The Model That Beat Claude & ChatGPT, The Nvidia GPU Crunch Continues, Beehiiv’s Social Tools

0:13Welcome everyone to the Informations TI TV. My name is Akash Pasricha. It is Friday, July 17th. It's been a great week. We had our one year anniversary party. We had our UBS shoot. We are back in our New York studio next week. But today on the show, Moonshot AI and Kimmy K3 is getting a lot of attention as it challenges closed source models. We'll bring on the CEO of benchmarking company Arena for the latest on what he's seeing. We'll then unpack Netflix's latest quarter and separately some other big headlines from the week on this week's edition of The Editor's Cut. We'll dive into our latest exclusive reporting on the traction that Google is getting with its TPUs.

0:56And we'll close out the show with a conversation with the CEO of Beehive as they make a big bet on social. It's going to be a great show, so let's get right on into it. KimmyK3 is the big news of the last 24 hours. Moonshot AI released the model. People are suggesting it could be a big threat to existing models' dominance and an open source threat at that. I want to bring on Anastasios Angelopoulos, the CEO and co-founder of Arena, for all of his thoughts. and astatias welcome back to the show it's great to have you here thanks for having me new day new model new day new model you've been busy man i mean gosh this is like the fastest uh pace of model releases i think we've seen in a while it's incredible six models in the last uh seven days it's been grok 4.5 kimmy k3 you know previously we had glm 5.2 it's the the the new series of models the gpt 5.6 is the soul you know the luna the terra uh it's been an extraordinary piece of model development and it's left us with a chinese open source model on the top of the coterie and it's i mean it did it come out of nowhere here were we expecting we'll talk about the reviews in a second were we expecting this to be this good initially upon reviews well you know there were there have been murmurs of the great performance of kimmy k3 for quite a while i'm not but I'm not sure that people expected it necessarily from Moonshot.

2:21I will tell you what has been the trend, which is that over the past year, you've seen open source and closed source models had a big gap and then it started closing and closing and closing. And then they were kind of tracking with open source models just a few months or weeks behind closed source models. And people were saying they're never going to crack it. And it's because of distillation. They're just distilling, distilling, distilling. And so fundamentally, when you distill a model, it degrades. and so you're never going to get better performance. But this is the first time we're really seeing the narrative that breaks that mental model, that says that the Chinese labs might actually just be really good at developing models and not just distilling American intelligence.

3:03Right. So you run one of the most popular platform of benchmarks that is widely cited in AI. What is your data telling you about how good Kimi K3 is? Well, the way our platform works is that we have a user base of tens of millions of people that are coming on Arena to use AI after their real workflows. Their agentic workflows, they could do single threaded conversations that are, you know, tens or hundreds of turns long. They're getting to do, you know, workflow automation. They're doing coding. They're doing math. They're doing all these sort of economically valuable tasks. And we take all those tasks and then turn them into a benchmark.

3:40And one of the benchmarks that we released for Kimmy K3 has been Code Arena. and specifically Kimmy K3 is at the top. It's number one in front-end code arena. So it's the best at the sort of lovable vibe coding use case. Now the best model in the world for that, of course, to which hundreds of millions of dollars and inference spend will accrue is Kimmy K3. So ahead of 5.6, ahead of Fable too? Ahead of Fable, which is absolutely remarkable. So let's just pause here for a second. this is kind of, you know, Fable was the model that really caused a bit of a reckoning in terms of cost. This is an open source model, which, I mean, that means it's significantly cheaper, right?

4:28Am I correct there? That's correct. And I will tell you, the gap between Kimi K3 and Fable is not small. It's a win rate of in the order of tens of percent. You know, let's say 10 % is the difference in win rate between kimi k3 and fable 5 which is you know it's it's not the largest gap that we've seen on arena but it is substantial and indicates a real delta in the change in performance and now that comes at the cost of a model that is a sonnet level cost that's a double-edged what do you mean by that a sonnet level cost what do you mean by that oh like claude sonnet it's it costs the same amount of claude sonnet it's three dollars in oh god $15 output token.

5:07Same price as Claude Sonnet. So the significance of this is actually quite interesting. First of all, of course, the raw performance is amazing, but also the fact that it's no longer the case that open source model is 50 % of the quality at 10 % of the cost. It's actually that it's frontier level performance at a reasonable cost that you would see for a mid-tier frontier model. And so I think the ecosystem is evolving perhaps in an unexpected way in the sense that open source is actually also quite expensive, which speaks to the fact that the margins are actually not that high on inference. Right.

