Why Jensen Huang Rejects OpenAI & Anthropic's AI Safety Pacing Calls, Apple’s New Direction

0:13Welcome, everyone, to The Information's TI TV. My name is Akash Pastracha. It is Wednesday, September 16th. Before we get going here, I want to flag a scoop that we published late last night. The information has exclusive reporting that Instinct is in talks to raise$1 billion at a valuation of roughly$10 billion. Sequoia Capital and Benchmark are both in the mix. There is a lot more in the story. I encourage you to check it out on our website. Today on the show, we are breaking down all of the announcements out of Dreamforce and what Jensen, Sam Altman, and Dario all said at Salesforce's big event.

0:52We'll then look at how the prospect of AI pacing is forcing startup founders to rethink their businesses. We also have new reporting for you on Apple's potential return to the server market and why it is mending its relationship with NVIDIA. We're going to close out the show with robotics. We're going to be talking to Tudor Intelligence, which today unveiled its new generation of industrial robots. It's going to be a great show, so let's get right on into it. It is day two of Salesforce's annual Dreamforce conference. The CRM software giant has unveiled some new products, including a new model.

1:28But the event is also happening as the debate over pacing AI development remains top of mind. I want to bring on Laura Bratton, author of our Applied AI newsletter, for her takeaways. Laura, welcome back to the show. It's great to have you here. Hey, Akash. So let's talk Dreamforce. I want to talk about all the new products that they unveiled in a second here. But I mean, the big headline that I saw was that you had all three of Dario, Sam and Jensen on stage at separate moments with Mark Benioff. And this obviously comes after the whole flurry of perspectives on AI pacing shook out. So what did each of them say on stage about this topic?

2:11Yeah, so it was pretty funny. It was during the keynote that Mark Benioff, the CEO of Salesforce, spoke with Dario first and then Jensen at separate times. And it's as they were walking around the conference. And it was kind of a moment where Mark Benioff, who's like six foot six inches tall. Six foot five, actually. I looked this up. Six foot five. Okay, okay. He was just towering over these AI leaders. and it was in sort of looking down on them, which I thought was sort of a poetic moment for Salesforce. But it was really interesting. So Benioff, when he was talking to Dario and asked him about pacing, Dario just repeated what he'd said and took a moment to make a slight dig at OpenAI and said, you know, we may not have had this high profile security incident, but we think safety is really important and that, you know, we should come together and set standards for the industry.

3:09And I thought what was interesting is that just, and later on in the day during a fireside chat, Sam Altman echoed Dario's comments. But what was really interesting is that a little bit after Dario spoke with Mark Benioff, Jensen was with Benioff and Jensen was saying, if the model providers think their products are unsafe, then they shouldn't release them. And it's up to them and we don't need more regulations and safety and speed don't necessarily need to be at odds with each other, which I thought it was just really interesting to have the contrast between all the different leaders all on the same day at the same event.

3:51And then Salesforce itself sort of tried to take this middleman approach where Mark Benioff was saying, you know, if companies want to slow down, they should slow down. And if they don't want to slow down, They don't have to slow down and kind of just tried to stay really neutral, which it's kind of impossible to do right now. But that was what Salesforce tried to do. So he didn't seem to take any specific side on this, Benny Offer. No, he just told every guy leader, what you're doing is incredible and amazing. And, you know, tried to. We love it. We love all of you. We love all of our friends equally.

4:26Yeah, I'd be interested whether that approach is the same behind closed doors. But, you know, that was the public facing statement. Do you have any sense at all what the perspective might be behind closed doors? We're just brainstorming here. I'm just trying to think about enterprise software companies because you and I talked earlier this week about how investors were loving enterprise software stocks momentarily after this on Monday. Yeah, I mean, I won't pretend to know what Salesforce strategy behind closed doors is, but I will say like they've benefited greatly from this partnership with Anthropic and seem to be going deeper and deeper into that partnership.

5:13I mean, Claude Forrest. Claude Forrest was the big announcement on their earnings call a couple months, a week ago. And, you know, Salesforce is an investor in Anthropik. And so, you know, they benefit from Anthropik's success. Right. Well, so let's talk about then the new products that they did unveil at Dreamforce then. Where do you want to start here? What were the big announcements worth flagging? Yeah, I think the biggest thing is just AI Force, which all these new names at all the Salesforce conferences always sort of give me a little bit of a headache and I have to really sit down and try to define what's what and what's a renaming of an old product.

