U.S. Investigates Chinese AI Companies, Google Cloud Revenue Rises 82%, Anthropic’s Unusual IPO Plan

0:13Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Thursday, July 23rd. Today on the show, the federal government is looking into whether Moonshot AI is using advanced American-made chips. We'll talk to The Information's tech and politics reporter, Leo Schwartz, who broke that story. We've then got some quarterly results to unpack. Google and Tesla reported it was CapEx galore. Investors in both companies were unnerved. We'll talk about both those earnings calls shortly. We'll then unpack our reporting on Stripe's latest financials. And to close out the show, we've got a closer look at Anthropik's unusual stock plan for its employees ahead of its public debut that we are anticipating to come very soon in the months to go.

1:00It's going to be a great show, so let's get right on into it. The information has exclusive reporting that the U.S. government is looking into whether or not Moonshot AI, the maker of Kimi K3, is using top AI chips from American companies. U.S. chip exports have, of course, been a highly regulated issue for years now. I want to bring on Leo Schwartz, our AI and politics reporter, to unpack his reporting for us. Leo, welcome back to the show. It's great to have you here. Thanks for having me. Nice shirt. You too. Mine's half-sleeved, though. So it's a summer linen and the summer of tech policy, indeed, in the land of AI.

1:41What do we know about what the U.S. is investigating with respect to chips and moonshot AI? It's the summer of the open-source battle over AI. As you know, there's been weeks of speculation about whether DC and the Trump administration is going to crack down on this rise of Chinese open source. A lot of people are worried that labs like Moonshot, which develops Kimi, are creating these models that will challenge anthropic and open AI. And there's been a lot of uncertainty about what the White House response will actually be. Will there be a ban of these Chinese models? Will there be restrictions?

2:17Yesterday was really the first shot by a Trump administration official, Michael Kratzios, who's the director of the Office of Science and Technology Policy, did this pretty shocking post on Twitter, or X I should say, where he accused Moonshot of distilling or essentially stealing Anthropics' information to train its model. What we discovered in our own reporting is that behind the scenes, the Commerce Department's Bureau of Industry and Security, which basically handles all export controls of sensitive technologies, has been investigating this issue of basically whether Moonshot, a Chinese company, and other Chinese companies have been illegally or improperly using U.S.

3:00advanced chips from companies like NVIDIA to train their models. But there's an interesting layer to this story, which is that when Michael Kratios made his post, he wasn't doing so with information from commerce and their investigation. The investigation is still ongoing. We report if commerce is going to take any action, it would need to be completed. And basically the upshot of all of this is that there's a lot of division within the Trump administration on how to handle it. There's a lot of confusion even in between the different agencies on how to act. And sorry, Kratzios' tweet, it was that they had evidence that they were distilling or that he thinks that maybe they were distilling?

3:38So Kratzios was pretty clear. It said, we have information. We have information. Okay. Yeah. But it would be separate information then, presumably, from the information that you are citing in your own reporting. So we've got sort of two different lanes here of the government pursuing this issue. Yeah. So this information would come from the Bureau of Industry and Security under Commerce, which does investigate export control violations. But the information that Kratios was going with was not from BIS. Right. Right. Right. And so, I mean, let's just talk about the consequences here. So if it does come to be found that Moonshot AI was using chips that they were not allowed to be using, these would presumably be the chips above and beyond the threshold of, you know, we know that export controls have been a big topic.

4:33The H200s were sort of the most advanced chip that they were allowed to sell. So we don't know exactly what chips they are investigating them using. But if it gets found that they were using it, then what happens? What's the consequence? Yeah, so Kratios at least accused Moonshot of using these more advanced NVIDIA chips through Thailand. And as our colleagues have reported, there are these intricate chip smuggling rings, or at least ways for Chinese companies to try and circumvent those restrictions. I think all this is being figured out in real time. time. But but the upshot of this is that if the Bureau of Industry and Security does determine that Moonshot was violating export controls by training its model using these prohibited U.S.

5:19chips, that could result in different actions. One that's being considered is something called the entity list, which is basically a designation that other Chinese companies like Huawei are already on, which would mean that these foreign companies are blocked from accessing any further U.S. tech. That could even mean cloud services. And that's something that's on the table. Although, as we reported, these discussions are very preliminary and obviously would require this investigation to play out. So whether that is one option that the Trump administration pursues against Moonshot or other Chinese labs remains to be seen.

