Inside Nvidia’s $12.9B Hugging Face Buy, Nvidia Q2 Sales Double to $96.2B, Salesforce Shares Jump

0:13Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Thursday, August 27th. Today on the show, the information has exclusive reporting that NVIDIA is acquiring open source AI platform Hugging Face for$12.9 billion. We'll talk to the information's deals reporter who broke that story. We'll then unpack NVIDIA's quarterly results. Revenue more than doubled. We'll talk about that with an analyst. And we're going to close out the show with Salesforce's quarterly results. Investors seem to like them, and it's big, expanded partnership with Anthropic. Does this mean that the SaaSpocalypse is over?

0:53It's going to be a great show. We're going to find out the answer to that question very shortly. Let's get right on into it. NVIDIA is continuing its deal-making streak. The information exclusively reported last night that the company has agreed to buy Hugging Face for$12.9 billion. My colleagues Amir Afradi, Valida Pau, and Phoebe Liu wrote that story. I want to bring on Valida to share more about what she knows. Valida, welcome back to the show. It's great to have you here. Great to be on. So was NVIDIA always in the mix for Hugging Face here, or is this a surprise to us? I feel like it started with an unsolicited bit, according to our reporting, and then deals talks have happened between Hugging Face and NVIDIA, it happened throughout that time.

1:40But it makes a lot of sense for NVIDIA to be in charge of a strategic asset that hosts a wide range of open source models that like developers can use um nvidia has been kind of really pushing aggressively for like its own open source models and like you know jensen huang like nvidia ceo's beliefs that um having successful open source models would have preserved like nvidia's kind of dominance as like you know these models can run on nvidia chips who else was in the mix do we know in the bidding process um hacking phase counts a lot of like kind of strategic investors are like um already so we do know that like salesforce is one of like the potential bidders and um our sources have said like you know lots of the um big hyperscalers like you know our amount of parties are interested but like towards the end nvidia rent like won well they can pay a lot more than anybody else too i imagine so that that certainly helps um i want to ask you about the the multiple here i mean uh you reported that hugging face is generating 150 million dollars in revenue right now which would put it at about 80 times revenue the the multiple i mean is this just a reflection of how in demand open weight models are right now i feel like um for an 80 times ford revenue is a pretty expensive deal but i felt like you know for NVIDIA, the multiples here doesn't matter because like, you know, it's a very strategic move for them.

3:13And, you know, HuggingFace kind of helped like for their open source kind of approach from the very beginning. And then like, you know, they are like wide distributors of like all these open source models that like develop can use, you know, develop their customer pay like a monthly subscription fees and like, you know, and pay, you know, how much computing powers they're using. So like, you know, for NVIDIA, like, you know, this can allow them to get access, like a wide distributors of like these open source networks that will like, you know, ultimately benefit them. But Huckin' Fee has been growing like faster, you know, in recent months because like these exploding interest in like, you know, cheaper, you know, open source models that like others have been using to, you know, cut costs from close source models, like from Anthropic or OpenAI.

4:00So like, you know, they have been growing faster. Right. Recently. So you've been watching and scooping much of the NVIDIA deals chatter as of late. We had the poolside licensing deal, I think, a couple weeks ago. We had the investment in perplexity that I believe you reported, and now there's this. Are you expecting NVIDIA's deals to continue here? Do you think it has enough? What are people in your deals orbit thinking about this? Well, this is like a series of deals that NVIDIA has done just in August. So they have put in lots of investments in kind of these land and power firms that help develop AI data centers.

4:43You know, that includes Lansium, SB Energies, and of course, that like$100 billion backstopping deal for open AI data centers. They also invest in like, you know, data labeling, start up Mercore, do the licensing deal with Poolside and also like, you know, discussing investing in like perplexity. So it seems like Jensen Huang is on a deal-making spree here, and we don't feel like he's going to stop as he built out his vision for his empire and relies on or now increasingly more important for his open-source ambitions. Right. And then the other area that I'm personally watching, and we did have a company, Emerald AI, on the show earlier this week.