6:00But AI Force was pretty significant because it's this software layer that lets you work with any AI agent you want to connect into Salesforce and work through Salesforce. So, you know, Claude Force is one version of AI Force. It's, you know, this connector that lets you plug into Salesforce through Claude. And then there's also Slackforce, which is the same version of that, but for Slack. So really, this just shows Salesforce is stepping away from its attempt to force you to use its own proprietary AI products within the four walls of your Salesforce account and is really opening up its platform for any customer to use any AI agent they want.

6:48And the software layer makes sure that whatever AI agent you're hooking into your Salesforce account has the proper permissions and is doing the things that you want it to for whichever employees are using that agent. Okay. And how does that then differ or contrast with the model that they released? Um, so it's just separate. So then they, they also released Koa, which is a new reasoning model that they developed on top of NVIDIA's open source Nemotron model. Um, and it was basically trained using synthetic data specific to, um, customer relationship management. So it's a model that will be cheaper for customers to use specifically to do different sales tasks with, maybe like resolve a customer service issue or track down a sales lead.

7:44The model is specifically post-trained to handle those kinds of tasks and more cheaply maybe than like a leading Anthropik or OpenAI model. And this is all part of Salesforce's strategy to be a software layer that's independent of whatever AI you want to use. That's kind of the message they were pushing. You know, or at least Matthew McConaughey when he came on and did his whole highly produced video. He was there this week? Yeah. He's done the commercials, right? Yeah. So we got Matthew McConaughey, Usher, Gwen Stefani, Miley, no shortage of celebrities at Dreamforce. Miley Cyrus is - That's what I saw in Mark Benioff's tweet.

8:30I haven't seen her around town, but apparently she's there. Okay. So Laura, I mean, as you think about AI Force and about COA, and you think about Salesforce's strategy traditionally as a software company, and we've talked about outcome-based pricing pricing with you as well, sort of their transition to try to be top of mind here. Is this a shift in strategy for Salesforce or is this very much the same playbook? It's really a shift in strategy. Like I said, Salesforce previously was pushing this idea of wanting you to use the AI agents that you built within the four walls of Salesforce rather than making their platform open for any AI agent to connect into.

9:16But really in the last six months, We've seen Salesforce shift to try and make it easier for customers to use any AI agent they want to tap into their Salesforce data. And this is really interesting because the whole debate in the software industry is whether software companies going, quote unquote, headless or basically their user interfaces becoming less important as companies start to use AI agents to tap into their software systems. whether or not that's going to be a bad thing for software companies and their ability to monetize the sort of seat-based subscriptions that they've historically charged customers.

9:55But as companies have shifted their pricing models, we've seen that maybe it's not a bad thing if the user interface isn't quite as important. The fear is that companies like Salesforce will become glorified databases, but Salesforce is trying to get ahead of that argument. Yeah. Well, and it sort of makes me think that, you know, the same way that startup founders, I mean, they build the product first, they worry about monetization far down the line. That sort of seems to be the approach that Salesforce is taking here, which is that it's sort of unclear what impact this could have on their ability to monetize their data.

10:31But I guess the hope here is that if they open the platform and they get enough people on the platform, they will figure out down the road some way to make money off of it, which, as you just pointed out, remains to be the big question, essentially, because the old thing they were making money on isn't really the moat anymore. Yeah, I mean, I will say, you know, I've been to Palantir's customer conferences, I've been, you know, to Salesforce's and ServiceNow's. Seeing Salesforce's AI Force demos and, you know, the new model they're unveiling all this stuff, it was definitely felt a lot more sophisticated than previous sort of demos at conferences they've had in the last couple years with AgentForce.

11:20So we'll see what that actually looks like in practice for customers. But I do think, you know, they seem to be thinking really deeply about how to charge for outcomes or usage. So I would say they're in the middle of that transition and they seem to be trying to be really flexible with customers about how they charge for their AI. I do think, like you said, you know, Salesforce historically will introduce products and then figure it out later behind Salesforce. So we'll see how this all plays out. Yeah. Laura, let me ask you one last question, going back to where we started the conversation, which was about the AI pacing discussions on stage.

