for consumer use. Like we change this for enterprises all the time. So I do want to note that, but it's something that is in real contrast to the OpenAI contract, where at least what they said publicly, and they made a big deal out of this was like, oh, we maintain full control over our safety stack. And so they were saying, it doesn't really matter what's in the contractual language because we can put whatever safeguards we want into the model, but Google does not even have that. So big picture here, Erin. We don't have a lot of details, I think, on the size of these contracts unless you want to jump in here. Yeah, it's part of the contract that was signed last summer. So it's a $200 million ceiling. And so that's like the overall contract. My understanding is that this one's less than that. But so a substantial amount of money, but like very, very small, vanishingly small in the context of Google's overall business. Okay. Yeah. Well, so that's where I was getting at was the size is what it is. I mean, just remind us of the bigger importance here of Google doing work with the government in this climate, competing against the other AI labs. I mean, why are these deals so important for the company given their relative size? Right. So this is obviously not that important to the company from a monetary basis. I think the real importance of this is it shows the moment in tech that we're in, in terms of how companies are willing to work with the military, how companies are willing to work with the Trump administration. There's great reporting in the past couple of days done as well about Google co-founder Sergey Brin's rightward shift. And then I think the other thing is that, I mean, this is important for people who live in the United States. This is important for people who live in countries where the United States military operates. I mean, like, these are deals that could have, like, real impact in how, like, war is waged and how the military operates. I mean, like, we already see that the military via, like, Maven Spirit Systems, via Palantir, like, is using Claude and a lot of things. Like, this deal now allows them to use Gemini. And so, I mean, like, that, I think is also a reason why people should care about this beyond the like, this is the monetary value or like, this is what it says about kind of the political climate attack. It's like, no, like if you are a person who is impacted by the United States military, which could kind of be anyone, like this is a story that's relevant to you. Do you think this sets a precedent for more deals to come? In other words, I mean, you know, we've heard so much about, about Anthropic, OpenAI, and Google's negotiations with the government. I mean, these are the big companies. I'm sure that there are not just defense companies, but look, all these fast-growing startups, they're probably vying to get some government business as well, you know, for the same reasons. What kind of a precedent do you think this sets for that? Do you expect that people will cite the language here and try to replicate it somehow? I think maybe. I mean, like, I think what's, so one of the things is like, there are not that many frontier labs left to strike a deal. So Google just struck a deal. OpenAI has one. XAI has one. Anthropic very famously does not have one. I think it's possible that they've reached some kind of agreement, but I think that that is kind of being negotiated very much on its own terms and kind of like the ship has sailed in a lot of ways for them to just be like, yeah, we'll take the Google deal. What I would expect is that this, if anything, maybe shifts the Overton window of like what companies feel as if they can do. Like when OpenAI signed their deal, they were already making a big deal of like, oh, like we want the Department of War, the Department of Defense to like offer these same terms to other AI companies. And so like what Google did is kind of similar to that, like has again like these changes as i mentioned so like i could see yeah like another company coming in and like trying to do something similar but there are like not that many like ai companies like left to do something like this like meta is an obvious company but like meta has also been really struggling with its ai models to the point that i don't know if they are like lining up to be the next one to sign one of these kinds of deals yeah i think i think you have to focus on getting the product right first before you worry about selling it. But who knows? We'll see how it all plays out. Aaron, I want to thank you for coming on. That is Aaron Wu, our Google reporter, here at The Information. Opening statements start today in the trial that pits Elon Musk against Sam Altman. Elon Musk claims that OpenAI has abandoned its original mission as a non-profit and has turned it into a, quote, wealth machine. Musk provided a significant sum of seed money to OpenAI for that mission and is now seeking damages. OpenAI says that Musk's claims are a baseless attempt to disrupt its growth. One of our AI reporters, Rocket Drew, is in the courtroom. He's going to be keeping us updated on the trial all week long. Here is what he had to say about jury selection. It was a thrilling day. The court that's overseeing Elon Musk's breach of charitable trust lawsuit against OpenAI has selected the nine jurors that are going to decide the outcome of the case. So those jurors include two men and seven women, and we know a bit about them. Many of them are immigrants coming from countries including Mexico, the Philippines, Guatemala, and Pakistan. They come from a range of different career backgrounds, including a caretaker, a nurse, a painter, and a couple people who are retired. The judge was very careful to ask people about their attitudes toward AI. Are