5:28They are a data center energy optimization company. NVIDIA has invested in them. I mean, that sort of goes along with the thesis that NVIDIA, the next gap they might be looking to acquire companies in or maybe invest in is the energy sector. And so that's an area that I feel like they could be making more headway in the weeks to come. Oh, they could be. I feel like they have done lots of kind of like aqua hires in the past for their own teams. You know, they can also still kind of do more on chip sites and, of course, like, you know, on their AI efforts. Right. I want to talk about another story that you scooped yesterday.

6:08So you and our colleagues Rocket and Amir wrote that SoftBank is looking to take a majority stake in a humanoid robot company, OneX. Tell us about what we know about this this chatter. so um softbank is going to is talking about buying a majority stake in one x which is a humanoid robot makers um what is interesting is that um one x um previously had talks with kind of like an open ai last year about like a possible takeover but like you know that deal didn't pan out so like you know this is interesting in a way that like you know softbank's major shareholders is kind of like discussing a majority stakes in this company that open ai is backed and had mna talks in um but broadly speaking you know just fits into like you know softbank ceo masa's grand ambitions and like you know physical ai is going to be the next frontier like the future of softbank and he has been kind of like assembling you know pieces of like you know different robotic startups and investments to kind of like you know build out this vision of him that like you know believes that you know robots and like physical ai will be like the next um frontier of the ai development and and a breakthrough in technology.

7:21How does the$6 billion valuation compare to past valuations that 1X has had? So last year, 1X had to raise$1 billion in the$10 billion valuations, but they never really raised the full amount. Our sources tell us that it's like, you know, less than half the target. So in a way that this is a down round, but in January 2025, they're only like valued in under$1 billion. So it's still a huge step up for like a company that's really not shipped many robots or their products. And like, you know, they're not generating lots of revenue. So it's still like a steep price to pay. How far is 1X along in its own humanoid development right now?

8:03I've seen some of their models on their web. I mean, they seem to have some prototypes. I mean, do they have something that they're selling right now? How far are they along? long I think they have like started pre-orders for like you know their robots and like they have promised or like told customer they're gonna like ship it by the end of 2026. so um we'll see if they can like make that timeline and we'll if customer have the robots in the homes and I mean Valida just to take a step back here given that you uh spend your day talking to everybody in the deals orbit I mean Jessica had a great column this week talking about the deal landscape broadly, why she expects it to continue and in some cases even pick up.

8:46I mean, from your view, why have deals picked up so much this summer? It seems like it's not just NVIDIA. I mean, these acquisitions are fast and furious. What has changed in the last couple months? I feel like tech deals has been sort of slow for the past year. And now suddenly there's like all these rage about like, you know, big price being paid to AI startups. I felt like, you know, still fundamentally, you know, tech companies believes in like, you know, AI is worth investing in, they're worth investing in, like they still need like great teams to help them kind of achieve that. And there's like so many companies, for example, like there's so many, like maybe not older, like, you know, maybe older, like, you know, after 2022 startups that, you know, they may not be able to take off, you know, they could find homes like you know in these strategic players and like you know these perfectly high in demand startups you know it's become more and more important so like you know tech companies feel like they need to make the jump here but of course like you know dc like those people talk about like capex's new m a so maybe like you know as tech companies spent so much on capex like in early this year they like you know slow down the deal machines but like you know they could start back up again right well valida i want to thank you for coming on that That is Valida Pau, our deals reporter here at The Information.

10:05Okay, speaking of NVIDIA, shares are up today after the company once again managed to lift investor expectations. Revenue more than doubled, although free cash flow fell over 50 % because of payment terms from customers. I want to bring on Ray Wong, CEO of Constellation Research, for his thoughts on all of this. Ray, welcome back to the show. It's great to have you here. Hey, thanks for having me here. And I'm actually here in New York. Look at this. There you go. Well, we like the East Coast here as we do. What stood out to you from the results? Look, 23 out of 25 beats, six quarters in a row.

10:45What was really standing out was not the results. We all expected good numbers. It was the guidance. 70 % revenue growth, Colette said, versus 45, 46 % was the expectation. I think that was the reinsurance that this AI-fueled stock market needed. Even with the supply constraints, 70%. So, I mean, this was kind of interesting that they're still seeing constraints. You know, we are so fickle in this market. We're like, there's not enough demand. This is all fake. What the heck's going on? Then we're like, oh, my God,$700 billion this year,$1.2 trillion next year,$1.5 trillion the year after that.