5:49So that was coding. I mean, just give us a brief overview here. Other benchmarks, what other benchmarks does Kimi K3 really excel at based on what you're seeing? Well, you know, it does sort of like, it's definitely a frontier model on a lot of the static benchmarks. So if you're looking at things like Sweebench and Tau and so on and so forth, it's sort of in the pack along with Sol and with Fable, although not necessarily leading on some, not leading on others, but still right there in the competition. In Text Arena, which is our benchmark for sort of chat GPT-like capabilities with user interaction and so on, Kimmy K3 is top 10, but not top one.

6:31And in fact, it's not even the top model outside of OpenAI Anthropic. It's about tied with Sol, but it is beaten by MUSE Spark 1.1 and the basically entire bench of Claude models, 4.6, 4.7, 4.6 Thinking, 4.7 Thinking, and Fable, all better chat models for chat interfaces. So right now, for text, who's at the top right now? Right now, it's Fable. It's Fable 5. Yeah. So Kimmy K3, from what I'm hearing from you, I mean, really, I mean, it's coding that has really impressed me. Of all the use cases you could use this for, that is the one that's standing out. Let me ask you this. Are there caveats to this at all?

7:22I mean, being atop a benchmark is certainly one way of measuring performance. Are there other considerations, though, around the way the model is made, you know, how easy it is to use that, you know, sort of might tell us something about how it's actually adopted by businesses? You know, we have yet to see how it's going to be adopted by businesses. It's too early to say. But we're going to have more to say on that over the next days and weeks about how steerable the model is and how companies are finding it easier and easier to use. Of course, the one thing that I will say is KimiCampro is quite a large model.

7:58And so it's not like you're going to be able to run this locally on your laptop. You are going to need a cluster. It's going to need quite a bit of infrastructure. And I would expect that many businesses will be relying on third-party inference providers to give them the compute they need even to do the inference on this model. Let me ask you about the comparison. DeepSeek came out. This was a model that was not developed in the United States. Obviously, it caused big market swings. People asked the question, hey, do I need all the compute that I initially thought? Do you see any similarities with this model and the impact that DeepSeek had on the ecosystem?

8:42Are they two separate stories? What do you think? Well, I think it will cause a reckoning in the capital markets. And the reason for that is that it brings into question what the dominance will be of the closed source models. In a world where there's a narrative violation against the distillation story, then what will happen is that people will say, well, open source models coming from China, They have so many benefits. Businesses can incorporate them into their own infrastructure without relying on a third-party service, without worrying about privacy and data leakage, without worrying about evil third-party companies training on their data and stealing their businesses.

9:27There's so many reasons why people would want an open-weight model. And in a world where that can be done for free, then the question comes, why would we be paying for closed-source models that are worse for our businesses, that are less private, and that allow these companies that want to be in every single business, in FDE for every single vertical, why would we be allowing them to witness our business and see our data, learn from it so that they can improve and eventually one day steal our business? Why would we do that? And so then what's likely to happen is an accrual value to these open-weight models from which, of course, the revenue model is quite different and worse than the closed-source models, which could then cause a collapse, you know, in the way that people are seeing the compute markets because, of course, the compute markets are driven by the optimism and the revenue of the closed-source models, so on and so forth.

10:27So it could cause a cascading effect. Does it matter that it wasn't developed in the United States? Absolutely. It absolutely matters. And in fact, we should all, you know, if we're patriots and Americans, be rooting for some American open source models to come out as well. Of course, it's fantastic that we see the Chinese developing these models and it's socially positive net-net. But at the end of the day, because of the geopolitical risk, we are not going to see businesses comfortable building their entire stack on Chinese models. do you think that the the i i hear you and that is certainly something we hear from from companies as well do you think that that is becoming less of an issue though because if i think of the last no if i if i think about no absolutely not okay why do you think it's less of an issue no no i i'm asking based on um this is this is purely based on you know uh narrative observation what i i feel like people are more willing to use models not developed in the United States.