5:54So entity list, that means it's a restriction on what the companies on the entity list can access in the U.S. would it at all mean anything in terms of US customers accessing the tech in China or wherever the entity list is from? So theoretically, it wouldn't. This is focused on how foreign companies can access US tech. But at the same time, when you look at how BIS treats companies on the entity list, it's essentially putting a red flag on it. And there's all sorts of other reasons that a US company might be dissuaded from interacting with a company on the entity list. And I think something else to keep in mind here is that a lot of this is very legally novel, this idea of software companies being restricted in these ways.

6:41A lot of these rules are written in mind more for things like weapon manufacturers or actual hardware. So I think agencies like commerce, like BIS, and like the Office of Science and Technology Policy are really figuring this out in real time. So the entity list is one option. Are there other consequences that they're considering at all? Yeah, so after Kratios posted on X about the information the government has about Moonshot, Scott Bessent, who is the Treasury Secretary, did a quote tweet. I don't know if we call it a quote tweet anymore, a quote post. Quote post. A quote post sounds less fun.

7:20And basically said that the entity list is one option, another is sanctions, just sanctions in general. And there weren't more specifics there. But this just goes to show that the Trump administration - That sounds like it would be a more serious consequence, though. Yeah. I mean, we don't know if sanctions would be monetary, if sanctions means restrictions from who in the U.S. can actually access these models. But that certainly is a big fear among a lot of both U.S. companies trying to access the models and U.S. open source developers. So we've been tracking this issue now on the debate that Silicon Valley and D.C.

7:55are having around how to respond to Kimi K3, to the power of open-weight models that have been developed abroad. The idea, should there be regulation against models developed overseas, regulation against open-weight models, period. What have been the developments on either side of that debate in the last 48 hours? Where do people stand on this now? There's a really interesting division line being drawn between people defending open source in general. Obviously, there's a lot of U.S. companies that are building an open source or use open source or defend it. And then also people defending Chinese open source itself.

8:36Maybe not directly, but there's figures like David Sachs or Bill Gurley who both basically say competition is good. Capitalism finds a way. If anything, we need to be pumping up the capabilities of U.S. companies rather than focusing on Chinese open source. But I do think yesterday was significant because really for the first time you see two very senior Trump officials, I guess Besant actually did also the day before, come out and criticize Chinese open source at a time when a lot of the tech community is trying to defend the ability of open source and make sure there aren't restrictions. But now you're really seeing these senior White House officials come out and use the distillation issue of Anthropic as the reason for saying we need to have some sort of restrictions or actions against Chinese open source.

9:24I got to tell you, the David Sachs perspective is kind of interesting to me here because he's obviously his term in government, whatever the cryptozar. What was the term exactly? AI and cryptozar. AI and cryptozar, sorry. So that term has now finished, right? But it's kind of interesting that I was sort of thinking last night about, do you think he could be as outspoken on this issue if he was still holding that title within government? And if he knew what he knew now about the position that certain people in government are taking, would he still have wanted that position or not you know because taking that position sort of to me felt like it was it was his reflection on the perspective that the government had taken towards crypto policy at that time and you know it was sort of like yeah you know i'm on board with this i think you know i'm here to sort of further the cause i don't know like i do you think he would have taken that position now knowing what he does about how ai policy is shaking out in the administration It's a good question.

10:36And I think we have to remember the timeline, which is his special employee designation ran out in March, which is when he stepped down. He still is in this sort of semi-official position called PCAST, which is an advisory board to the president. From all we understand, he is still incredibly influential in these debates and is spending time in DC. But I think the other important thing to keep in mind is when he did take that position and for really the first, say, year and a half of the Trump administration, AI was a more segmented issue that I think a lot of the senior principal officials in the Trump administration didn't really think about.

11:13And as we reported in our big story over the weekend, that really changed with Mythos in the spring and this idea that this growing realization, I think, among senior officials that AI was this huge national security issue. And now Sachs, rather than really being able to have his own fiefdom in the White House figuring out AI policy is competing with folks like Besant to try and figure out what the future of AI policy is. So would he have taken the position now? I think maybe even more so. That's true. Yeah. It's more of a turf war. It's more important that his view is able to win out over other very powerful voices.