11:22Okay, do we have enough supply?

11:24Jackson Ader:I mean, that's something like, oh, great, we got to worry about supply now. You know what? They'll figure it out. I mean, we're seeing the capacity. They're investing in the plants. They're doing the AI factories. But I think what's interesting is the way NVIDIA is layering their future revenue. You can see it by the investments they make with the startups. You can see it with the investments. I mean, in the last four weeks in the Valley, I cannot tell you how many people I know that were going to visit NVIDIA to get a Nematron partnership in place, right? And that's the open weight models that they're pushing.

11:51And they're going into that space pretty quickly. So you can see that not along with the investments, the AI factories, along with the fact that chips are now a asset class on its own. I mean, all these things they're doing is designed to continue to further that demand and make sure that there are no surprises in the market. Right. I want to get to their broader investment strategy here in a minute. But I mean, if we just stick with the results. So the day sales outstanding was interesting. I mean, they seem to have looser payment terms here for some of their biggest customers. 45 days to 60 days.

12:25I mean, our executive editor, Martin Pierce, made the point last night in his briefing column, what, you need 15 more days to pay off billions of dollars in these contracts? I'm not sure what difference it actually makes. I mean, are we missing some context here on the ground? What's your understanding of why this is coming together? I don't know. Maybe someone's making money on the float. No, I'm just kidding. That's a lot of float. I mean, I don't know. I

12:49Jackson Ader:think it's probably maybe changed the way that they're billing. I don't know for the ground in particular, but I do know that people were focusing in on that number, trying to understand, you know, is there a shift in terms? I would say that probably the biggest shift we're seeing right now is just how people consume and what kind of outcomes. And there might be some promotions and other MDF that's built in there that we might not have seen. But to me, that wasn't necessarily a red flag. I think for me, what I was really interested was just looking at, you know, gross margins coming down. That was a great set that expectation.

13:18I think it was also important to see that, you know, even with the adjusted gross margin, these forecasts were kind of like on the big side. I mean, data center was at 43 and a half billion and that's going to get to, you can see that getting the 50 to 60 industrial was at 41.7 billion. So in our models, I mean, we're seeing something we're seeing like, this is, this is real. And now we're worried about supply chain costs being passed on. And so the rate increases, that's the one that we're going to look at to see how much more it's going to be to buy something on Rubin or how much it's going to be when we get the frame in.

13:48So I think the next set of chips that are in the product root. What about the 25 % of revenue coming from the AI Frontier Labs? Was that higher or lower than you were expecting? You know, it's actually online. We're going to see the AI Labs continue to get better in terms of token usage, and they're going to need more capacity. But it's not a concerning amount to you, 25 % from the labs. I would be concerned if that was 40%, 50%, because we're hoping for diversification, right? We wanted to go to physical AI. We wanted to go to sovereign AI. We wanted to get into CUDA and software over time. That's going to be important.

14:26Right. Now, the other number that was interesting to me was they talked about sort of the per gigawatt revenue opportunity and that getting to$40 billion per gigawatt, I believe it was, for the Rubin family of chips. Just help us unpack what that means. I mean, is that just pricing power from NVIDIA raising prices with a premium product here? Is there other stuff baked into that? I think the best way to think about it is like you're buying chips. You're willing to pay more for chips if they're actually able to get to some level of scale. Is it faster? Is it better? Is it cheaper? Am I consuming more energy?

15:08I'm consuming less energy. Am I getting more MIPS and I'm getting more processing per unit? That's what people are looking for. And I think NVIDIA has built out these curves over time. They're starting to introduce that terminology so people understand that, hey, it's okay if you're two generations behind because you buy the latest version of our chips, it pays for itself. So help me understand. So is it just that the Vera chips are more expensive or is this... I can crank more out on the Vera chips than I can on the other chips. I'm going to crank more out on the Feynman chips is that 10x might be like 3x, 4x.

15:45I mean, that's what they're looking at. Okay. And so in other words, is that saying that even for the older family of chips, I mean, we should expect that per gigawatt figure to go up in the future as well? Or how should we track that? You expect that the performance is definitely going to go up. And if that gigawatt has a correlation, remember, we're also going to have a used chip market as well. And so the question is, how do I value my used chips in my used data centers? versus what's in the new data center. And I think we're going to be able to start seeing those comparisons against performance.