11:34That's just based on conversations we've been having on the show. The governments of both sides would disagree. What I see happening is actually more hawkishness on both sides, that the Chinese government is actually talking about export controlling Chinese models. And the US government is also talking about banning Chinese models because neither side wants to incur the geopolitical risk. and of course there's so much competition happening between the us and china it is it is absolutely becoming higher and higher risk we even have the u.s government banning u.s models from release let alone chinese models i think it is actually a huge topic right let me ask you about another model anastasio so uh thinking machines lab came out with their model inkling uh what's been the reviews that you've been seeing on on that particular model listen inkling is the first open source model to be released uh by thinking actually the first model period to be released by thinking and of course it's incredible to see that we have a company that is leading the american open source charge the other contenders would be nemotron rc reflection mistrol of course uh in europe and there's a couple other players um but england came out at the top of the u.s open source race today that said it is not yet a frontier model by any stretch it is if you look at the wider field of top 10 including chinese models it is number 10 number 10.

13:09so there's nine chinese models above this and so they have quite a bit of room to climb now if you know anything about thinking machines and it's sort of you know common knowledge it's not something that i know specifically because of our business but they've had quite a bit of restructuring so on and so forth people leaving people coming in restructuring their team and so maybe to some extent it's impressive that they were able to do this in a relatively short amount of time since the current structure that they've had six months or so that said um there is a long long way to go and if you look at the field of closed source models as well as open source models, it is number 37 model.

13:45So we still have, you know, there's a ladder to climb, but let's all root for them. Because that is, we need a model within our shores to support the enterprises that want to avoid the geopolitical risk that we just discussed. If we don't have an American open source model, businesses will simply be incurring greater risk and be less comfortable adopting AI. Is there anything about the AI ecosystem in China that has, you know, made them so effective at developing open source models specifically? It feels like in the open source race, I mean, it's always been a question of which models are coming out of China.

14:26Why do you think they have always succeeded in that particular arena no pun intended yeah well i think that um the strategy is actually quite a good one which is to diffuse the model everywhere and then become the best place for running that model and therefore uh gain resources and you know they need to have some kind of an asymmetric advantage the chinese model providers over the u.s model providers um there and but because they have a lot of drawbacks too so they are absolutely limited by the amount of gpus that they have they're absolutely limited by the fact that there's there's a legal regulation side the u.s that keep you from shipping data to China um and so because of these facts they need to have some way of attacking the market and it may as well be open source I think that's also compounded by the fact that by there's a lot of central planning in China um and IP is treated differently in China than it is in the US and And perhaps there's less of a focus of IP from what I know of the Chinese market than the U.S.

15:27market. So those are all reasons why the open source strategy would be better for them. And then how do you monetize? Well, there's two ways of monetizing an open source model. The first is become the best place to run that open source model. And then you can earn money through inference. That's what a lot of these companies are doing. And the second is through licensing. So you say, okay, I'm not going to run my open source model, but i'm going to partner with inference providers in the u.s that are going to run this open source model and if they reach a revenue of above x dollars in running my model then i'm going to do a rev share with them and that's going to be in the licensing of the weights and of course because we respect licensing laws here that is exactly what those companies are doing and that's going to accrue back to revenue for those companies i want to ask you where we started the conversation we talked about the pace of the release of models lately.

16:15It feels like they've been coming fast and furious. And we've been talking on the show about where AI research is at, the idea that continual learning is very much something that people are looking forward to. Why do you think models have been getting released so quickly? And let's put competition aside. I mean, there is a lot of incentive for these companies to put things out to continue to stay competitive. but is there anything about the current state of you know what these models can do whether or not they can help develop other models that is sort of uh influencing the pace of the release of these new products of course i think you're getting at the concept of recursive self-improvement the idea that the models might be able to train themselves and so on and so forth i don't think we're quite there yet right although we are even seeing in our business that the ai um that the existence of ai is actually making our employees more and more productive by the month we're seeing productivity per employee actually rise even at arena at our small scale we're 70 people and we are still seeing that effect happening so i'm sure it's happening at the big labs as well that said i will say there's a funny dynamic you know first of all you know even competition aside with these labs which is that the release cycles tend to be kind of bulky in the sense that you'll get a quiet week with a couple models here and there and then you'll get a week with 10 models and why is that i think these companies are all hearing about each other oh kimmy k3 is going to release the following day oh we know that we're not quite as good as kimmy crate 3 so we should probably front run them so that we get a little bit of good press imagine if inkling and kimmy crate 3 switch right if they switch on the release date and kimmy crate 3 comes first inkling doesn't they don't release they're not they're not coming out with it yeah exactly they don't they don't look quite is good and so it creates this dynamic of a mad rush everyone hears that everyone else is releasing and so everyone is competing to be first especially the players that know that they're on a weak foot and not to say that that thinking machines is i think that the model is great it's the number one u.s open source model and again we should all be rooting for that but it's also true of the closed source model providers as well right so last question for you anastasia is what does kimik 3 i mean what does it mean for the businesses of anthropic and open ai as you see it well i said this uh online and i'll say it again i think we're going to see a lot of companies that are distilling american companies distilling chinese models now instead of the other way around i'm you know i'm sure open and anthropic uh have their own way of training models but even beyond those two uh you know the the kimmy k3 is going to serve as a base model for many many u.s models um and we're going to see that happening now and if the trend continues that the chinese models uh get better and better over time it may pose like a serious threat to the to the closed source business model because of all the drawbacks that the closed source uh providers have hey are you expecting a comeback for for google here they've kind of fallen off the map here with the absolutely i actually have a huge belief in the google team you should see those people they're really smart Their team is like excellent, excellent.

