Trump Allows NVIDIA H200 Sales in China, IBM $11B Confluent Deal, AI's Midterm Threat | Dec 9, 2025

0:13Welcome, everyone, to the Informations TI TV. My name is Akash Pasritch. It is Tuesday, December 9th. We have got a big show ahead of us today. First up, President Trump is allowing NVIDIA to sell its H200 chips in China. The real question is whether Beijing even wants them right now. We will break that all down with a top strategist and our editor-in-chief very shortly. We'll then dive into a crucial story we published this morning about how AI backlash could shape the midterm elections. We also have a conversation with CoreWeave about its new software features, and we have an interview with executives from IBM and Confluent to discuss their$11 billion deal.

0:55And last but not least, we will close the show with news about the latest company to jump on the chat GPT train, Instacart, which will soon let you shop for groceries directly through the chatbot. It is a big show, and so let's get right on into things. President Trump has given NVIDIA the go-ahead to sell its H200 chips in China. He, of course, has cut quite the deal with the U.S. government taking 25 % of that revenue, he says. The FT is reporting this morning that China will limit H200 sales in China to make the domestic chip sector more competitive. That is, of course, a story that we have covered in great deal through our coverage at The Information.

1:34And so to break it all down, I want to bring on our editor-in-chief, Jessica Lesson, and and Patrick Moorhead from Moor Insights and Strategy. Good morning to you both. It's great to have you here. Hi, Kash. Thanks for having me. I'm excited to talk about this. Jessica, I want to start with you because it feels like we've kind of been on this train before with the H20. We had the cat and mouse game. We, of course, had our own reporting that China says we're not even really that interested in the H20 chip right now. And now, NVIDIA is saying we're not even making the H20. You know, it's kind of a standstill.

2:08It almost sounds like it's some kind of posturing a press release, Akash. But no, you know, when I saw this yesterday, a year and a half ago, I went to Beijing with then Commerce Secretary Gina Raimondo to watch her negotiate these export controls with her counterparts in China. And while it was a very different administration in a very different time, even then it was clear there was not much action going on. there was just a lot of talk and diplomacy. And I think, you know, it's a fascinating move. We've been reporting in the information for weeks that the U.S. government was leaning towards allowing these chips, obviously showing the lobbying power of NVIDIA.

2:49But there are many questions, and I think you nailed it. In the U.S., we tend to forget that Beijing is the other side of the coin. And, you know, this happened with TikTok and ByteDance. It's not just what Washington wants, it's what Beijing wants. Now, China does have a problem. The local Huawei chips and other domestic chips cannot beat these chips when it comes to training. And so I do think there is some possibility of China letting these U.S. chips in for training, definitely not inference. But right now, it's just a bunch of statements and we really don't know yet. Patrick, what do you think of this?

3:28So I don't know if I take this as reality yet, right? We've seen a lot of yeses, a lot of noes, a lot of posturing, a lot of negotiation. But I am surprised that we somehow went from the H20 being a bad idea to the H200 being a good idea. I think that people realize you're not going to build a next generation frontier model using an H200. You really need a Blackwell for that. And as Jessica said, even though Huawei Cloud Matrix is pretty good when it's completely networked, but it takes 6 to 10x more power, And it's a lot more difficult to train a next generation of frontier models. So until they start shipping and revenue gets cited by NVIDIA, I'm not yet a believer that this is a done deal.

4:31I want to talk about then, Patrick, what needs to happen then for this to be a done deal. What is the next step here? Yeah, so it looks like there is, I mean, there's always been friction between the Hawks, the Defense Department, and the Commerce Department and others. The irony is that even the H-20 was very much a green based on the federal government's BIS regulations about density and performance. But what I think has to happen is that the – I'll call it the observability and the tracking mechanisms need to be in place to make sure that these H-200s go to the right place. Listen, it's not a national security issue if the Uber of China or the Instacart of China is using H-200s to even do inference or create a decent model.

