NVDA GTC, M&A Wiz / Goog $32 B Deal, April 2 Tariff Uncertainty; Huawei Belt & Road; ChatGPT

28:51Within that space, you know, you say, how could we differentiate ourselves? And, you know, AWS was first, I think AWS is considered a little more developer friendly. They kind of started in that place from the very beginning. They've been more flexible and open and what they offer. You know, Microsoft, I think mostly leverages their corporate customer base and kind of uses that to drive their business. So Google, how can they differentiate security, especially in a distributed world, seems like a very reasonable answer to that question. And so I guess, you know, to put it all in perspective, watching all that, it makes a lot of sense that they would come to this place.

29:34Now they're still saying it might take until next calendar year to get this done. And I think they put up a $3 .2 billion break -up fee that they lose if this doesn't get approved. So it's not like this is easy going. I remember when they tucked in ways in 90 days, like even this is difficult, but I can very easily see how they got to this place. And I, and you know, I'm not deep in cloud security, but I can also have a lot of conviction that that's a good idea for an area of differentiation. Well, I mean, I won't tell you this. I hear it from every single company I'm talking to. They're going to be watching this one very closely because people want to get back to the MNA game.

30:17And basically, there's been no MNA over the course of last several years. You know, we certainly, one area I want to cover here is, you know, just the pace at which we're building monster companies, right? Hat tip here to Shardal Shah at index to Doug Leoni for another legendary investment. You know, this is a business that was started in 2020. Okay. And they're selling it five years later for $32 billion. $32 billion. One of the largest outcomes in the history of Silicon Valley, you know, and those guys are both repeat offenders at home run power law outcomes. So hat tip to both of them. Green Oaks was in this and listen, Bill, I'll share a little vulnerability here about about altimeter and myself because I've talked a lot about the search, you know, what can we learn from search as we go to invest in these model companies?

31:13And I say, you know, think about all the people who got like us in altivist and all these logos. They were right about the internet. They were right about search and they didn't make any money. And all the money went to, you know, to to Google. Well, this is a case where altimeter was an investor with Sutter Hill in a company called Lacework, right? And it was competing in a very, very similar space. This was the thesis of Lacework. Lacework raised a lot of money, too much money I would argue, you know, at the time and they went hard at go to market, but they didn't have the product market fit and execution right relative to Wiz.

31:49And I remember when Wiz did that series be and I said, we've got a product issue. They're killing us, you know, in terms of how fast they're growing on product. And so this was a case where we lost money betting in the space. And some of these great great friends of ours made a lot of money betting on the competitors. You know, we've we've had plenty of wins here. And but this was one of those humbling times like what did we what did we get wrong? Right in terms of team and execution and shows how hard this business is bill. Right. It is a razor's edge between, you know, the promised land and and and not making anything because the winners tend to to walk away with, you know, a disproportionate amount of the price.

32:30Yeah. So any reflections bill just you've been around this for a while. 32 billion of enterprise value created in five years. Well, definitely. And I would I would qualify one statement. You you said they made a lot of money. Sounds like that won't happen until 2026. So but that's not counted chickens just yet. Yeah, look, one takeaway I have, which my partners don't they don't think it's fair, but I'm going to say it anyway. Repeat entrepreneurs in the enterprise space are a golden ticket. Like like it is so there are so many learned experiences about go mostly I think about go to market that are repeatable that you can line them up and do it again.

33:16And I always think of the workday example as like picture proof of this, right? It's the team that built people saw and and they just came and did it on the new platform and the technology stacks always changing and when it changes, it allows for a new intern to come along and be disruptive relative to that. And this has happened over and over and over again from mainframe to many, many to client server client server to the internet internet to the cloud. It just happens over and over again. And so if you have a repeat entrepreneur and you can, you know, establish a relationship where they're going to want to come back to you again, shout out to my partner Peter Fenton working with Brett Taylor again, which is a perfect example of this.

34:00Um, great. Sluteman. Yeah, the odds of success are way, way, way higher. And as you said, had tipped to Leoni, he's been around a long time having a win this big at the, you know, at the I won't put him to pastor, but at the end of your career, it's just super impressive. But I think it's, I think, you know, the key takeaway for me is that that the repeat entrepreneur with a new wave in, uh, in enterprise is probably the best way to increase your probability of says. And even then, you know, you may run into what you did. So it, you know, it ain't easy, but it's a great outcome here, great outcome for all these people.

34:40And let's, let's hope it get it, it gets approved. Yeah. So, so, so let's talk about that a little bit the approval process, because I've got a lot of questions because it's not only important to this one, Bill, it's important to the signal, right? Are we going to unleash animal spirits around M &A, or are we going to caution before they get out of the gate? So let's talk about the FTC for a second. Right, the chair of the FTC, Matthew Ferguson, I think there've been a lot of misinformation about, uh, about where kind of he stands on M &A. And so the headlines that I saw over the course of the last couple of weeks, um, in particular, there was a quote that I saw Bloomberg and a bunch of other people reporting where he said we're not going to be deferential to the C suite.