19:22I mean, it's also true of Anthropic and OpenAI. I mean, I just wondered if they... I think they're good. I expect the comeback. Wait for it. Just wait for Gemini 4. I'll bet you it'll be, you know, at the top. Well, hold on. We can't even get 3.5. You know, 3.5 is delayed now, according to Bloomberg. So we got to get there first. Yeah, that's right. That's right. All right. Well, Anastasios, I want to thank you for coming on. That is Anastasios Angelopoulos, the co-founder and CEO of Arena here on TITB. Netflix reported results last night. Investors were not pleased with their reaction today. The stock is down 40 % over the past year.

20:00It fell considerably today after the results. To discuss that and other big headlines from the week, I want to bring on co-executive editor Martin Peers for this week's edition of The Editor's Cut. Martin, welcome back to the show. It's great to have you here. Hey, Akash. How are you? I'm good. I'm in San Francisco. I'm not in you have not been here all week i know did you miss me are you very quiet office very quiet we're able to get work done okay well we'll be back monday so uh sorry um okay so netflix reported results last night uh what were the highlights um well the highlights were kind of what we expected which is that every quarter this year their growth is slowing it fell to i think 13.4 percent It's really come down by three or four points over the last few quarters, and they are projecting it will continue to slow.

20:53This is not a surprise. I mean, this company is now very large. At the end of last year, it had 325 million global subscribers. They are making more money than anyone else. They said that this year they would have$12.5 billion in free cash flow, which is more than Disney. um so they're really a giant company what is interesting about netflix now is that television used to be a hit driven business and the networks you know in the old days um their advertising revenues would reflect the nature of their hits and their popularity so how many people were watching in bc would influence the advertising it got in the year after you know So if it had the number one lineup in one year, then it would go into the upfront very strongly.

21:50But we're in a very different environment now. Netflix, it needs hits, but not in the same way as the networks need it to. It just needs to make sure that people continue to subscribe. And because they have such a vast array of programs, they're really able to keep people subscribing. The issue now is, can they maintain growth that investors have come to expect? I mean, I think over time that's just not going to happen because it's very hard. I mean, they've probably got pretty much everyone that they can get. There are a few. Right. And the interesting issue they have, which they highlighted yesterday, was that one way they can get more revenue is through advertising, which is still a very small part of the business for them.

22:49And for that to grow, they need to have more live programming, particularly in sports. And they made the point that they're going to spend, I think, 5 % of their programming on live, but that will only account for 1 % of the actual programming. But that brings in new subscribers. So it's an issue. Sports is extremely expensive. that the TV people cannot make money on sports really, but they need it to maintain an audience. The more Netflix moves into that area, the more their enormous profits will get diluted. So, you know. So it's a classic growth versus profitability trade-off here, basically, that the company is going to have to grow.

23:41And then the other issue is, we've already seen this happen, And they have to resist the temptation to do some big acquisition to, you know, as a way of trying to find a way to grow. They obviously fell prey to that when they did the deal to buy Warner Brothers Discovery. Fortunately, they got overbid by David Ellison. And now they're insisting again, we have a very high bar for M &A. Well, okay. Hopefully they will maintain it. What is it? an 80 billion dollar bar it's like we won't go above 80 maybe or something like that and you know yeah the stock is down a lot but really if you look at the multiples uh and where it has traded in the past it's still actually a fairly expensive stock for that industry so uh things are not as bad as some people might make them out to be so i guess just thinking about the levers here that you think they could pull i mean there's subscriptions there's advertising i mean uh they haven't yet done anything in the way of of like free um you know there's some speculation they could do some free stuff which would maybe help on the advertising but they made the point that they don't want to do anything that would um cannibalize the existing uh subscription so they can't put anything on the free that uh would prompt anyone to cut the cord and to uh cancel And honestly, the free services that are out there mostly show crab and will draw a particular part of the audience that doesn't want to pay.