5:29I do believe that the paperwork exists. It just has to be implemented and okay. But until NVIDIA recognizes revenue, I think anything can happen. So, Jessica, we had Channer from our Asia Bureau on the show a couple of days ago, and we talked a lot about this parallel ecosystem that China is trying to build with its own, not just chips, but its own models. And it doesn't necessarily matter if it's not as good as the stuff in coming out of North America. It just matters that, you know, they're all using their own technology. My question for you is how much of a risk do you think this is? And also, I mean, long term here, I mean, we've seen this play out.

6:12China is going to catch up at some point, right, with their own chips. Completely. I completely reject this war analogy. I've said it before. I mean, I understand its purpose in galvanizing innovation and competition. You talk to any CEO of a leading tech company privately, and they say China's right behind and is going to be as good. and this idea of keeping them at bay is foolish. Now, it doesn't mean the U.S. shouldn't try to be as competitive as possible. I mean, another aspect of this that I'm confused about is the 25 % cut the Trump administration is saying it's going to take from the sales of these chips.

6:50Now, it's not unprecedented to have the government take a cut of, say, foreign defense sales from defense contractors. But the thinking there has been that the government has been doing some of the R &D and backing those technologies. That's not the case here. So I think this is also something that people should really be questioning and honestly seems quite unprecedented when we're talking about the private sector in the United States. Patrick, we hear a lot about NVIDIA's sales to China. What about all the other chip companies? I mean, they all have skin in the game, too. We don't talk about AMD sales to China or even the hyperscalers.

7:28We know that Google's TPU is getting traction. I mean, what do we know about their international efforts to sell it abroad? Yeah, so I was an executive at AMD for over a decade. And even back then, we had very clear restrictions on a metric called FLOPs, right, that was used to develop supercomputers that could do anything from weather simulation to nuclear simulations. So this is very much a thing for AMD. And it will also, I mean, if we ever see exports of something like a TPU, which I doubt we will, it would apply to them as well. And the restrictions are very clear on the schedule. It's a two-by-two matrix on performance and density.

8:14And that gets updated every year. But AMD very much is going to be part of this game. And when Intel in 2026, when they bring out some of their new technology, they will to some of the startups like Grok will be covered, too. So, Jessica, with your reporter hat on, and I say this because you are the editor-in-chief, but you're a reporter at heart, with that hat on, besides trying to figure out what the Trump administration is thinking with their next post on Truth Social or on X, what are the big strategic questions that you are trying to get answers to right now? Well, I mean, obviously, I think Patrick nailed it for will these chips flow?

9:01Will the revenue flow? Who's going to get them? I mean, all our reporting suggests that you can try and control who gets them, but China's getting the chips. So, and, you know, the deep seeks and the Alibabas and the Tencents, right? So I think there is a huge gray market as well. It's been well documented by many publications, including ourselves. And so I think that's worthy of more scrutiny because in the end of the day, you know, these models are very, very advanced in China. And that's also, and stay tuned for more reporting this week on some of the progress of some of these major model makers in China.

9:40As well as Huawei, of course. I mean, while we pointed out that they're not a viable competitor in training and probably won't be because of a range of export controls that are going to prevent that lower down in the stack, you know, they are making progress. And that is another important topic to look out for. Great. Well, Jessica and Patrick, I want to thank you for coming on. It is a fast-moving story, and I anticipate we have only just seen the beginnings of it. That is Jessica Lesson, our editor-in-chief, and Patrick Moorhead from More Insights and Strategy here on TITV. Okay, midterm elections are coming up next year, and politicians are already starting to think about how AI issues will start to shape voters' perspectives.

10:24Today, our Washington, D.C. correspondent, Sylvia Varnum-O 'Regan, is out with a story about how fear around labor displacement, electricity costs, and the impact of data centers could influence the 2026 elections. And I want to bring her on to talk all about it. Sylvia, welcome back to the show. It's great to have you here. Hiya, Kash. Great to be here. Thanks for having me. So I have to tell you, Kash, the backlash begins. I mean, you know, AI is on voters' minds. You've been talking to people all around the country that have been thinking about this issue. What are they telling you? Yeah, I think that this is a huge issue going into the midterm elections.