35:28We are going to be the cop on the beat for big tech. And people combine those two things and said, Oh my God, this is a continuation of lean a con, right? Here we thought we're moving into a totally different business environment, but that doesn't sound good at all. So I went back and I read the transcript on, you know, a couple of his long interviews. And of course they pulled those two things out of the interview, but I don't think it's reflective of where he stands. So just a couple data points that I would add to the conversation. One is when asked specifically as to whether or not he differed with lean a con, he said, uh, really quickly, I, I've written hundreds of pages of dissents with lean a con.

36:15And he said the number one area I differ with with the prior administration is that they lingered. They created a regulatory fear state that nobody wanted to do anything because literally they would bring an action and then nothing would happen. It would just die over time on the vine. And it was like purgatory for all these companies. And so no deals got done. He said, I'm going to be a cop on the beat, but he said number one, I'm going to go to court or get out of the way and let business thrive. That is a totally different mindset. Go to court because you're clearly in violation of the law or I'm going to get out of the way and let the business thrive.

36:59So that sounds to me like we're going to get pretty quick action here. Like we're going to know one way or the other what they want to do. And then the second point he made, he said, I'm not a regulator. Think about that. I'm not a regulator. I'm a cop on the beat. I'm looking for fraud, monopoly or collusion. And I think those two things may well define a very clear and differentiated approach to the FTC. And I think it's going to be reflected here. I think we have an unbelievable opportunity to unleash an incredible amount of economic growth and capital finding its best home and best place to grow businesses with a lot of pen up M &A.

37:43And I will tell you this, when I look in the public market, I see great businesses that are growing 10, 20, 30 percent that are trading at six or seven times revenue. This deal just got done it over 30 times revenue. There's ample room for buyers and sellers to meet in the middle if the United States is truly open for business. Well, and I would argue that at least in terms of Google was the it passes that test. It's none of those three things. So hopefully it will move through quickly. And it sounds like we've got on the flip side on the IPO front, there's talk of core weave in the shoot. Rumors about Clarenna, I think Sirribris is in the shoot.

38:29It'd be great if we see some is remember I'm on the buy side. So when Goldman or Morgan Stanley want to sell one of these things, our phone rings and they want to check market conditions, they want to check what we're thinking, you know, and then eventually, you know, the road show occurs, etc. And I would say all of that, our phones are ringing. There's a lot of activity. The ones you mentioned and many others. So again, think about this. The NASDAQs down 10 % bill in a month and notwithstanding that we get a huge emanate deal announced and we have the pipeline for on the IPO front. So this to me is a strong signal that people believe this is much more of a pro growth, you know, administration that's going to be supportive of deal making.

39:24And if that stuff happens, then I will say whatever you think about tariffs, whatever uncertainty is sitting in your mind about tariffs, there's a lot of upside from this activity percolating through. One thing I would bring up and that is pertinent April second, which is you and I just said that that Matthew Ferguson, the FTC may in fact clear the way here. But the question is, what are the Europeans going to say? Remember, they've been the ones who've been the pain and the ass on all this stuff, right? And these companies now have significant presence in Europe. And so if the Europeans continue to bite our ankles, I think this is one of the things the administration is going to point out on April second and just say, like, we can't have this, right?

40:06You can't effectively thwart business activity between two American companies because you're going to sit there and impose, you know, these harsh standards, right, from an M &A perspective. And that I think is going to get kicked into the domain of the trade negotiations and tariff negotiations. This would fall into the bucket bill of a non tariff trade barrier that we're going to punish American companies, right? With excessive regulation that that in fact hurts our ability to do business. So this is going to be interesting to see how that one unfolds. All right. Now let's switch gears. I understand you were down in San Jose yesterday, were you at the SAP Center when Jason, I hear there were 18 ,000 people there.

40:53And I saw a photo his his leather jacket, I think, has more zippers and buttons than the previous version. He's clearly the new Steve Jobs. I will tell you, in fact, I was just down there this morning. I saw our buddy will Dan off from Fidelity and Gavin Baker. You know, it's, it's, you know, everybody is there. I will tell you this, whatever you think about the upside or downside of the stock and we're going to get into that, we're going to get into their announcements. What I said to Gavin, as I was on my way out, is, you know, hands in the air in prayer position to give thanks that this is an American company.

41:30There's an extraordinary company that that's at the heart of, you know, our competitive advantage when it comes to national security, economic security, all things AI, Jensen is in full founder mode. They're executing brilliantly and they're an American company. And so, you know, and by the way, we take that for granted. It's not yet, you know, like I don't think we should take it for granted because we just got done talking about how competitive the Chinese are with BYD. It could very well turn the tables and we could be dependent upon Huawei for next generation AI chips instead of the United States.