11:59Mark Benioff trying to take sort of a middleman position here. How much confidence do you have that Salesforce and Mark Benioff will be able to keep staying, you know, towing the line here between the two perspectives? I mean, do you think that there is a world here where Salesforce has to ultimately take a side? Have you thought about that at all? I think that software companies are able to sort of play the neutral ground because they are sort of acting as these intermediaries and trying to act as if they're agnostic to whatever AI model that you want to use. So they're in a really good position to say, we're not going to take any side and we'll benefit, you know, whether Anthropic and OpenAI succeed or they don't succeed because you can use any model through our software platform.

12:57But I do think, you know, any drama from Anthropic and OpenAI, it does feel like software companies behind the scenes might be rolling their eyes a little bit. But I couldn't speak to that, of course. Right, right. Great. Well, Laura, I want to thank you for coming on. That is Laura Bratton, author of our Applied AI newsletter here at The Information. the prospect of ai pacing has startup founders thinking about what this could mean for their early stage businesses too my colleague julia hornstein who covers venture capital went out to her sources this week to figure out how early stage founders are thinking about this i want to bring on her bring her on to share more about what she learned julia welcome to the show it's great to have you back tell me a little bit about how silicon valley is reacting here i mean we know the reactions from Mark Benioff, from Jensen.

13:54These are all the big fish. What's going on with some of the earlier stage folks? I mean, I think that their reaction has been pretty mixed. Some of the investors and founders that I spoke with for my column yesterday said that regulation and cooperation across the labs may dampen competition and give the labs even more power over the future of the broader tech ecosystem. Other investors that I chatted with said to expect more consolidation across the tech ecosystem, especially as smaller labs don't have the same capital and talent pool that the biggest fish have to keep up with this regulation that they may look out into their prospects and think that selling is the best option for them down the road.

14:45But then others think that regulation makes it all the more important to invest in the apps on top of and the infrastructure underneath AI models and to help them, you know, run better to make apps that, you know, the labs don't necessarily make themselves. So there's definitely been a widespread reaction across the tech ecosystem. I want to get into both of those perspectives here. I mean, on the first side of things, so this idea that the smaller labs, smaller startups, their businesses could suffer here because the big will get bigger. Anthropic OpenAI, I mean, they have the loudest voices at the table.

15:23Is the fear here that basically whatever rules Anthropic OpenAI would encourage the government or as sort of an independent reviewing body, the rules that they would be set, what is the fear exactly? Just that Anthropic and OpenAI would have more influence over making the rules benefit them and they would sort of leave out the smaller startups? Is that the fear? I think that the fear is that, you know, in creating some sort of oversight body that the labs and that one of our colleagues reported over the weekend that they've been in talks to create or at least cooperate on some sort of, you know, pacing or regulatory matters.

16:13I think the thinking is that they will all be collaborating and sharing ideas with each other, but that may leave out, you know, the smaller players who obviously will need to, in some ways, conform to regulations if they're, you know, dealing with the government. But these smaller players will be not at the, you know, initial... They just want, I mean, their voices, it's not even that, they can't definitively say that it will hurt them, just that they won't have a voice at the table and they won't be able to sort of represent themselves in these discussions. Exactly. Right. Okay. And then on the flip side, though, you have VCs telling you there's still a lot of stuff to invest in around the edges here of the actual model itself.

16:59What are some of those pockets that VCs might look more to now? Yeah. I mean, there's always more stuff to invest in, right? So another investor that I I chatted with, who's backed several applied AI startups, said that if the frontier models are in some way pacing their development of products and models that could allow app layer startups to have a lot more time to make their tech better. So basically, if the labs are slowing down their rate of research and development, this investor described to me. Perhaps an app company focused on a specific niche within their product offerings could take advantage of this lull, this pacing to make advancements of their own that the labs may not be as focused on.

17:54He seemed really animated by this. And then, And then, of course, there's also the argument of it seems like so many in Silicon Valley are really interested now in infrastructure plays. So things that help the models run quicker and better. So I think that that's also an area that people are, of course, looking into that feels maybe a little bit like pacing proof. What about security? Did anyone mention security to you? Yeah, some people did mention security to me. I think that, you know, especially as we see kind of more of these hugging face like incidents come to light, that will become all the more important for startups to focus on.