11:49But certainly he's not able to have, I think, as much direct influence as he did before. Yeah, no, that's a good point. Last question for you. So now looking ahead, I mean, we had the executive order and we've been waiting now for, I think, the guidelines on AI policy are basically supposed to get determined here in the next couple of weeks, right? And then following that, we've got a meeting between China and the U.S. as well. How do you think the Moonshot AI, Kimi K3, the discussions that you've reported on, how do you think all of that is going to impact the framework for the AI policy that we're expecting and then ultimately the China-U.S.

12:33discussions? Yeah, well, as you said, we're supposed to have this framework in early August. What I'm hearing is that it's likely to be delayed. I think there's fears that all this China open source debate is postponing that, sort of how the export controls on Fable did. What was that a few weeks ago at this point? It's hard to keep track of time. So there's still, I think, a lot of optimism within the AI industry that the release of that framework, whether it be in early August or a bit later, will help create better rules of the road, at least for how US companies release models. But it won't answer the question of what to do with Chinese open source.

13:10maybe there'll be something in it about protecting against distillation, but that remains a novel question that has to be solved and would likely require some other type of policy, whether it's the entity list or sanctions or an executive order or a full-out ban. That remains to be seen. And also, as you said, there's now these looming talks with China on AI, which is set for September. I think one way you might look at this potential action is it's saber-rattling ahead of those talks and basically saying we like we have this power to crack down. So be careful. But at the same time, if they go too far, I think there's a risk of setting off a trade war or going to those talks in a bad position.

13:52But as we know, President Trump does like to have a forceful approach, especially in those types of bilateral talks. Yeah, well, it'd be an exciting dinner conversation for them, I'm sure. Leo, I want to thank you for coming on. That is Leo Schwartz, our AI, tech, and politics reporter here at The Information. We're turning to a couple big quarterly results. Alphabet and Tesla shares are in the red today as investor concerns about AI spend overshadow different areas of growth. I want to start with Alphabet. To unpack the results, we are chatting with Dan Flacks, Senior Research Analyst at Neuberger.

14:32Dan, welcome to the show. It's great to have you here. Great to be with you. Thanks, Akash. So let's talk about the numbers. What stood out to you here from the Google results? We're seeing very strong growth in cloud, which accelerated again. And I think that is fundamentally transforming this company. And as we look out later in the decade, you could see actually the cloud business approach the size of search. Of course, though, you have to pay for all that. And so you're seeing significant capex. They raised it again this year. Next year, I think there will likely be well over$300 billion in spend.

15:08And so this push and pull around the spending and the cloud growth, I think, will remain debated in the months and quarters ahead. But overall, we like what we saw, and I think the stock remains attractive at current levels. Remains attractive, but it's down this morning. What do you think investors are reacting to? I think it's the capex and the concern about this arms race and, of course, the negative free cash flow that results from this elevated capex spending. If we step back, though, and think about what's going on, the search business, I think, remains healthy. They are reinventing that again.

15:43And you have this ability to put longer, more complex queries into it. And so you get more value for the user on one side and more value for the advertiser with increased signal on the other. YouTube remains healthy. Their subscriptions are continuing to see nice growth and, of course, cloud. But the investment required and the concerns around when these investments will peak, I think, is pressuring the stock. I want to get into the ads business in a moment here. You know, as you – well, actually, let's talk about the ads business. So, I mean, the ads business, it's still double-digit growth. You know, I think it was a slight deceleration from last quarter, although it's exactly where it was two quarters ago.

16:28So, I mean, it's still very strong. Are there any caveats here to the ad growth that you see that we should be flagging? And I say that because, look, the numbers are strong. They also have more inventory now with the AI overviews, ads play that I think many are expecting to be another avenue for growth. I mean, what caveats should we be thinking about here? From a technology standpoint, there's some concerns that Google 3.5 Pro is delayed and the company's talking more about the Gemini 4.0. And we'll certainly learn more about that in the months ahead. So there are concerns around their ability to keep up with some of the leaders with the models.