42:07And so protecting that level of innovation here in Silicon Valley is important. But let's talk about what we learned. Just very quickly, while you're on that topic, you know, Jensen was born in Taiwan. And as many have highlighted, many of the founders and leaders of all these incredible American tech companies are either either first generation immigrants are second. And so I do hope I'd harken back to when President Trump told the all in crowd that he wanted to staple a visa to every diploma. I hope those activities are underway. I have not seen any visible activities on that front, but I hope those activities are underway.

42:49No doubt about it. I will tell you when you think about in particular the chip companies bill, lip -butan, the new CEO of Intel, I think born in Malaysia, you know, I think every one of them has a CEO who is born in Southeast Asia, China or Taiwan. And so I couldn't agree with you more immigration, smart immigration, smart immigration, right, is a huge national competitive advantage. And we ought to, you know, we ought to continue to focus on that. So what came out of GTC that's the same or different than what we knew. Let's just start at the top by saying the stocks basically flat, right, the stock peak peak last year, around 150 bucks a share.

43:30It got as low as 105 the other day. It's back to 115 on consensus numbers and video trading at about 20 times next year's consensus earnings estimates. That's really low relative to the S &P. It's trading about 24 times this year's consensus estimates. That's again, lower than the S &P average. So this is not a demanding multiple. Why is it trading there, Bill? Well, people are worried about the demand. They're like, is AI overhyped? Is the demand going to continue to be there? What about deep seek? What about the competition for M6 et cetera? So there is a big wall of worry about this and, and, you know, and even among the Mag 7, Apple et cetera, all of those stocks, Costco's trading at 50 times earnings.

44:15So we have a lot of non -tech grump companies without a lot of growth that are trading at much high earnings multiples at this point. Then in video, okay, so a few of the key things that came out of this and then I'll tee up a few of the debates. Number one, we talked a lot a month ago about deep seek and he was very clear on stage. He said there was a profound and deep misunderstanding of deep seek R1. Because remember when it came out, there was a lot of fear that, oh, you can get intelligence on the cheap. You don't need as much compute. You don't need to train these big models. He was vehement that that was untrue.

44:54And in fact, he doubled down on that bill and he said the amount of compute we now know that we need, right, today is a hundred times greater than what we believe to be true a year ago. Okay. So that was one of the things that was buzzing that just what is happening to the aggregate demand, the aggregate tam, if you will, for the market. And so I want to talk a drill down on that for a second because he showed an important slide that we'll have here which was industry estimates on what the tam is for AI data centers. And he said, remember we did this podcast on February 22nd bill last year where we laid out the data center rollout.

45:40And he said that the amount of compute he was in Saudi Arabia at the time, the world was thinking it was going to be about $250 billion a year. So a trillion over four years. And he said, no, that's wrong by half. We're going to do $2 trillion of tam over the next four years. So something closer to $500 billion a year of new AI workloads, right? So data centers supporting new AI workloads and then replacement data centers. So these are, you know, X86 architected data centers that are going to move to accelerated compute. Well, he doubled down on again at GTC. And now he's saying that it's going to be a trillion dollars per year by 2028.

46:24So we'll include this side, but we re reforecast it. The slide we did last year. This is a both new AI workloads. And all of the existing data centers that he believes will be rebuilt and when they get rebuilt, they're going to be more accelerated. So that's going to a trillion a year. So the market is growing much faster than people think. Now, why is this happening? And I think he used a really great example. He said, as of today, you think about somebody who's writing software. And they're basically hand coding the software and they can do that without accelerated compute, right? And so you're just working in a normal on a normal database in a normal data center.

47:08And he said, but by the end of the year, right? Most code is going to either be refactored or written from the start using machine learning. It's going to be coding agents that are helping to do that. And those all require accelerated compute. And so if we listen to Mark Zuckerberg or Elon Musk or Sam Altman, they're all saying the same thing that coding agents are going to come. They're going to just place a huge amount of the handwritten code that was previously getting done on CPUs. And so that's the type of thing. He showed some queries and the reasoning models and why that's multiplicative and leading to much greater demand.

47:47But those were, I would say, at the high level, he really made the case that demand is exploding across all of these different workloads, literally from he said this morning that robotics and autonomous cars was already greater than $5 billion revenue run rate for Nvidia and growing very fast, he said. So they have from synthetic biology to robotics, to physics, to all the stuff we see in consumer, to enterprise IT. They have this incredible stack across CUDA, which is multiple layers of software. And then finally, he said, remember, we're not in business building chips anymore. That ship sailed a long time ago.