16:17Got it, got it, got it. Jalapeno, OpenAI's chip. We saw some results from that. Well, we heard from OpenAI, their results on that. Do you think it is a threat to, maybe not NVIDIA, but to its competitors in this market? You know, everyone's going to build custom silicon, and that's really important, right? Not all of them are going to be able to get it to the level Google got to TPUs actually being the one big threat to GPUs. But the demand is so big that these are still dropped in the buckets in terms of the impact to NVIDIA's revenue. But what I'm most worried about is if someone in China figures out how to actually get the lithography right, build their own chips, crush TSMC on a multidimensional chip that can actually perform better than an NVIDIA chip, that would worry the crap out of me.

17:11And I think you're starting to see the crackdown now on exports and the export bans. We saw recently in Taiwan, there were several individuals that had been caught at TSMC for smuggling chips throughout. We've seen smuggling going on in Dubai. We've seen smuggling coming through Singapore. And so they're starting to crack down on those pretty hard. Yeah. Hey, I was intrigued by a tweet last night, an ex-post from Nikesh Arora, the CEO of Palo Alto Networks. Did you come across this on NeoClouds? I did see a little bit of that. It was kind of interesting. I mean, the idea that he was basically saying that, and he did confirm, I mean, lower down, he said, CoreWeave and Nebius, two of the biggest neoclouds, he sort of suggested that they are doing more than just cloud services.

17:59And so they weren't really included in his analysis here. But broadly speaking, he suggested that the business of neoclouds, I mean, once the supply and demand economics, economics equalize, that the valuations for these companies will drop considerably. What's your read on that? Do you agree with him? Because this is the big question, right? Well, I mean, 400 % revenue growth, that's not sustainable, really. I would liken this to the housing supply shortage that we have today. Like today, I can go to a large builder. They can build track homes. They can build big-ass apartments and condos, and it's like the same thing.

18:37There's always going to be demand in these neoclouts for something that's an empty shell that I can build to my suit, so I can build to suit. Something that's actually turnkey, that's tuned for my industry, tuned for my performance requirements. Something that's actually beyond turnkey. I think all these things are going to happen. So I think as the neoclouts continue to differentiate, you're looking at what CoreWeave and Nebius are doing in terms of getting down to the details. Like you can custom build something out of the box. And the fact that they're always going to have the latest chips, there is definitely still going to be a market for it.

19:06Now, is that at 400 %? I don't know if it's going to be at 400%. So in other words, you disagree with his thesis here that valuations will drop markedly. I disagree only because we're seeing a very different kind of demand emerge as customers are actually buying things that wouldn't have seen possible before. Just like, for example, if you want a trillion parameters for one tenth of the cost on-premises, you'd do it as Samba Nova Systems. And to be able to do that before, and that's air-gapped, that's protected, I would go there if I wanted to get to a value game and be on-premises. If I wanted the latest version of GPUs and I was only going to use it for a year or two or I was going to use it for a short period of time, or I needed to actually get a burst capacity, I'm going to CoreWeaver and that is.

19:54Right, right. Last question for you. So the NVIDIA shares were higher on the results. If you look at the multiple, the multiple is steadily declining. Do you see the multiple continuing to decline? What should we expect? So we've got this theory, and we've been talking about this, like these Silicon Valley barbecues, right? This PE compression, the fact that NVIDIA is trading at a multiple at 23, which is below the average of 24, right? Like that makes no frequent sense. On earnings. On earnings, yeah. Price to earnings is ridiculous on those ratios. They got PE compression. They were at 45 at one point, and we all thought that was high, but they should be in the 30s.

20:35And so I think the rationale and what's really going on here is because there's not a lot of volatility in the NVIDIA stock. So no one's actually swing trading NVIDIA. And so because of that, it's kind of steady. It's kind of holding where it's at. But if there were one of those like a meme stock, I'd definitely think that they'd have a higher PE value. But there's no exciting news out of them. But that's great that there's no exciting news out of them because most people are holding these kind of like for retirement. These are institutional investors. They've got a good group of those that aren't just going around and trying to like drive stock, you know, play.