17:14From my perspective, I'm relatively comfortable given their performance. And in my view, the most important thing is taking the models and taking the IP and translating that into the product roadmap in search. And that is where we're seeing strong results. I think the competition is certainly going to remain fierce. You have OpenAI trying to build a digital advertising business. And of course, others see this opportunity. And so Google has to keep investing. It has to keep experimenting. AI mode is seeing good traction. And I think the experience has to continue evolving. So it's going to be in the months ahead, I think, a combination of the roadmap around the models, new product features, and really continuing to transform search to make it a richer experience.

18:02How confident are you in the company's ability to get three, five, and then ultimately four back up to the top of those leaderboards that so many of us watch with respect to benchmarking? I mean, they've lost some top talent to OpenAI. Of course, they still have, you know, I think we were talking about it. I mean, they still have at least one or two Nobel Prize winners on their team. So it's not like they're hurting. But, I mean, they seem to have fallen out of favor here in the conversation around. So how confident are you that they can get those models back to the top? With respect to talent, I expect many of the top researchers and scientists, they're being poached by the existing large public firms.

18:46And, of course, the private firms, some of which, as we know, in the case of OpenAI and Anthropic, are enormous as well. And I would expect that to continue. That's been the history in the valley, and I suspect that'll be the case for the foreseeable future. In terms of the models, what I think is important is that Google has to try to isolate the elements that are most important for its businesses and its products. And I think they're doing that quite well overall. There's improvement needed in areas like coding, for example, to keep up with what Anthropic is doing. But I think for what Google needs to do, they're making very good progress.

19:23I think the key is really balancing the cost, the utility, and really the ability to infuse it into the product roadmaps. And the company has generally been masterful at that over time. I think the concerns, though, around the leaderboards for the models, that's going to come and go in the weeks and months ahead as it has been now for some years. But I think the urgency... You're expecting them to come back to the top of those leaderboards? I think they'll be amongst the leaders for some of their models. And for me, that is important. But what is most critical for the investment is translating that into revenue growth across its products.

20:03That matters most. Right. Hey, the headcount growth, they grew by 4%. Did that surprise you? No, they're continuing to invest in technical talent and researchers. We talked about poaching talent earlier. I think modest headcount growth is going to continue. If you're growing the cloud business, as one example, at over 80%, the need to invest in people and talent remains high. And so the key for Google is to continue to allocate its resources, its people most effectively and try to think through their ability to navigate the short-term pressures which all companies face while continuing to invest for the longer-term opportunity.

20:49And that's tricky. And, of course, you'll always have cyclical elements at play as well. Right. I mean, you know, and I hear you on the cloud growth, which is extraordinary. On the flip side, then you look at the margins and the company's burning cash, their operating margin also shrank a bit. And so that's sort of where I was surprised with the headcount growth, at least. I mean, you could see another flavor of tech company that lays off people to keep their margins in check. I mean, how do you square those two things? So a couple of things. In terms of the cloud, the margins there expanded nearly 1 ,500 basis points.

21:25So you're actually seeing very good leverage, very strong returns from the cloud business. More broadly speaking, though, the increase in CapEx and the associated depreciation, of course, compensation expenses, those are headwinds. And that's something the company is going to continue to work to optimize. I don't think there's an easy solution given the investments required to drive the growth. If I step back, though, and look at what's happened over the last several years, they've been building out new businesses. You're seeing it with cloud. You're seeing it with subscriptions in areas like YouTube.

22:02Search itself, it's this constant reinvention. And so the margin dynamics, it's something we watch very closely. But what will matter most, in my view, is the revenue growth. And finally, just to round it out on the free cash flow, we are in a period of aggressive investment. and I expect that to be the case for the foreseeable future. Great. Well, Dan, I want to thank you for coming on. That is Dan Flax, Senior Research Analyst at Neuberger here on TITV. Tesla reported second quarter revenue jumped 26%, but manufacturing and AI costs mean the company is now burning cash. For more on that, I want to bring on Ross Gerber, President and CEO of Gerber Kawasaki Wealth and Investment Management.