11:02And what I really wanted to do with the story was to look across the country and to better understand how these issues were playing out for voters and how certain candidates who are running for Congress next year are positioning these issues as part of their campaigns. I spoke with a congressional candidate in Georgia where there is a huge number of data centers. And she said that she was going to meetings where this was coming up almost on a weekly basis. A lot of concern among constituents out there about the impact of having these data centers and having so many in the works. There are several proposed data centers also.

11:44And she was looking to really make this a focus of her campaign. Similarly, I spoke to a candidate in Memphis and also talked to a candidate out here in New York who was really pushing for AI regulation around safety issues. Right. And so you're talking to all these people. These are people from both sides of the aisle. I mean, how do they imagine this manifests itself in the upcoming elections? Well, it is still very early, so I think there's a lot to come. But I do think that we're seeing, as AI becomes such a big part of all of our lives, we are seeing concerns about this technology, whether that's job losses, electricity costs, as you mentioned, also child safety and privacy issues.

12:33So these candidates, and I should be clear that the candidates I spoke to were Democrats. However, it's true that these issues are something that are of concern to both Democratic and Republican voters from my reporting. Because these issues around electricity costs are concerns for so many at this point. But there also are Republicans who have voiced concerns about AI. So you're seeing divisions in that party also about how to kind of handle this rapid pace of AI development as well. So we just had an election in November. Were there any candidates that really pushed this agenda? Did we see the impacts of it at all?

13:18I think this, I think these races was what sort of catapulted this issue and started to raise these questions about will these issues become more prominent voter concerns? Will they influence more voters going forward? Because you saw candidates out in Virginia and in New Jersey really focusing their campaigns around electricity prices and the impact of data centers and data centers, you know, having to pay their fair share. And this seemed to resonate with voters. Of course, there were other issues, too, that these politicians focused on in these areas, but there was a sort of current of these concerns being at the center of their campaigns, and they were successful.

14:06Now, whenever this comes up in discussion around politics, there obviously is the company angle to it and companies trying to push candidates that are in agreement with their own platforms. Look, the big tech companies, I mean, this is a story that's been around for a while, given that these big tech companies have been there for decades. The newer tech companies like OpenAI and Anthropic, they've only lived through a couple election cycles in their lifetime. How much are they involved in politics right now? Well, that was another reason why I wanted to do the story, because there is so many layers to this.

14:45And one of them is the angle around regulation. who controls AI companies and what they can and cannot do. And there's really a big fight playing out in Washington right now and across the country about this because the Trump administration wants to block states from being able to regulate AI. President Trump, in fact, posted on Truth Social this week that he intended to put out an executive order to this effect. And there have been several attempts made to get this sort of over the line. The tech industry is supportive of this effort. A lot of politicians in states, state governors, for example, including Ron DeSantis, are not supportive of it and say that states have the right to implement guardrails around this really important technology as they see fit.

15:36The administration is pushing for some kind of national legislation. Not sure of the details of that or how that would come together. But technology companies have come out in support of that. So you are seeing these companies sort of get involved. Executives from these companies are also getting involved. There are a couple of super PACs that have been formed, both to push this sort of anti-state regulation line and to counter it. And OpenAI's president, Greg Brockman, for example, has donated to the former, the PAC that's pushing back against state AI regulation and looking to support candidates.

16:17who are in line with their views on that. It is kind of funny to me to think about the amount of money that these companies are raising and will have to continue to raise to satisfy their operations. And then squaring that with whatever direction the political climate ends up going, it feels like a growing percentage of that may have to go to lobbying the government effectively. Look, I don't know if the percentage is actually going to grow or not, But it is kind of interesting to think about these companies fundraising to lobby the government in some capacity and the fact that that is totally something that happens.

16:58I mean, it's something that is not talked about sometimes. Right. Well, I mean, a lot of my reporting this year and why I think that my area of coverage is so interesting is because we've seen these companies really increase their presence in Washington, D.C. A lot of these CEOs are making personal inroads with Trump. And they certainly have increased their lobbying spend, absolutely, because they want to be able to push for this technology, to grow their businesses, to secure government contracts, and really to have fewer restraints on that. And so the states, which have been looking to pass legislation in lieu of any kind of federal standard, they pose roadblocks to these companies.