25:18And those are people that advertisers, a few advertisers want to reach. It's not like it's nothing, but it's not a great part of the business. So I wouldn't overdo the idea that Netflix is going to go heavily into free stuff. i think well and this is the i mean if you think about like you know now they're doing the podcast too i mean are you sort of think about the brand of content that you find on netflix i mean on the spectrum of like yeah they're just putting more and more stuff on and yeah their aim is to be the place that you can find anything um there's always going to be something there for you and i think they've done a pretty good job uh of that so um i think they're going to be hard to beat let's uh go to a different story uh that also made news this week we didn't cover it uh stripe reportedly is possibly considering a bid with a private equity company to acquire paypal it was uh the the number that reuters reported 53 billion dollars is what's being considered you wrote about this in your column this week what was your reaction well it's what stripe and advent the private equity firm are thinking about it's right right I mean, I saw a report this morning that PayPal's board thinks it's not high enough, which is kind of obvious because it is a very low bid.

26:39I mean, PayPal, that price, I think, is$60.50 a share. And the stock had fallen recently in the 40s, but that's very low compared to where it has traded in the past few years. So Stripe and Advent are coming in when the stock is weak. I think PayPal appears to have been struggling. It's lost ground. It's facing Apple Pay and Google Pay and everybody else. It's barely growing. But look, this is a company that makes a lot of money and it wouldn't be that difficult to revive it. They own Venmo, which is a very strong business. And I think people believe it can be, you know, Like, sorry, Venmo is growing and people believe Venmo can make more money.

27:32So there's a lot of potential here. I think what's more likely to happen is that the PayPal board will open a process. They will invite people to make offers. They really have to do that now that somebody has made an offer. They can't just— Who else do you think could make offers? Well, look, it could be anyone. SpaceX is an obvious one because Elon was one of the original investors and executives of PayPal. And he has, with the years, been talking about starting a payments app through X, which, you know, he's talking about doing this X money thing. So PayPal would fit. He could easily afford it if he issued stock.

28:16And the cash that he got, if he bought it, would help pay for some of the AI investments that he's making. So I think that's one, seemed to be obvious possibility. There's other possible bidders, could include the banks, maybe. I, you know, I'm not an expert in that area, but I would guess that there would be quite a few people lining up to make an offer. Right. If that happens. Well, we will have to see how it all shakes out. Martin, I want to thank you for coming on. That is Martin Piers, our co-executive editor here at The Information. Google is ramping up its competition with NVIDIA on the chip front, and it too is getting creative with the deals it is striking to get chips in customers' hands.

29:02Amir Afradi is our other co-executive editor, and he published some exclusive reporting about that effort in our AI infrastructure newsletter. I want to bring on Amir to share more about what he knows. Amir, welcome back to the show. It's great to have you here. Good morning. Okay, so let's talk about this meaty newsletter that you had in our AI infrastructure column this week. There was a lot to it, and I want to see how much we can get through. Who is Google targeting with their TPUs right now, as far as you're hearing? Yeah, so the Tensor Processing Unit effort, as you probably know, is a very long-term journey for Google, which was one of the first NVIDIA chip users for AI.

29:45I found out about NVIDIA only because I was covering Google. And fairly early on, they made this investment to make their own chip that was more tailored to their machine learning models. That has turned out to be quite a smart move, not only for Google's internal margins because it runs a lot of its search and other AI workloads and advertising on TPUs, Now it sees an opportunity in this moment with AI to actually capture a significant portion of the AI chip market. So as part of that is trying to figure out how to scale up this business, but do it in balance with the fact that Google is also one of the biggest still buyers of NVIDIA chips.