48:31We're in the business of building supercomputers. Blackwell is 40 times more capable than Hopper was. And the demand is off the charts. And so, you know, notwithstanding that fact, I think the markets checking them at the gate and they're saying, we'll believe it when we see it. And I think a framework for thinking about Nvidia at this point is this is probably a company at 2 .7 trillion that's going to grow at whatever the rate of its earnings growth is bill. Right. You've seen all the multiple compression you're likely to see. I think in that regard, you know, they can grind a lot. I have my reflections and feel free to push back.

49:10One, as you already said, the market didn't move, which implies, if you believe in the efficient market hypothesis, that there was the full absorption of data that neither positive or negative. Right. Of what was said relative to what expectations were. Two, I do think there continues to be, and maybe we should just find a way to put a pinnacle, but this argument about scaling or not scaling that I think no one's in disagreement on the reality, but a bunch of people like to argue it in funny ways. So I do think most people believe that the pre -training scaling if it hasn't topped out, it's at least slowed materially.

49:55And even Nvidia is talking about inference being 991. And so the scaling is more on inference. Everyone seems to agree with that. When they say everyone got it wrong, I think that falls into that category of that this kind of nuance discussion where the bulls don't want to admit that anything isn't scaling. And the analysts are just saying, well, it shifted a little bit and that might matter. Anyway, we can forget about that. On the token count thing, no one in the previous world that's called Intel's heyday talked about the CPU units of performance, like how many times the CPU transacted. If they had, you would have had crazy exponential numbers also.

50:39They didn't, it wasn't discussed that way. And so no one's counted the CPU clocks that have happened over time. So I do find it like it's kind of being ultra promotional way to talk about things, to talk about the token count, just because we haven't done that in the past. On Rubin, you know, and I'm biased because Benchmark's an investor in cerebris, but and I would expect this in a more inference focused world. You know, Rubin, which is their next gen, they started talking about there's way more on the die and the die is getting a lot bigger, which is the direction of cerebris. So that that one kind of confirms to me that we're moving more towards inference.

51:19And then the other thing that I thought was interesting and I'd really love your view on, he got up on stage and talked about being the chief revenue destroyer. And it was very kind of provocative. And I don't, I don't fully understand. I think he may have created a bit of anxiety in his customer base because he was implying that this next generation is so good that it makes the previous generation quite unvaluable. The problem is the companies he's selling to, you know, within the past three years moved from a four year depreciation to a six year depreciation on these servers. And if he's going to kill him in two years, there's a, there's a, there's a kind of economic problem for these customers.

52:04Yeah. I don't, you know, I was there for that. You know, and it was I think an off the cuff remark that because the black well generation, the GBs that they're currently selling, not the future generation, the current GBs are 40 times more capable, right? When you look at TCO, then the hopper generation. And so he said, you know, we're probably not going to sell a lot of new hoppers for the cutting edge stuff that people want to do. So in fact, he showed a slide of the CSP demand. So the top four cloud providers in the United States. And what the year over year order was, you know, hoppers 1 .3 million hopper GPUs to the top four CPUs or CSPs and comparing like time frame, he said, we've done, you know, 3 .6 million black well GPUs so far this year.

52:56And I think his point really there was the pace of scale up and scale out bill and how much utility that provides to the end customer. And the point is like, because I, you know, we're investors in core, we, we talked to core, we've, or we talked to the CSPs. The reason they're elongating the depreciation cycle is because the way in which they build these is all the software upgrades flow naturally through to hopper. So hoppers continue to get better over time, even though, you know, you've got the same chip deployed. And remember for inference, hoppers great. You just wouldn't choose to buy more hoppers today because the new chip is a better TCO than the old chip was.

53:38It's not that the old chip wasn't the right thing to buy at that moment in time. And I would say the same thing's going to be true about black well ultra. What I think is when you're on the pace that he is, he's got 35 ,000 employees, they're executing blisteringly fast. They've got the best supply chain in the world. And I think it puts tremendous pressure. Like I think the gap between him and AMD, you know, Nvidia and Intel, Nvidia and all these custom Asics. In fact, he made a comment that I think royals and people, he said about the custom Asics, he said, don't you know that the vast majority of custom Asics that get designed and get taped out actually never scale into any commercial production.

54:21And we hear that Trainiam is now cutting their cost, you know, on AWS is cutting the cost on Trainiam chips to try to get a lot more customers. Usually don't cut costs on things, right? When, when things are going great and you got a really competitive product, I think Trainiam is not even competitive with hopper, let alone with where black well is today. And I think, you know, jassy would be the first to admit they're buying as much as Nvidia as they can get their hands on. So I don't, I'm not worried about what that comment means for their release cycle. I think this is one of the few companies in the world that paints a, you know, can you imagine if Apple took the stage today and told us what was going to be in the next four iPhones, right?