17:40And so they obviously don't want that. So there is a lot at stake here for these companies. Right. Well, Sylvia, thank you so much for coming on. We really appreciate it. It's an issue that we don't talk about enough, and I'm excited to have you back again on soon. Thanks. Appreciate it. Okay. CoreWeave announced some new features today, helping businesses manage and track the ways in which they use GPUs. CoreWeave, of course, falls into a category of companies that we have been tracking very closely at the information, the NeoClouds. Here to tell us more about the news is Henn Goldberg, Senior Vice President of Engineering.

18:14Chen, welcome to the show. It's great to have you here. Thank you so much for having me. Good morning. So I'm excited to talk about the new features today. Tell us a little bit about the mission control expansion that CoreWeave is unveiling today. Sure. So if you think about what's happening right now with AI is that the type of workloads requires a different type of architecture. That's what we've been doing for a long time is thinking about the stack as one. So not only being intentional about every part of the stack, but how we think about it holistically. Today, with Mission Control, we are actually providing visibility and transparency to all of our customers the same way that we have it internally.

18:57Which means the detection becomes simpler and they really have the data in real time that allows them to understand and really stop the guesswork of what's happening in their environments from metal to model. meaning they can see anything from the data center up to their workload. I'm curious, when you were talking to customers, what was the pain point that you had that prompted you to make this feature in the first place? You know, I was just talking with an enterprise customer yesterday, and they mentioned three things that they care about. The first thing is performance. Two reasons for that.

19:34Of course, GPUs are expensive, and they want to optimize and make sure that they utilize the environment the best way, but also thinking about the latency and how quickly they can respond to customers. The second thing is reliability. AI workloads are becoming more and more critical for our customers, and they want to make sure that they can provide the right guarantees to their customers. And last but not least is security. And really, those three things just come up again and again for all of our customers. That's top of mind for them. And that's really aligned very well with the value propositions of mission control.

20:12Now, I want to take a step out of the future and talk a little bit about CoreWeave more broadly. You know, one of the things that I always wonder with the NeoCloud class of companies is we know that they're the big cloud providers, and now we know we have the fast-growing cloud providers. How does CoreWeave think about differentiating itself against these giants? The advantage that we have is that we can really focus on what these new workloads demand. You know, there's a big difference when you think about an AI workload. If before every node or a machine would run a single task, now all of these tasks have to run together, orchestrated.

20:54And it means that, for example, if a single node breaks, the entire system slows down. Okay, so this is just one example of how things are changing. Also, the way we think about compute, network, and storage has changed. Before, we thought about it as a commodity. Now, performance, how much data is getting to the GPU is becoming a high priority. So the opportunity that we came to the market and to our customers is to say, hey, we are building now this new stack. that is really optimized for these new type of workloads. In addition to that, I think that our size is actually an advantage for us, which allows us to really partner with our customers.

21:34And that's something that we hear over and over again. There are customers appreciate their opportunity to work directly with engineering and really troubleshoot together those really complicated environments. Now, when I think of CoreWeave, I obviously think of NVIDIA and the focus that the company has on GPUs and helping customers access that compute. Are you working with other chips at all, like TPUs? At this moment, we are really driving that decision based on customer demand, and we are seeing the best results with NVIDIA. Our infrastructure is not tied to a certain chip provider, but we really appreciate the partnership with NVIDIA, and things have been working well for us providing that system.

22:21So the infrastructure that CoreWeave has built, it could be used with TPUs? 100%. And so has the company bought any TPUs? At the moment, we haven't done that. But if a customer would raise that need or we'll see a better performance or any signals from the market, we can, of course, explore new options in the future. And broadly speaking, I mean, I am just curious about how you think about the future of these chips. NVIDIA has the lion's share of the market. It has a very powerful chip. But, you know, speaking, you know, personally, I mean, how do you see this evolving over the next three years?