22:46Ross, welcome back to the show. Good to have you here. Yeah, thanks for having me. Okay, so what kind of a mood was Elon Musk in for this quarterly result? I think he said something that he was like under the weather. So he was like kind of bowing off sickness. And it reminded me of the old days in the old Tesla days when I used to send emails to them like, you know, it'd be better if you did the conference call with a little bit more positive energy, you know? Right. So, you know, it wasn't the typical call in the sense of lots of ridiculous claims and optimism. So the sobriety, I think, hit the stock a little bit today.

23:23So Tesla's got a lot of challenges in front of them. Well, and we'll unpack them in a minute. But what's actually coming to mind now is now we have earning calls probably two, three weeks apart with Tesla and SpaceX. And so we actually have more of an opportunity to track, hey, was this just a bad day for Elon? Was he under the weather? Or is this a mood that he's in, you know, this quarter? And so to the extent that people track that, I mean, that's certainly more data. Let's talk about the numbers for Tesla. So what numbers stood out to you for this quarter? Well, good or bad. You know, so the good numbers.

24:03Stock is down, so I guess the bad ones. Right. Well, you know, but the good numbers is that they're selling EVs. And I've been saying this a lot. Like, if they just focused on their core business, they're kind of crushing it. You know, like they don't advertise. The CEO turns off consumers and yet their sales are still very strong. They still make the best vehicle services revenue up 50%. So you can see that Apple effect of buying a vehicle and then all the ancillary revenue they're now generating through charging and other services they provide. So so the business model that used to be Tesla is continuing to work.

24:36They just stop focusing on selling EVs and sustainable transportation. So the bad is the fact that RoboTaxi doesn't work. And so there's this misperception that this massive scale is about to happen of all these cabs everywhere. But the fact of the matter is the software doesn't work yet. And so because of that, if you listen to the conference call, he does not want to be responsible for killing somebody. And rightfully, rightfully. Nor should Tesla be, you know, I use the software almost every day and it's not good enough to scale on a cab business as it is today. And so knowing that he's playing a risky game because if he kills somebody, people are going to want to shut this down.

25:20But at the same respect, he can't really scale and he's not getting the data he needs. And just putting one cab in a city with a small square is not scaling robo taxi. And so you see the increase in expenses involved with putting out logistics and all this kind of stuff and running a service, but the service doesn't generate revenue. So then there's that excess cost. And then Optimus is a cost that just goes on and on into the future where we just don't know when revenue will ever happen. And saying they're going to manufacture this thing in a year is just absurd. Well, and so they talked a lot on the call about building out the supply chain, and that's obviously what they're investing so heavily in.

25:57and the reason why they're burning cash now. I wonder, as you look back on how they built the supply chain for the core automotive business, I mean, that business, like you said, now they've built it. I mean, they've done it. Well, yeah, especially around batteries. Right. So, I mean, what do you think we can infer about this next big supply chain build from how they did the automotive build? I mean, does it suggest that they're going to do the impossible again? Is it a harder build? Like, how do you compare these two? Yeah, building a human versus a car? I mean, come on, dude. Okay, so I've spent a lot of time on this, a lot of time, because, you know, obviously I have to talk about it in the public a lot, and I'm not an engineer, so I don't like to approach things like, I know engineering-wise better than Elon.

26:50I definitely don't, okay? So, but what I know is about humans and I know humans better than Elon. And, you know, if you look behind me, you see these guitars and I can play guitar pretty well. And I've been doing it my whole life. And my fingers are just like amazing. Okay. You can ask my wife too, you know, like amazing. And so it's like trying to build this without any model, like God was pretty good at making humans over millions of years of evolution. hands are tough, feet are tough, and eyes are tough. And these are the three things that really make us special besides our intelligence. So you take out intelligence and you look at what makes humans really special, hands, feet, and eyes, okay?

27:35And they're not going to be able to build these things. So most robots have been built with wheels or with tracks or other systems than walking. And I think they're going to struggle trying to build humans. And this could take a very long time. And even after they master it, what's the application? To me, folding clothes, spending a trillion dollars to build a robot to fold clothes doesn't make any sense to me at all. Military, space, industrial applications make a lot of sense to me. So where this goes, I don't know, but they're spending a lot of money and there's no revenue coming anytime soon.