55:03They'd never do it. They don't know, right? This guy is planning out where, you know, the next four generations of the product is going to be. And I think he's got an incredible path. He introduced this new improvement around photonics, which I, which I think is pretty incredible. If they execute against the plan, it's going to be a lot bigger business. And I don't see many people able to close, you know, to close the gap on them. And I think they'll stay ahead of the rest of the world. So that was, that was my, you know, my take on things. I, you know, question back to you. Yeah, go ahead, go ahead.

55:36You know, he said this morning that he doesn't think he said on CNBC, I don't think tariffs will have a near -term impact on our business. And there's a lot of talk, Bill. We've talked here about the Biden AI diffusion rule, right? That makes it much harder to sell these chips to like 50 countries, this convoluted system. We've talked about, you know, the fact that Chinese export controls are likely going up. We're going to get this, this announcement on sectoral tariffs on April 2nd. Do you have any points of view or any thoughts about what we should be doing with regard to, you know, those things and Nvidia?

56:13If you were, if you were the president, would you be imposing higher export controls, you know, would you be getting rid of the diffusion rule or keeping the diffusion rule? How would you think about it? Let me make one brief comment. And then I'll answer that question. I do want to, to agree with you on one point. I'm looking at a, oh my God, you're going to agree with me on a point. I don't think our job here is to just agree with each other. It'd be a very uninteresting podcast. So I try and maybe create an alternative viewpoint. But I look at the mag seven. And I'm looking at them, ranked by market cap.

56:46And if you asked the question, I think you would agree with this. Yes, the question, which of these companies is executing the fastest right now? And which of these companies is most exposed to global trends in video would win both of those, correct? Correct. Yeah, I mean, listen, it's a smaller position than it was the last couple years because all of our position sizes are smaller as I took down risk at the start of this year. But again, I don't expect this is going to be one of those flash in the panel. It's up three acts because multiples expand a lot. I'm, I'm assuming multiple stay largely the same.

57:23And I just get the benefit from the great execution and the top line and the earnings growth of the business. And I will say, you know, this slide that I showed you here, Bill, Wall Street consensus expectations are basically the Nvidia tops out at $250 billion of revenue. Like somehow there's like this, this ceiling there. And if you believe that to be true, then their share of the market falls off a cliff because the market's growing really fast. I don't think they're going to lose share of market. If anything, I think they probably gain share of market in a fast growing market. But I do think back to, you know, these the diffusion rules.

58:00There are a lot of people who were around the rim at GTC and they are very worried about tariffs. They're very worried about the Biden era diffusion AI diffusion rule. They're really worried about Chinese export controls. And, you know, so I just thought I'd get your look. I continue. I said it a few weeks ago. I think the number one risk on the stock is government action from DC. And there are a lot of people have been asking questions about the percentage of their revenue that goes to Singapore and then they reply and say that's billing, not shipping. And then people say, well, where is it being shipped to?

58:37And there's all these questions about deep sea, can where they trained. And I certainly think that it's impossible to stop a startup from any country from traveling to Europe or Malaysia and running a model. Like, I don't know how you're going to prevent that and move in the bits back. And so I don't think there's a solvable problem. But I also think there's a lot of angst in DC about China. I think a lot of it's misplaced and overly angry. I think your friends at OpenAI added to that last week by putting out an anti deep seek paper that I thought was quite sad. Like if a company like GM or Ford were to put out a paper like that about BYD, we would look at them and go, oh, you just want the government's help.

59:22You must be uncompetitive. And so I don't know why we would think about that differently from one of our leading AI models. But there, you know, I do think that these people have a lot of power. I think it's a bipartisan issue. And I think it's the number one risk on on the stock flat out. You know, my view on this is when you come, if we impose high structural tariffs, and if we allow this crazy Biden era diffusion rule to stay in place, which makes it hard for us to export our chips, I literally think it's unilaterally disarming America in the race to AI. Yeah. I think it's a very bad decision.

1:00:09And positive Huawei, right? Like, correct. It's going to lead to it's going to lead to a Huawei belt and road. Listen, I already think this is backfire against us. Right. Deep seek is running inference on Huawei 910s, you know, because they may not be as efficient as a video, but they just throw a hell of a lot more power, which they got a lot of in China at these chips. And they can do it. And part of the reason China has been forced to build a vertically integrated domestic supply chain literally from design to fabrication around chips is because the United States made it super hard for them to get their hands on and video chips.

1:00:46And so we really have to ask the question, did we achieve our mission? And there are a bunch of people arguing, well, we need to even be through a higher export controls on China. You know, listen, I don't, you know, I don't want to do anything to make it easy on China. I'm fine trying to slow China down a little bit. I just think it's a task in futility. Like, there's no denying they already have frontier models. The release of them every day, their frontier models are smaller. Like, it's the deep seek moment. Now, Bob has one, by dance has one, right? So that that horse is out of the barn. As we've already seen with BID as we are, all with CATL, right?