they positive? Are they negative? Two of the jurors said that AI is actually giving them more work to do at their jobs because they don't trust the outputs from the AI systems, so they have to spend more time than ever double-checking it. One person said they use AI at work and they feel broadly positive about it So some big decisions now rest on this collection of nine people And what I watching out for next in the coming days at court are when we're going to hear from Elon Musk's expert, a computer science professor at UC Berkeley, who will be an expert on AI, the technology itself. And soon we should hear from the man himself, Elon Musk. So stay tuned for that. that was our ai reporter rocket drew here at the information the information held our flagship finance conference yesterday at the new york stock exchange hosting a who's who of ai's biggest financiers and finance stakeholders on stage it was a great series of discussions that spanned ipos neoclouds data centers debt markets and more i want to bring on cory weinberg our deputy Bureau Chief of Finance for some of his reflections on the day. Corey, welcome back to the show. It's great to have you here. Hey, Akash. All right. So I want to get to all of your reflections from the day. I do want to play a quick clip from one of the sessions. Let's take a listen. I think those IPOs aren't going to all be created equal. I think that people are more bullish on some, more cautious on others. I think it's potentially, there's a chance, and of course, None of us know for sure, but sort of my working model is that it will create a lot of caution. The performance of some of those IPOs may actually be like real cold water on reality. And I think there's a state of the world where it actually makes the capital around it more cautious. Do you think that's a good thing, bad thing overall for the? I mean, I think with anything, there's great decisions being made. And I think there's decisions where, you know, you're giving companies hundreds of millions of dollars of valuation when they have 100K of ARR. There's probably a reality making moment that happens, which is good. I mean, it happens every few years. We see this happen in the market. And I think sort of our base case is that the market grows caution. But we'll see. And again, not all of those IPOs will be equal. okay so a possibility maybe for some cold water that was one of the things that came out of it what were your broader reflections on the day cory my takeaway from the day was that all of these massive pools of capital that have been financing ai are so interlocked and interwoven together that it's going to be a huge story when the mood shifts, whether it's because of a sort of struggling massive IPO or some shift in sentiment in the debt markets. This is the story in markets and finance right now. It's AI and it is impacting venture capitalists, but more so it's impacting public markets investors and credit investors. And right now the mood is peak buoyancy. And I think we started to hear on stage folks say out loud, what happens if the mood shifts a bit? When you say what happens if the mood shifts a bit, tell me a little bit more about what you gleaned from that. Yeah, I think right now you have, you know, sort of, let me take a step back. I think this is the second year in a row we've had this sort of theme of financing the AI revolution. And it's just really useful to gather people in person and talk through, you know, sort of the issues of the day and what they're thinking about, obviously. But if I rewind back to a year ago, Wall Street wasn't fully on board with the AI revolution yet. Like, it was, there was like more hints of skepticism. You know, CoreWeave was about to go public. this time last year, and the mood on Wall Street was a little bit like, we don't know how we feel about this company, how much debt it has, and how much its growth could keep going. Fast forward a year, it was striking how much the bankers, the debt investors, the asset managers on stage were all systems go. You know, their bonuses are now tied, you know, hugely to the next deal and ensuring that this AI money train keeps going. And that always just gets me wondering about sort of how they're managing the obvious risks that come with the side of the market that OpenAI, Anthropic, and SpaceX represent, which is, you know, leading edge of this AI revolution, but very unprofitable and volatile and unstable. And that just creates this whole really interesting mix of issues to kind of work through. Yeah. You know, one of the interesting set of discussions that I thought were really helpful were around not just the neoclouds, but the AI infrastructure story at large. And to your point about how interconnected these things are, I mean, this idea of investment grade debt and the non-investment grade debt was something that kept getting repeated. And it sort of it really clued into me towards the end of the day that this this link between, you know, hyperscalers making commitments, it actually is the mechanism through which smaller companies and startups can also take on more leverage, which to your point, again, if any of this goes already along the chain, right, we not only have the hyperscalers implicated, we not only have the data center companies implicated, it's startups too, you know, and then it's venture capital funding. And so it was shocking to me, you know, how much people were taking a very wide scope of view across this ecosystem. Yeah, that was a good, that's a really good observation, Akash. And I think you've had a lot of the markets kind of meeting the technology in terms of what its needs are, which are a lot of capital and a lot of debt capital. Something that was striking to me on stage was billionaire investor Glenn Hutchins, who's