NY’s Data Center Moratorium, Microsoft’s New Security AI Overhaul, The American AI Dividend Debate

0:13Welcome, everyone, to the information's TITV. My name is Akash Pasricha. It is Thursday, July 16th. Today on the show, we have exclusive new reporting on DeepSeek. Our Asia Bureau has gotten wind of the company's latest financials, as it potentially looks to a Shanghai IPO as early as next year. We're then covering New York's attempt to limit data centers from being built in the state and President Trump's reaction. We'll also dive into a major shakeup at Microsoft. The tech giant security division is undergoing a significant restructuring. Microsoft reporter Aaron Holmes will join us shortly to unpack his exclusive on what is happening behind closed doors.

0:55And finally, we're going to take a look at the debate on whether AI companies should give Americans a share of their profits. We'll talk to OneWeepBC with a strong opinion on this. It's going to be a great show, so let's get right on into it. DeepSea could be marching towards an IPO, and the Information Publish exclusive new reporting this week on the company's latest revenue traction. I want to bring on Jing Yang, our Asia Bureau Chief, to walk us through what we know. Jing, welcome to the show. It's great to have you here. Hi, Akash. All right, so what do we know about DeepSeek's financials?

1:29Yeah, so we reported earlier this week that the company's annualized run rate is approaching$500 million. And that revenue mostly came from selling cloud-based access to their latest deflection model called V4. And I just want to give you the context. because V4 was launched in about three months ago. So if you're capturing that, yeah, that is not a small amount. What about margins? What do we know about profitability? So we know that the gross profit margin from selling cloud-based access to V4 is somewhere between 70 to 80%. And again, I want to give you the context to that percentage. We know that in general, Chinese open source AI models charge for excess token costs at such a tiny fraction of leading models from Anthropic and OpenAI.

2:31So if you think about how low that they're already charging customers for, this high gross profit margin is actually really remarkable. And we revealed in the story that they're able to achieve this high gross margin because DeepSeek has really strong capabilities into essentially, you know, making, running their AI models, running the inference of their models at a higher efficiency. because they apparently worked out how to, you know, run, you know, response to air currents and have that output a lot more efficient than, you know, on a single chip basis. Well, and Jing, I should point out, this is staggering.

3:16I mean, you know, 70 to 80 % gross margins. These are the software gross margins that we are so familiar with on this show. And notably, it's the margins that no AI lab has been able to even get near. I mean, we're talking about even having positive gross margins on this show for some companies. So that is very much a staggering finding. Tell me, if the financials of the company are so strong, at least on a unit economic basis, why is it then looking to raise money again? Well, so let me just say that there's actually a lot of people waiting to see if they can actually keep up this kind of high gross margin, number one.

3:58Number two, as you know, we're living in an era where there's a new model, you know, that drops every couple of days. So the company, you know, with V4 that has been super popular, they cannot just say still, right? They need to continue to invest in compute, in data, in talents, in order to continue to improve and scale up their models. And this is why they have been raised. Right. Now, last time we had you on the show, well, might not have been last time. When we had you on the show previously, you talked about the first funding round that DeepSeek was looking to raise. And I remember our conversation.

4:37You were hitting home the point how unique the structure of that funding round was and how important it was to DeepSeek's founding team to preserve control over the company. They obviously have a very specific ambition here to keep things open source. is this second funding round is it as unique is it the same you know covenants around the fundraise that we talked about last time i mean yes and no so uh let me just step back a little bit um so they closed on raising you know about um 7.4 billion dollars uh equivalent so 50 billion just about a month ago, and now they are raising the same amount of money is their target, but at a higher valuation, at about like a 30%, 35 % higher valuation.

5:32I'm just saying like nothing has happened in the last four weeks, but they're coming out, raising the same amount of higher valuation. And we actually reported that the company started plotting raising the second round before the first round was officially closed. and then but in terms of a question about the structure so this time they are actually becoming a lot more open um so they are so there are two major changes um compared to last round so this time they are going to allow they're going to utilize um there or they're going to let investors utilize this special structure allowed under china's um stringent foreign exchange control which essentially may give foreign investors who invest with US dollars or any other currency, but not the Chinese yuan, to convert their capital from outside of China into the Chinese yuan and invest in PNBC deals in China.