1:01:26China is going to have frontier capabilities. The bigger issue is this bill, this on the diffusion rule, right? Which the Trump administration ought to throw away and start over. This diffusion rule will make it hard for us to get chips to Saudi Arabia, to the UAE, to India, to, you know, our friends in Southeast Asia. And it makes it hard through this regulatory capture that some of the US CSPs put in place, you know, all these hoops they have to jump through. And I just see Huawei walking right through that door, you know, and beginning to, you know, run the table globally. Like, frankly, they've done, remember, they said this would never occur in telecom equipment bill because of Nortel and all the highly capable telecom equipment of the United States.

1:02:13Huawei ran the table across the globe with Huawei gear because the US did this. Like, we ought not make the same mistake twice. I think it can go further. Like, every time I've listened to ASMR, who's, I think, a Dutch company, is that correct? And someone that has dominant market share like 80 or 90%. When you hear them talk about the restrictions that the US government has put on a Dutch company to sell into China, I always hear this kind of reluctance and a bit of being angered that this, the US thinks they have that authority. And then you take what I said earlier that came from many, at the economists, you know, I could easily see us provoking Europe to the point where the ASMR team just says, well, screw it.

1:03:03We don't care what you say. We're selling to China. And I think that could easily happen. So I don't know, I think we got to calm all this down a little bit. Well, one of the things that, you know, maybe we'll wrap this little section here. When it comes to, you know, what we ought to, what we ought to be allowing China to participate in here is like, I totally agree back to the tariff issue that it should be a strategic objective of the United States to re -enshore fabs, right? And you and I have a bit of a disagreement on this, you know, it gets back to this Morse Chang. Morse Chang says, it's too hard to do US workers don't want to do the work.

1:03:45They don't want to live in dorms. They don't want to do the six days a week. I don't think we need to do that. I think with the level of automation that we now have in fabs, and I think given the strategic national importance of having these fabs, I'm happy that this administration has put the pressure on TSMC to build their next generation R &D lab or fab in Arizona. I think it's achieving the desired results. What I would hope that they would do is maybe announce tariffs, but make them conditional bill. They say, if you don't do these things, TSMC and video, et cetera, and prove to us that your building manufacturing capabilities back in the US, then we reserve the right to hit you with tariffs.

1:04:23But we're not going to impose them today. We'll have a two -year delay or whatever while we want to provide the incentive, but we don't want to slow down US companies in the race to AI. I think that would be a terrible mistake. And so hopefully that's the middle ground, the administration guided by folks like Scott Besson, David Sachson, others find their way too. I think that would be a reasonable middle ground. And I think it is a great objective to get this stuff back in the United States. But maybe to wrap here, we can hit on a topic that you and I talked about the other week. And this gets back to this couple of questions.

1:04:59One, just thinking about, we had the launch of GROC, Big Bang, the launch of deep sea. Maybe we just check in a little bit, Gemini's had some updates on where we are in the state of consumer AI demand. And importantly, the topic that I've been reading about around contribution margins, like even if revenue is growing, are they selling dollars for 50 cents and are they losing money on each incremental unit of business? I'll kick it off by saying this, when I look in the App Store, deep sea jumped up to number two and was there for a couple of days. Chat GPD was at number one at the time in the App Store downloads.

1:05:43Now, deep sea isn't in the top 100 last time I checked. When GROC jumped up to number two for a few days, last time I checked, it's now around 65. Gemini never really jumped up. It's hanging out around 55. The point being, and Chat GPD is still number one in the App Store. The point being, I've said about consumer markets now for a while, that all my pattern recognition from Google and Matter is that these tend to be winner -take -most markets, that it's almost impossible to dislodge the inertia. You can't get there by being slightly better. At this point, you're going to have to be 10x better than Chat GPD to slow down that inertia.

1:06:27That was the learning from prior, from these prior periods. And now, we've seen the reports over 400 million weekly average users. And I would tell you, I believe the OpenAI is massively supply constrained. I think they're building these two huge campuses, data centers, one in Abelene, Texas, the other one in Denton, Texas, one with Oracle, one with Core Weave. And I think they have to build all that just to actually launch the products that they currently have. That's not even supporting future growth. It's just like supporting the demand they currently have. I think they have four or five products literally sitting on the shelf because they don't have the compute to deliver it to customers.

1:07:09Any thoughts there, Bill? Have we seen, do we know the winner already in consumer AI? Well, I think your argument is very, very solid. I don't think there's any data points that would suggest there's an immediate threat to that. I would think that the powers that be at Google and Groc and Meta and Amazon must be up at night all night trying to solve this problem. And so I guess, Netscape had Microsoft. OpenAI has four or five of the most powerful companies in the world aimed at them. And so it's a, you're right, they have a lead. It could be insurmountable. They have all the King's horses and all the King's men behind them.