30:35And that's because it has its own cloud business for all kinds of other companies that want to use those chips for AI workloads. It's trying to get some of those customers to use TPUs in addition to GPUs. So it's a very, very complicated balancing act. And so who specifically has Google been targeting lately? You had some reporting on this. Yeah, so there's this whole array of NeoClouds. As you know, these very kind of GPU-specific cloud providers. And they exist only because NVIDIA allows them to exist. These are companies, many of them are former cryptocurrency mining firms that have pivoted their business to just servicing other companies that want GPUs.

31:22So these companies were using GPUs for cryptocurrency mining. Now these GPUs are used for everything else. So they have some experience in running data centers. There's a whole group of, there's probably hundreds of them. There are only like half a dozen that actually matter. And Google has been talking to those companies for a while about whether they'd be interested in bringing on TPUs, which I should mention that we're in this huge CapEx boom as well among the biggest cloud providers in particular. and Wall Street is going to be scrutinizing CapEx more and more and more and more. So for a company like Google, which is, again, one of the biggest CapEx spenders as it develops its own data centers increasingly, working with these other companies, these neoclouds, is a way to kind of lower a little bit that CapEx, get somebody else to take on data center development, in particular with TPUs, which increases Google's operating expenses, but the benefit is it could decrease its CapEx expenditure.

32:29So anyway, Google's approaching these companies and saying, hey, you should work with us as well. It's nice to not be completely dependent on NVIDIA. Jensen Wong, as you may or may not know, is a fairly competitive person. There's a reason NVIDIA is the world's number one most valuable company. What's the pitch? I mean, take us inside the pitch for Google. What do they say? Is it cheaper? You know, is it just diversity of chip that is the root of their argument? Like, what's the pitch they're making? Yeah, a few things. So, look, NVIDIA is the gold standard and does not seem to be, you know, near to losing that title.

33:11But Google says that it has a few different advantages for companies that want to use TPUs specifically for, you know, transformer-based AI models. These are chips that have not changed as much as NVIDIA's chips have in terms of the design and the kind of networking and other hardware that you need around it. So if you're a company that is installing them in data centers, you may just have fewer surprises. Whereas with each new NVIDIA generation, you're dealing with a whole host of radical changes, which is part of what makes NVIDIA great, but also is incredibly painful for the companies, including these NeoClouds that want to install them, as well as the traditional cloud providers that also are installing them.

33:56So Google is saying we have a lot more consistency here, and these chips are, according to Google, much more efficient in some ways compared to GPUs for running AI models. Some of this has yet to play out, but I think there is something to that. And if there's one single competitor that Jensen is paying close attention to, it's Google and TPUs. Now, you had some reporting that Nscale is a neocloud that Google has been talking closely with. Do they have a deal in place or what's the status of the negotiation? Yeah, there's no deal that we know of. And one of the things that happens is Jensen has Intel across the industry.

34:37He generally, not always, generally knows who's talking to whom. And in this case, he did find out that Google was talking to Nscale. And there's some moves that happened since then that made some folks around this feel like he was providing additional incentives to N-Scale to maybe get them to not use TPUs. N-Scale says on the record that's not officially their perspective here. But yeah, Jensen's not silly and he has not been shy at all to use his company's balance sheet to help his biggest customers. And now Google's really targeting some of its biggest and most loyal customers. It's just extremely nerve-wracking for someone like Jensen.

35:24Although, another thing you point out in the newsletter, though, is that while NVIDIA is throwing its own balance sheet at the problem here of getting people to adopt their chips, that is also possibly an approach now that Google themselves is considering, no? Yeah, yeah, absolutely. This is another reason why Jensen cares about Google probably more than any other company. Google has lots of cash. It generates lots of cash all the time, just like NVIDIA does. So if anyone is in a position to really seed the market with competing chips, it's Google. So it has been discussing these kind of backstop deals where it will essentially be guaranteeing to lenders who are going to lend to different businesses that may want to buy TPUs that Google will pay out if those companies that are borrowing for TPUs can't find buyers or renters or what have you for these data center chips.

36:26And so, yeah, so those conversations are happening. Now, important to point out that Google also has a joint venture with Blackstone to essentially develop a TPU cloud provider. This is another leg in this stool of getting TPUs to be used externally outside of Google. And I guess from Google's, and I mean, for Google to backstop these TPUs, I mean, for them, I guess the argument is, hey, we can just use them ourselves for our own cloud offering, right? So, I mean, it's really no, we're not losing anything because we could just use the chips ourselves, basically. So they, and that's something that NVIDIA, I mean, they don't really have as big a cloud business, obviously.