6:33This is a pretty well-established route, but it just wasn't allowed previously in DeepSig's first round, which was predominantly funded by Chinese investors in Chinese yuan. And then the second change in the second round of fundraising is that they're also going to allow special purpose vehicles, which, again, is pretty common in big VC deals, that SPVs essentially are vehicles that allow you to, You just create a legal entity where you can pull money from multiple sources of capital into one stock, into one legal entity, and then invest. And the last time, DeepSync was pretty cautious about use of that kind of structure because they want to know where exactly is the money coming from.

7:25But this time, we can see that they are loosening up a little bit, or pretty much actually quite a lot. And we also reported that they particularly are interested in attracting investments from the Middle East. So with these two new changes I just talked about, receiving investment from the Middle East or anywhere in the world, maybe except for the U.S., will become possible. I want to ask you about the IPO plans. What do we know on that? And also, maybe, Jing, help us understand, broadly speaking, it feels like DeepSeek has really broadened its ambitions, at least in terms of fundraising, certainly an IPO, over the past couple months very quickly.

8:10What's behind this new, wider ambition for the company? And is it any different based on what the product is hoping to be? So the company has already hired investment banks to start preparing the paperwork and what else that is needed for an IPO. They're hoping to file the application this year. And then if it's approved, then they're hoping to lease as soon as next year. And the background here is that in China, stock market, any company that wants to go public has to go through this so-called guidance process, process, which essentially the first step is you have to hire investment banks. And these banks basically then act as your teacher and teaching you what do we need to do in order to be IPO ready.

8:56And that process usually takes quite a number of months and even up to a year or longer. So they are already sort of in that preparation stage. And then to your other question, the simple answer is that we don't know if anything changed. which we reported just a month ago again, that after the closing of the milestone first round, that they are very AGI-peeled, that they are not rushing into commercialization. So our understanding is that this sales and the margin, despite how remarkable they are that we just talked about, it's not something that they deliberately, you know, trying to pursue. They weren't like, you know, going all out marketing or whatever.

9:44to get all of this revenue. It's just like, you know, sitting there and the V4 is popular and the people want to use it and then the money was coming. And then we will be obviously closely watching this and I would love to come back again to talk about whether anything has changed in terms of mandate or the mission of the company after we find out. It's the next big reporting question. That's certainly for sure. Jing, before we let you go, I want to ask you very quickly about a story that Channer Liu, our colleague, published about asml uh her reporting suggests that the the the chip equipment manufacturing company this is a different brand of company that we've talked about the show they make uh the tools that make the chips they are now open to raising their prices what do we know about that so asml is a dutch company and then they have an absolutely unique position in the global semiconductor conductor manufacturing chain.

10:41They are the only supplier of EUV extreme ultraviolet lithography machine that is essential in the production of advanced chips, especially AI chips. You didn't even stutter over that. You just said that as if it was like a MacBook. That was impressive. I did rehearse before I got up. Okay. They make the tools. They make the chips. But let me just explain very quickly what these machines are. Think of them as extremely sophisticated, complex, intricate printers that can print transistors of the width of just a couple nanometers onto light-sensitive silicon wapers. Okay. Okay. So, and then these machines, these machines sell at least a couple hundred million dollars a piece.

11:40And then now, the company, you know, ASML now wants to raise prices of these machines and their customers are companies like TSMC, Samsung, HK Hynix, and Houdan, whose customers are, you know, NVIDIA, Apple, et cetera. And so let's talk about the reaction from the customers. I mean, what does TSMC think of this? So TSMC has resisted in ASML's demand for raising the prices. But then ASML wants to raise the prices because, you know, this is how the whole thing works, right? The explosion in demand for AI has caused explosion demand for chips, which then has caused explosion demand for manufacturing capacity of the chips.

12:31And TSMC on that end has already committed to expanding their manufacturing capacity, which is like a years long endeavor. But in order to expand their chip making capacity, obviously they need to order more machines to make their chips. And then this is where ASBL comes in and they are seeing as well surging demand. And then they also face pressure to invest, increase their capital spending. and that's why they're coming to their customers asking for more money. Right. Well, Jing, I want to thank you for coming on. There's a lot to it and there's a lot more to find out as well. We look forward to having you on again.