1:08:04And so it'll be interesting to watch. We've talked about the things. I think if someone crushed voice, you might give them an advantage to move fast. That might be particularly compute intensive, which we'll get into our next topic. But for now, I don't have any ability to take the other side of yours. Well, you know, who the couple I didn't mention there, I didn't mention Meta, right? And we know those guys are in full beast mode, but they unusually quiet. Like, I'm actually shocked how slow we've seen any change there in terms of a standalone consumer app. I know it's coming, but I've heard it's been coming for a long time.

1:08:46And or even the integration within their existing apps, I'm, you know, I'd love to talk with was Zuckerberg about that. And the other one is Apple, right? And we've talked about that again, just not seeing any real change there. And in the meantime, I'm seeing, you know, I think the stuff that we're going to see over the course of the next eight to 10 weeks out of open AI as they get these releases teed up, I think it's going to be pretty profound and, you know, continue to add, you know, to add pressure to that lead. But, but let's let's ask the question. Does it wait? One thing I have been thinking about.

1:09:22So I do think that deep seek by choosing to be more open than llama and the key area they're more open than llama, they don't have these caps. And so, llama and mistraw on a few others, they claim to be open. But if you get too big, you have to go pay the piper. And so, you know, and there's always been a continuum of open source. Like it's always every company chooses their spot on that continuum and tries to see what they can get away with. Because deep seek is so prevalent in the enterprise right now, my point of view. And it's been forked like 1500 times on hugging face. Like, and for all the reasons open source works, that's why that's happening.

1:10:05I do think it's an interesting moment in time for meta in particular to consider maybe going left and getting even more open with llama, which I think would be really powerful for them. And this probably won't happen. But even open AI could consider doing that. If you're right, the Derek consumer product company, there's not much risk in taking that leap. And it would undermine anyone else that's trying to compete based on the quality of their proprietary model. So anyway, I wanted to bring that up. Where were you? Well, listen, you've been appropriately tough in asking questions about the gross and the contribution margins for these businesses, right?

1:10:54And so I saw a report recently on Anthropic suggesting that they had lower contribution margin. So, so let's just first just unpack what that is. So, you know, contribution margin is basically, you know, what do I have to pay as a variable cost to, you know, provide the service that I'm providing? So in the case of Anthropic, most of of their enterprise API, I believe is sold through AWS. And they have to pay a huge customer acquisition fee in the world of Google we would call it, tack, probably to Amazon. And then, of course, you have all the costs that you have to pay to actually serve the model, right?

1:11:37And there's a variable cost associated with serving the model. And when you take those two things into account, because of the large percentage of their revenue that is indirect, that they have very little or no contribution margin, right? So no net revenue really flowing through the PNL. Now, almost all of their businesses direct. So they don't have to pay traffic acquisition costs or revshare to anybody, right? Which obviously you have to, that has to be the business model, because the cost to serving these things are expensive. Now, of course, both of those things ignore bill your overhead and your training costs, right?

1:12:20And so that's truly fully loaded. But I'm just looking even at the contribution margin level, because if you're not, if you're not meaningfully positive at contribution margin, then you have no chance of covering the cost on a forally burdened PNL. So, so here's, I've been thinking about this a lot. I will introduce, let's just call this loosely and a first drafts, an idea that I'll call the GERLY negative gross margin AI theory. Okay. And because my brain, I can wander in some pretty interesting places all by myself. But a lot of people have been comparing things to the internet revolution. And, you know, I know why they do that.

1:13:07There's so much excitement, you know, for all similar reasons. And so they want metaphors and comparisons. In addition to that, I would say, and you just explained it, the unit economics are messy here, because you have, you know, you have variable costs that is, especially if you're buying, you know, your compute from someone else, you may have CAPX. And then there's a question of, are you depreciating that over how long and do you charge the unit economics against that? Some of these players have credits that they were given by, by the large platforms as investors. Do you really take that into account or not?

1:13:49And in addition, you have tons of VC money. And so if you're going to take, if you're going to raise billions of dollars, you're going to have high burn rates like these things just are part of the game. In addition, I think there is a mindset that most VC markets are winner take all. And so you can't lose market share. Right? It's just, you can't lose market share. So if you can't lose market share, how do you price? You know, do you price to market? Or do you, you know, do you price to cost and value like you would in an economic textbook? Or do you just price to market? Meaning, I can't lose because of price.

1:14:28And so, you know, it becomes a buyer's market. I'll create a theoretical example because I think there's also an issue of stacking. Let's say you have, and let's not pick on any company. Let's say you have a large model provider that's buying compute from a hyperscaler. And then let's say there's a, you know, startup doing voice for the enterprise on top of that model provider and all these people are buying compute from the one below it. If those second layer and the third layer are both negative gross margin, the consumers buying compute from the hyperscaler at a price that's lower than they would if they bought it directly.