21:04They're really not trying to arbitrage the style. Right, right. Great. Well, Ray, I want to thank you for coming on as always. That is Ray Wong, CEO of Constellation Research here on TI-TV. Salesforce shares jumped after its quarterly results last night. Revenue rose 11%, slower than the 13 % from last quarter, but the company raised its revenue guidance for the full year, suggesting it could grow as much as 12 % for the year. I want to bring on Laura Bratton, who authors our Applied AI newsletter, and Jackson Ader, Managing Director of Software Equity Research at KeyBank Capital Markets. Welcome to the both of you.

21:42It's great to have you here. Good to be here. So Jackson, I want to start with you because we spoke to you, I think a couple of weeks ago, I had asked you what to watch for with Salesforce results. And you had said, well, we're all going to be watching to see if AI can raise total revenue for the company. We've got this deceleration compared to last quarter. So did it live up to your expectations? What do you think?

22:08Jackson Ader:Yeah, yeah, it did. I mean, and I think, you know, the stock's reaction today kind of, you know, tells the story. What is important here, and I think what investors are taking away is that on an organic basis, the company still expects to accelerate its subscription revenue growth in the second half of their fiscal 2027 year. year. And even if that level is going from maybe six and a half to seven or 8%, it's not accelerating massively, but we're still on an acceleration train. That's what we heard last night. And so clearly we're getting a nice relief in shares after a tough start of the year.

22:51In other words, the deceleration from last quarter, it's less about that. It's more about what they're guiding to ahead.

22:59Jackson Ader:That's right. And they acquired Informatica last year, and they've also made some, or well, they're planning to make some acquisitions that are also being factored into guidance for the second half. But overall, organic results, both in the quarter, were slightly ahead and organic results or organic expectations for the rest of the year are also moving up slightly. Again, I think the size of the reaction you could debate, but I mean, the company said what I think they needed to say last night. Okay. Laura, what stood out to you from the call last night? I mean, the loudest part of the results was the Claude Force announcement.

23:47Was that what stood out to you or walk us through what you saw? Yeah, I think the Cloudforce announcement was definitely important. It's just basically showing that Salesforce is going even deeper into its partnership with Anthropic and they're making Salesforce more available in Cloud. So you can do more with your Salesforce accounts natively from Cloud than you could previously with this announcement. And we're gonna find more about how Salesforce is gonna monetize that partnership further at Dreamforce in September. I think the other things that stood out to me were just what Jackson was mentioning about organic revenue growth, because we've really seen Salesforce kind of rely on its acquisition of Informatica to boost revenue.

24:33And now we're seeing, you know, it's organic revenue, not necessarily re-accelerate yet, but they're expecting it to. And I think early signs of that were just that they're seeing, you know, contracts with existing customers expand. They said, you know, net new AOV grew its fastest in four years. And I think Jackson could probably speak more to that. And then, you know, the other thing that I thought was interesting was just agent force recurring revenue growing faster this quarter than the previous quarter. And Jackson, I mean, just make clear for us, what is your understanding of the difference between Claudeforce and Agentforce, how one fits into the other, how revenue is recorded between the two?

25:17Do we have any clarity here?

25:19Jackson Ader:Not yet. Not yet. So the product isn't going to be launched. When we get to Dreamforce in September, we will probably have a better framework for what we're looking at here, right? Not just how are we going to make money from this, but also how is the integration actually going to work. What's interesting about Cloudforce is that for the most part, existing or incumbent large software companies have allowed the model developers to bring their models or their agents into the software environment. And this is really the first one that I can think of, or at least the first large company that I can think of that is saying, actually, we are going to allow the model company have its user interface, right?

26:05Jackson Ader:Claude from Anthropic access all of the Salesforce data that it needs to, but the user themselves actually doesn't have to leave the Claude environment. So it's different from what we've heard from other software vendors, but we have no details. That's okay. This is more about the signal that Mark Bayhoff is saying, we are not afraid of disintermediation. I didn't realize that. So this means that you're accessing Salesforce through the Claude platform. That's right. So, I mean, Laura, let's go back to the data sharing wars that we've written about the information. I mean, does this say anything about those walls between enterprise software companies, model companies, those search companies?