13:13That is Jing Yang, our Asia Bureau Chief here at The Information. President Trump is criticizing New York Governor Kathy Hochul for a new executive order that temporarily pauses environmental permits for up to one year for large-scale data centers. The president wrote in a Truth Social post that the policy should be changed immediately. Our tech and politics reporter, Leo Schwartz, is all over the story. I want to bring him on to talk about it. Leo, welcome back to the show. It's great to have you here. Great to be back. Okay, so tell me a little bit about this executive order from Kathy Hochul.

13:49So the executive order is the first time we're seeing a statewide moratorium on new data centers. We've seen over 300 such actions at more of the local or county level. Our terrific journalist Shane Burke did a whole data infographic about this that you should check out. But Hochul's order is the first time that we're seeing this at the statewide level. It lasts for a year. It's just a pause. And like you said, it would pause any permits from going out. Interestingly, the state legislature in New York had passed a more severe restriction that would have lasted two years and encompassed more types of data centers.

14:28But Hochul did this as almost an intermediary, something that can be done for just a year as they work out a more long-term solution. So were there a lot of data center plans in New York? I mean, we live in New York City, and I try to get out of the city as much as I can. But, you know, I haven't seen them, but I guess there's a lot of them upstate. What's the deal? You can go for a nice scenic drive in the Hudson Valley and try to find your own data center. I was going to say, I was like, I hope they don't ruin the Hudson Valley. I'm a big fan of the Catskills. New York is one of the most active states for data centers.

15:03The Pew Research Center found that it was in the top 10 states, which we referenced in our article on it. And it's also one of the top states for a lot of concerns over what data centers could mean. And I think one of the most interesting dynamics here is that Hochul is a Democrat, but she's certainly a more centrist, pro-business Democrat. The way that she framed this moratorium is not what you're seeing in more extreme rhetoric against data centers and fears over AI or full-on opposition to AI. It's basically saying, let's have a pause so we can figure out the impact that data centers are having on communities so that we can try and work together between the companies building these, the hyperscalers, and local communities and make sure they're being done in an economically safe and responsible way.

15:48What's the reaction been to the executive order? Well, Trump didn't like it. I know he didn't like it. He took the truth social to practice it, which might have been expected. besides that data centers are a really interesting issue uh they have both bipartisan support and extreme opposition this isn't just a right left issue uh you've seen a lot of right-wing states or right-wing led states i should say like florida and texas consider their own moratoriums well so and so let's i mean let's talk about that so so trump didn't like it but but it seems like there's support on on both sides to this issue to some extent depending on what state you go to You're seeing a strange alliance form.

16:31It's very multipolar in terms of where people stand on AI and on data center buildouts, where a lot of the anti-regulation people normally, like Ron DeSantis, the governor of Florida, are also wary of the economic impact or job displacement or other fears of AI, as well as just what it means to have a giant looming data center in your community. and they've joined a lot of more left-wing people, I would say, in opposing build-out. Hochul is more in the middle. She says she's interested in AI but wants to do it responsibly. Trump, of course, vacillates between the two. You have both the extreme anti-regulation part of the Trump administration, exemplified by David Sachs, who's no longer officially part of the administration, who's very pro-build-out of these things.

17:18And then there's also people like Steve Bannon, a top Trump advisor, who are pushing against the build-out of data centers, not necessarily for environmental reasons, but because of things like job loss, which is a populist issue itself. So, I mean, let's just play this forward a bit. I mean, it's a one-year moratorium, and we should say, so executive order. So does this literally mean that you cannot build a data center for a year? Tactically on the ground, what does this mean and what doesn't it? So it's not legislation, so it can't be that far reaching. What it does is it uses executive power to say that it's not going to give out new permits for certain types of data centers, some more than 50 megawatts of energy, which would encompass the larger hyperscaler ones for one year.

18:06The legislation that was proposed would go even further. And again, she actually was on the Odd Lots Bloomberg podcast just today stressing this isn't the type of thing where she wants it to be a one-year pause and then another one-year pause and then another one-year pause, she really sees this as a stopgap to try and come up with a more sustainable solution. And it seems like her vantage point is not a full-on moratorium on data centers, but doing it in this more deliberate way. Right. Leo, I want to ask you of another story that actually came the same day as the New York executive order, and that was the Trump administration's executive order on IEI cybersecurity uh this was an initiative called gold eagle what is the gold eagle very very fun name for unfortunately i think a more boring project i mean you gotta you gotta get people to pay attention to cybersecurity somehow a very exciting topic i'm sorry everybody should care about this uh this you know what i think we've got aaron holmes coming on in a minute i think he would contend that cybersecurity is the most interesting part of the tech ecosystem but let's Let's talk about why Trump is so interested in it.