on the board of CoreWeave, really basically putting it pretty bluntly and saying, if you're an investor and you're not exposed to AI, then you are just sort of totally missing the boat. this is the next industrial revolution. You know, this is you also had another investor from, I think, Tomasik talking about how, you know, a lot of the sort of debt financing solutions that he's watching to remind some of the advent of the LBL, the leverage buyout, you know, sort of that that kind of came of age in the 80s. And so, you know, folks on Wall Street, I think, are pretty proud of how they're meeting this moment and are able to use essentially the cash flow streams and the credit stability of the largest and most important tech companies in the world, a lot of the Mag7, and use that to sort of backstop and provide certainty around the less profitable, less credit worthy, less investment grade part of the market. Right. Let's go back to that initial clip that we played at the start. This was one of the speakers on stage talking about some of the ways in which the SpaceX IPO or these AI IPOs broadly, if there is cold water, what that could mean for the sector. We also had, I know you moderated a panel on IPO specifically, and you had a speaker talking about the implications of what these IPOs going bad could mean for the AI sector at large. I thought that was a pretty, you know, interesting point here, that there are many ways in which the AI IPOs could go wrong and only a few ways that they could go right. I think that's true. I think this is going to be, I think the story of a lot of these IPOs, because they're so large and by these IPOs, I mean, obviously SpaceX kind of coming most near term in June, Anthropic and OpenAI, more loose timelines, but we expect to kind of see movement on those this year or early next year. And a lot of the story wrapped up in these IPOs, you know, sort of is around the technical trading dynamics and this question of can the market absorb these companies, which will have this very new dynamic of coming to the market as new issuance IPOs as some of the largest companies in the world, you know, with trillion-dollar valuations and how investors sort of have to absorb that is like a huge question mark. You know, and I think, you know, sort of panelists were able to, I think, sort of lay out a pretty clear, you know, sort of bull and bear case on sort of how well these could go or not go, as you saw in the clip. Right. Well, Corey, it was a great event. I want to congratulate you for putting together such a great lineup of speakers and hosting a great series of discussions. I encourage people to check out our coverage on our website, and I'm sure we'll be talking about it more and more in the weeks to come. That is Corey Weinberg, our Deputy Bureau Chief of Finance, here at The Information. The chip news continues to come fast and furious Just this week there have been reports that OpenAI is looking to work with Qualcomm and another chip company MediaTek on smartphone hardware We've also done a lot of reporting ourselves on the GPU supply crunch. Joining me now to break down the current state of play in chips is Ben Paladian, founder and CEO of BEP Research. Ben, welcome to the show. It's great to have you here. Yeah, great to be here. Long-time listener, first-time caller. We have a great program and I'm excited to talk about the stuff that I cover in my research space. Wow, we're excited to have you on. So look, let's start with the news yesterday. So OpenAI potentially working with Qualcomm and MediaTek. These are not the GPUs that have really encapsulated AI, but it was certainly a market-moving bit of reporting. What did you make of that? There was always a lot of talk. I think OpenAI is still trying to find its strategy of distribution. Obviously, they had a device that they tried to make of Johnny Ive, and now they're trying to enter the phone segment. But getting into something like that at scale is very challenging, and it's something that Apple and other handset manufacturers have a strong chokehold on at the moment. So talks for now are very preliminary, probably. Right. Right. Well, let's go to a couple of the other, you know, bigger chip stories. You know, one story that I was curious to get your take on is we had Glenn Hutchins on the show yesterday. He was the co-founder of Silver Lake. And I asked him about his travels to Asia and the rivalry between the U.S. and Asian AI ecosystems. And he made the point, he said, look, China is learning how to do more with less in an environment where they cannot get a hold. of the best chips in the world in NVIDIA. What's your sense of that? I mean, you know, as you studied the Chinese chip landscape at large, I mean, do you think we're underestimating them? Yes, because if you go back a few weeks ago to the Jensen Dwarcush podcast that created the huge discourse, China has an abundance of electricity, but doesn't have the compute or the chips to run them on. So what they're essentially doing is trying to parallelize H100 early stage type NVIDIA chips, which is at like a seven nanometer scale, and put them together with very sophisticated algorithms and weights and methods of compute to use breakthroughs in that to leapfrog what we're doing here in America. So we have more horsepower, less actual electricity. They have a lot of electricity, but less horsepower, and they're working with those constraints. And the Asian open source labs from, you know, ByteDance or Alibaba Kuen or even DeepSeek V4 demonstrated that, and they said they trained all of those new models with Huawei chips, essentially. So how do you think this ends up playing out then? it's going to be a bifurcated world of compute and standards i think that's what jensen has been saying for a while