23:04Is it always going to be NVIDIA that owns the market here? Or do you see, you know, AMD increasing its market share over time, for example? I definitely don't think that I can predict how things will look like. What's important is that we'll continue to partner with our ecosystem to make sure that we deliver the best service to our customers. Great. Well, Hen, I want to thank you so much for coming on. It's a fascinating feature and a great time for NeoCloud. So we look forward to having you on very soon again. That is Hen Goldberg from CoreWeave here on TI TV. Thank you so much for having me.

23:41Okay. IBM this week announced it is buying data streaming company Confluent for$11 billion. Confluent shareholders were obviously happy with the purchase. It was a 31 % premium to what the company was trading at before the deal. Joining us now are Jay Kreps, co-founder and CEO of Confluent, and also Rob Thomas, SVP of Software and Chief Commercial Officer at IBM. Welcome to you both. It's great to have you here. Great to be here. Thanks for having us. So Rob, I want to start with you. Talk to us about the rationale here for the deal. We're incredibly excited about Confluent. It's an incredible team that Jay has built.

24:18And I'll give you an analogy. If you think about the Industrial Revolution, what drove all the growth was when the railroads connected factories to cities and to people. And that's exactly what's happening to AI right now. This is the new Industrial Revolution. and we need the rails that will make AI successful. Confluent and what they've built in terms of real-time data is the rails of this AI moment. We're incredibly excited to have this become part of IBM once we complete the closing process. Great. Well, Jay, I want to talk to you a little bit about the business that Confluent is in. If you were explaining it to someone who isn't in tech, how would you explain the data services that Confluent provides?

25:01Yeah. The service we offer is all about the real-time flow of data between the parts of the company. You can imagine if you're a big retailer and something is sold, what are the different software systems and databases and parts of the company that have to update or react to that? Well, it turns out there's hundreds. That's not just a problem for a retailer, that's a problem for a bank or an insurance company or a tech company. That's ultimately where the software was born. So, you know, this is all about kind of harnessing and reacting to data in real time as things are happening in a business, being able to build applications, AI agents that can react to that, use that data.

25:42That's what the software is for. And what is it that you feel you can do with IBM that you couldn't have otherwise done as a public company? Yeah. Yeah. Well, I think this is an opportunity to take this to a larger stage. IBM has built a fabric for applications, infrastructure, security that works across the different cloud providers out into on-premise and edge environments. Confluent had done a similar thing for data. The two things fit together. If you think about what's the future of these applications that these organizations are building, I think data is a big part of that. That's the thing each organization has to figure out how to combine with these new language models in the age of AI.

26:24And so putting those two things together was a natural fit. And of course, IBM has done a great job with open source. A key aspect of the platform they've built is having these kind of open standard layers that the world wants to build around. And that's true of Confluent as well. So there was a certain strategic fit element that I think was kind of obvious on both sides, even the first time Rob and I chatted. So Rob, I want to talk to you a A couple months ago, Jay and his team spoke on an earnings call about how AI-native customers were finding ways to actually reduce their spend on Confluent by managing the data themselves.

27:04I wonder how you think about that issue within IBM. I mean, how do you think about combating that issue or how you might be able to help with that issue? I think these things always go in waves. And we saw this actually when we did the acquisition of HashiCorp as well. Cloud consumption is going to go up and down based on the moment, the macroeconomics, the specifics of a company. That's always going to evolve. But when we acquire, we are acquiring for growth. And we kind of step back and we say, what does this look like over a five or a 10-year view? And in our view, Confluent and the capability they provide is going to be up and to the right when it comes to AI.

27:41Yes, there may be some modulations here and there, but the need for data to support anything that a company wants to do in terms of intelligence and how they operate, that's not going to change. And if anything, the demand is only going to go up. Part of what we've done in IBM, building on Red Hat and HashiCorp, is we've delivered this multi-cloud platform, partnering with AWS, with Azure, with GCP. And so as cloud consumption grows, Confluent's going to grow, and the other capabilities that IBM provide will also grow. Rob, I want to stick with you for a minute here because I kind of want to ask you about IBM's place in the AI story at large.