19:19Apologies to Aaron. I am not denigrating cybersecurity in any way. I think it could be the most exciting technological development. What's happening with Gold Eagle in particular is not the sexiest subject to come out of the executive order from early June on AI and AI cyber risks. The idea of it is it's a clearinghouse. So basically, if you have a ton of different AI models and labs who are each identifying different vulnerabilities on their own. That's being done in a sort of slipshod process. And of course, they might not really want to coordinate and share the vulnerabilities for competitive reasons.

19:53And the idea of this clearinghouse in the executive order is that it would be centrally housed, it would be run more by the government. And basically, labs could submit vulnerabilities and you could come up with a shared response to them or work together on it. Gold Eagle is the first iteration of that. The Trump administration announced that earlier this week on Tuesday. But really, I think they hosted a press conference, and it seemed more to tout their support of open source models. There's been a lot of fears or rumors swirling around that there could be some sort of crackdown on open source.

20:28And on this call that I was on, I would say the White House defending open source was much more a topic of conversation than this very exciting clearinghouse that's happening. So are companies likely to participate in this? I mean, what's their perspective on it? So the press briefing as well as the press release was very short on details despite reporter questions to that effect. People tried to ask who's involved. They wouldn't share. They did say that Carnegie Mellon is helping run this. But it's unclear at this point how far reaching this is or even what it's been able to catch or do so far.

21:05So we'll have to find out those details. um tell me leo what kind of additional guidance are you watching for here on this topic i mean where where is this story likely to go from here so if if viewers will remember uh this executive order that came out in june had a lot of different components the most significant part was this voluntary framework so this this system in which models can submit their models pre-release on a so-called voluntary basis, and the government would assess them, and then they could launch them. This would be for these new frontier models like Mythos or Fable or 5.6 out of OpenAI.

21:42This still has not been announced. It's supposed to happen in early August, and it will be a lot more significant than the clearinghouse that was just announced earlier this week. Great. Well, Leo, I want to thank you for coming on. That is Leo Schwartz, our AI and politics reporter here at The Information. Microsoft security division is undergoing a significant restructuring. Speaking of cybersecurity, the team cut several hundred roles and is also refocusing its product strategy. Our Microsoft reporter, Aaron Holmes, published exclusive reporting on that. I want to bring him on to talk more about what he knows.

22:17Aaron, welcome back to the show. It's great to have you here. Hello. All right. More cybersecurity. So what did you find out is going on in Microsoft's security unit? So Microsoft got a new security chief, Hayat Gallat, about five months ago. And since then, she's essentially cleaned house. She has replaced at least eight or nine senior executives in security. She also cut several hundred rolls last week. And she's basically overhauling the entire business to try to cash in on this frenzy around, you know, protecting against AI threats and also using AI to secure enterprises. So all in all, this is a pretty, you know, sharp pivot and direction for Microsoft, which it's easy to forget is the biggest security seller in the world.

23:07And essentially, you know, I'm told that the goal here is both to cut costs on things that are seen as less important now and also to hopefully jumpstart security revenue, which has not been growing as fast as many senior executives at Microsoft would have liked. Sorry, Microsoft is the biggest seller of security in the world. I did not appreciate that at all. It's bigger than all those other crowd strike, all these other surging companies. Yeah, their annual security revenue as of 2023 was$20 billion, which is more than CrowdStrike or Palo Alto Networks currently brings in. We don't know exactly how much Microsoft makes in revenue from security now, but it is north of$20 billion.

23:49So that makes it the number one. Okay, so if it's so big, then how is it tracking right now? I mean, tell us about the performance of the business and why ultimately there have been layoffs on the team. Yeah, so most of Microsoft's security revenue comes from bundling their security software with other products like Office and Windows that they sell to enterprises. And, you know, that has worked pretty well over the years. But I'm told that, you know, since 2023, revenue has grown, but it hasn't been quite the acceleration in revenue that, you know, Microsoft would like. And at the same time, they basically, they stopped reporting the security revenue.