just because if we give them chips it doesn't mean it's the end of the world the whole point is to get everyone on the cuda standard cuda itself operates on nvidia silicon through the chips and the whole system as a factory and it improves and it does all those things and at the same time the United States can monitor what's being done with those but if you're leaving everything to China develop their own semiconductor ecosystem you're going to go with those constraints and build those chips and build a competing uh system and try to sell that uh basically these these basically the same Chinese playbook that we've seen over and over And they take any technology and they deflate the cost and they go around the world to frontier markets like Africa or Southeast Asia and say, oh, do you need AI? Oh, you can't afford NVIDIA? No problem. We'll lend you all this money. You can buy all those Huawei chips. We'll build you a data center and make sure you run on DeepSeq or Quen or whatever it is. And that's it. America lost on the AI opportunity there in those markets. And then in those models, they put their own flavor of whatever they think history is based on the Chinese perspective, as we've seen before. I want to get your thoughts on a couple other stories that are trending. So Intel reported earnings last week, and their stock is, I mean, it is the stock to own right now. What do you make of their comeback here? it's uh if you look at my history on x you know i've been invested in video since 2016 because that's when dgx1 came out and the challenges of intel is they they lost the whole uh mojo of the gpu and in ai and now through all this management change and having the government taking a stake in it now and even an nvidia has a stake in it they're kind of in the right place the right time with the right product it's more of luck i don't think it's more of lipboop tan management which everyone talks about it's because if you look at the 2025 gtc keynote jensen showed the next inflection in ai compute is agentic ai and to control the multiple agents and tool calls it's not a gpu based thing it's it's a cpu based idea so everyone's running rushing out and saying okay, what's the next bottleneck? It's CPUs. Who makes CPUs? It's Intel, AMD, and then obviously ARM that licenses cores for those types of CPUs. So this is something I didn't appreciate, but so the agentic workloads that we talk about, those are actually even more dependent on CPUs than sort of the, you know, I guess the 2023, you know, AI workloads, the chatbot workloads. Correct. Yeah, because we're going from AI to actually doing things for us and opening up apps and orchestrating tasks. That's more of a CPU-based task. And because it's AI controlling other AIs, to do that controlling, you need a CPU. And that's something that, again, I tell everyone to just watch GTC keynotes and listen to every word. I think that the new NVIDIA NBL, the new Verirubin system that's coming out later this year has a Verirubin CPU that NVIDIA also designed on an ARM core, which is specifically designed for these agentic tasks. So like they're already there for that and they have the whole stack. So I think that's something that people are missing the forest through the trees again. So you talked about, you know, the demand for CPUs. GPUs, we've done reporting on the GPU shortage or crunch right now that everyone is going through. We obviously know that memory chips are also in short supply right now in high demand. What about networking chips? You know, is that where you see the shortage going next? Or where are you looking ahead for the next bottleneck? I don't think there's a bottleneck in actual GPUs from NVIDIA. I think the bottleneck realistically is when you think of building data centers, it's not something that's done in a factory. It's actually taking land and getting permitting, building concrete walls and infrastructure and actual human labor to stand up a building and wiring new people. And we don't have enough, obviously, skilled tradesmen around the United States to really build these things at scale. Once those things are built, everything else, the last things that go in are pretty much the GPUs. I think people are missing that point. I think we have all those things. And as far as networking, there's been this big debate on CPO, which is basically silicon photonics, of taking light and electronics and mixing it together. because the whole point is as models get larger and we have more data and we need faster compute the question is how fast do you move the data between the gpu and the memory chip you need it to go faster because you don't want to wait as long so the question is does it go fast on copper or if it doesn't go on fast on copper anymore then what's faster than light the speed of light so we're trying to use light and i think that's more of a 2028 story which you'll see in probably in the Feynman generation of NVIDIA GPUs, which we're trying to move the industry in that direction for that in around 2028. Right. Great. Well, Ben, I want to thank you for coming on. That is Ben Paladian, founder and CEO of BEP Research here on TI TV. Thank you. Well, that does it for today's show. A reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. If you can't make it then, episodes are available on theinformation.com, our YouTube channel, or wherever you get your podcasts. Make sure to follow us on social media on X, Instagram, and TikTok. I am off for the next three days, but you will be in good hands with our senior finance editor, Ken Brown, and Rocket Drew, who you know very well on this show. They will be taking over the show for the next couple days. I'll be back on Monday. Have a great rest of your Tuesday. Have a great week. I'll see you soon. Bye-bye for now. Thank you.