1:15:13Because it's being subsidized by two players in the middle. And on top of that, if this is true, your triple counting revenues. Right? Like there's a transaction that a consumer is making against this model. But because it's negative gross margin, you're literally adding it up three times. Same thing would be true if you were, you know, any really what people would consider a shitty business if you were a just, you know, a global distributor of retail products and you're taking 5 % although here it's negative. And so, this theoretically could be happening out there, you know, and you pile in the tons of EC money, the fact that everyone believes these markets or winner take all the fact that the founders are likely unsophisticated financially.

1:15:57Not, not, not, I'm not like taking shots at them. They probably haven't had a lot of finance classes. And you don't have great visibility on unit economics. It's set up. It's set up to be messy. And you could, you know, you could have substantial resets as people get in touch with the unit economics. When and if they're forced to, and that's a very unpredictable time window of when that when when that would happen. Yeah. This is a this is a theory. I'm not suggesting it's 100 print true. It's possible. This is going on. Well, listen, I think the one place I would absolutely agree with you, Bill, is that the unit economics and the ability to compare across these businesses is messy.

1:16:38That's number one. Number two, what I agree with you on is in no way are these business models proven yet to be anywhere close to as good as Google. Remember, Google raised less than 50 50 a million dollars before they went public, right? And so Google was able to do what they did with very little upfront capital. And then their unit economics from from the very start were profitable. They and this is a really important point in the internet. The businesses that work, the marginal unit cost to serve a customer were near zero, near zero for Google, which is part of why it was so economically efficient.

1:17:19I think that was true for meta also. Right. And I will tell you I will tell you so. So here's my my take away for whatever AI companies emerge with a consumer business model or an enterprise business model where they fully control the top of the funnel, where those people are coming to them at almost no incremental cost because their friend told them about it or whatever. There's no marketing costs. And so your only cost is actually serving the inference to support that customer. I think those will have a unit economics that are exceptional that in at maturity look a lot like the internet companies.

1:17:56But I will say that right now you hear these top line revenue numbers. And the first thing you ought to be asking yourself is who do they have to pay to get that revenue? Right. And what are they loading in terms of their variable cost against that revenue? Time will tell, but my sense is there aren't there aren't very many sustainable business models here today. You know, and I think what I mean by that is that you're going to have to support this level of CapEx that these companies are undertaking. You're going to have to have tens of billions of dollars in high margin revenue in order to support the reinvestment that you're making on the back end of these businesses.

1:18:36Obviously, I think the best position one at this moment subject to change in terms of the independent players is open AI. But remember, for meta, they can invest they have a printing press kicking out billion dollar bills in the back room. So they can invest against this for a long time. Amazon can Google cam to your earlier point bill, right? You're not going to have a clearing event here where it's winner take most because these guys will continue to throw money out at for a long time. And I would say among the other independent players, Elon and Groc have a great product have executed incredibly well.

1:19:11And he has a unique very, very unique ability to raise global capital for a long time along with x .ai. So you absolutely can't count him out of the race. But maybe we just wrap it there. I mean, there's, you know, it's a period of consolidation for sure. You got a lot of uncertainty in the world. You got Nvidia trading at 20 times earnings. You got a lot of questions about these companies. But we're going to turn over a lot of cards, I think, in terms of these products over the course of the next 8, 10, 12 weeks. We're going to turn over a lot of cards on tariffs on, you know, kind of status of the reconciliation bill and taxes, et cetera.

1:19:50And you know, that's what makes it interesting. I mean, we could find ourselves, you know, off to the races come this summer. Or I think we could find ourselves with, you know, in the middle of a trade war and, and, and, you know, and everything slowing down. So there's a lot of stake here over the next 90 days. Let me, let me, let me make three final thoughts on my, on my, my silly, silly theory one. I think one of the reasons this can happen also is this steep price decline we've seen on models and older models. And so I think it's very easy for a founder or a team to say, well, it's okay that I have negative gross margin or that I'm pricing here because look what happens.

1:20:33Like in six months, the compute for this will be a lot cheaper. And, and I think that's a very rational decision too. I think that also on lead to what everyone seems to be seeing what is when people go to production, they look for cheaper solutions. And this is why I think open source drives why deep seek is working on the enterprise. Like I think everyone when they go to production, they optimize. And then, and then third, just in terms of this theory, if we're going to compare to the internet area, if there are high variable costs, I think you have to think about e -commerce more than you think about, like Google, right?

1:21:11And there were a lot of businesses this gets back to selling dollars for 85 cents. They were growing by achieving that. And so I'm very curious how it all play out. All of that said, as you know, I'm a huge believer in AI. It's causing all kinds of disruption. It's going to create all new kings and queens in the venture market. I just like getting under the hood and seeing what's happening on the margin. Well said, well said, we're going to see a lot more in the weeks ahead. Good to see you. Good luck to Florida. All right. Let's appreciate that. Go G.