24:29Is that the idea? Yeah, they haven't disclosed security revenue since 2023, which raises questions about how fast it's growing. Yeah, that's one way to think about it, at least. But I mean, how is it that the team has fallen so short here? I mean, what are people telling you in terms of like what the core problems is? Is this a problem with product portfolio? Is it a problem with execution? Is it competition? What's the root here? So there's definitely a few kind of overlapping problems that Microsoft security faced before this new chief took over. So the previous head, Charlie Bell, who Microsoft poached from Amazon several years ago to run its security business, you know, he ran into a few issues.

25:14You know, we reported shortly after he took the helm that he faced a lot of internal resistance in trying to pivot the security org. At the same time, you know, there's been several kind of high profile hacks that have been black eyes for Microsoft. You know, most notably, a couple of years ago, Chinese hackers used a flaw in Microsoft software to break into tons of customers, including the U.S. federal government. And, you know, the Commerce Secretary's emails were hacked at the time. Microsoft was essentially, you know, given a run dressing down by the federal government, which squarely blamed Microsoft for those hacks.

25:51So that has not, you know, helped its security reputation. And I think also, you know, Microsoft now wants to sort of counter this narrative that it's not well positioned for the AI era. And it is facing, you know, new threats in security from the likes of Anthropic and OpenAI. And, you know, from what I've heard, this pivot and this overhaul that I just reported on is sort of meant to address that and make the company better positioned for this AI security era. And, I mean, you wrote that the company's new product strategy is actually trying to mimic Anthropic in some ways. What did you mean by that?

26:27Yeah, so Anthropic released its Mythos model a few months ago. And just to recap, you know, that sort of started this frenzy across the business world of companies being worried that models like this could be used potentially if they fall into the hands of hackers, they could be used to find vulnerabilities faster than humans could patch them. So, you know, what's been happening since then is a lot of big companies are budgeting heavily to pay companies like Anthropic to use these models to scan their own systems for vulnerabilities, even though they are very expensive to run these advanced models.

27:02So what we're actually seeing now is that Microsoft is starting to put out similar products. They released a tool called M-Dash, which is essentially a multi-model version of Mythos. It uses Microsoft's own models as well as OpenAI and Anthropik models to scan company source code for vulnerabilities. So, you know, they were pretty quick to sort of get on that trend. And I think we have yet to see how much sales that'll drive for Microsoft. But that's the type of direction that they're now taking in response to, you know, how customers are thinking about security. We were talking with Carl Kirstead from UBS earlier this week about Microsoft's stock performance.

27:43and I was talking to him about the underlying issues that he thinks investors are concerned about. He pointed more so to the broad AI question and Microsoft's place in the AI race as a key consideration here. How much do you think investors are focused on Microsoft's security business as something that is dragging down the stock? I mean, how many ripple effects does this really have for the company? You know, I don't think that necessarily investors look for security revenue as like something to peg the stock to. But I do think that it's sort of emblematic of what's happening more broadly across Microsoft's software portfolio, which is, you know, it essentially has to keep convincing its existing enterprise customer base to spend more and more on Microsoft software sequentially.

28:31And in doing that, they have to prove, you know, what we're offering is as valuable or more valuable than the similar things you can do with these new models from open AI and Anthropic directly. And so I do think that, you know, if they can make a strong case that their security software is state of the art, that strengthens their ability to sell this bundle that includes Office and Windows and get customers to keep renewing, you know, at higher price points to include more of those tools, which is definitely very important to the story that they're trying to tell. and to this effort to turn around the bad fortunes that their stock has seen over the past year.

29:09Great. Well, Aaron, I want to thank you for coming on. That is Aaron Holmes, our Microsoft reporter, here at The Information. The Information published a new op-ed this week from Scott Stanford, co-founder and partner of Acme Capital. Scott makes the case for a nationwide dividend that would come from AI companies. I want to bring on Scott to walk us through his thinking. Scott, welcome to TITV. It's great to have you here. Great. Thanks. Pleasure to be here. So the headline of your piece is, Americans deserve a dividend from AI companies' riches. How do you propose that this works, Scott? Well, take a step back.