18:41So some of the investors that I spoke to also mentioned security as well. Great. Julia, as you think about where you would like to focus your reporting next, I mean, what are the big questions that come to mind? I mean, for me, I guess one thing I'm thinking about is, you know, these staggering growth profiles that we've seen for these companies, just seeing how that really, if that does taper or accelerate. I mean, that's how I might be thinking about it. What are some of the questions that you have? That's certainly a question on our mind, especially as, you know, these biggest companies still seem to be burning a ton of cash.

19:22Um, I'm also really curious about one conversation that I had with another investor. Um, he was telling me about a startup that had recently found, um, customer sales to be increasingly difficult. And this is an app player company that sells to small to medium sized businesses. And, um, he was chatting with his founder who told him that, um, in the past week, because of all these like doomsday headlines, um, it's been really difficult to, uh, sell to these mom and pop shops because one of the first questions they have is, is AI safe? Is AI going to kill us all? So I think that that really speaks to how this discourse around safety generally has seemingly broken containment.

20:06And I think that startups, especially ones that rely on customers that aren't necessarily directly entrenched within the tech ecosystem, that's a dynamic that they'll certainly have to navigate going forward. Right, right. It kind of reminds me a little bit on a smaller scale, albeit of, you know, tariff uncertainty, whenever there is macroeconomic uncertainty, I mean, the reflex is hunker down, quit your spending, you know, cancel all your sales calls, stuff like that. It seems like that could be happening sort of on a smaller scale, at least with respect to AI spending. It's like, we don't even know what this tool is, you know, what the prospect of these tools are.

20:46So cancel the calls right now. And so that's certainly an interesting prospect if that's shaking out. Julia, I want to thank you for coming on. That is Julia Hornstein, our venture capital reporter here at The Information. Apple and NVIDIA have had a long and complicated history, but new exclusive reporting from my colleagues, Aaron Tilley, Shaner Liu, and Phoebe Liu reveals the relationship between the two companies could now be turning more collaborative as Apple considers returning to the server market potentially. I want to bring on Aaron to share more about their reporting. Aaron, welcome back to the show.

21:22It's great to have you here. Thanks for having me. So what did we learn about Apple's server ambitions, Aaron?

21:30Josh Gruenstein:Yeah, so Apple is looking to potentially get back into selling servers. Years ago, it would sell servers called XServe in the kind of the 2000s, And they weren't a hugely successful product. They were good, but Apple didn't spend or prioritize it very much. And Apple, with the demand it's seeing for AI on its hardware, it really sees an opportunity to start selling into that market. And when you say server, I mean, let's just define that for us because I think we know chips and GPUs and then we know the chips that Apple puts inside phones. What exactly is a server here that we're talking about?

22:17Josh Gruenstein:Yeah, I mean, it's a computer, but it's a box. It's a box you can sort of mount into a server rack and then manage. Like a data center. This is a data center server. Yes, that will fit into it. Precisely. And so what would be the chips that it's connecting here? Would they be like NVIDIA chips or what would be the chips here? So it would be Apple's internal chips. We have heard that it will be a collection of M8 Ultra chips. This is their Ultra high-end chip that they put into their Mac Studio products currently. but they're looking to connect these multiple versions of these chips into a single server.

23:04And again, basic questions here because I'm not a data center expert here. So these M8 chips, would they then basically be competitive with GPUs, TPUs, these AI chips that we talk about on the show? Are these a different class of servers or are they basically competitive?

23:26Josh Gruenstein:I think it would be more targeted towards inferencing. So that's the running of these AI models versus training, which is where GPUs really will continue to dominate likely and your TPUs also play. So more in the inferencing market. So this is will be targeted for customers who want to do sort of so-called sovereign AI where the data really stays within their enterprise, within their systems. and it doesn't leave to a public cloud. So very, first of all, if a customer is concerned about data leakage, if it's in a highly regulated industry, if it's a government, this would be an ideal sort of server-based product for them.

24:09Right. So where are they then in their considerations here to get back into the server market? Do we have a timeline or a status update on it at all?

24:20Josh Gruenstein:Yeah, so it's looking like they're targeting a few years from now, 2029. The key consideration was getting enough components in order to sell an external product. Apple, you know, obviously kind of maps out future supply pretty carefully. And this is something they needed to get enough supply in. So this is them going to their suppliers, asking for enough components. and it looks like 2029 and the inmate ultra chip will be ready by then as well. But just so we sort of make clear what we know and what we don't, so this is still in the works right now. I mean, there's a possibility that they may not pursue this or is this looking like a sure thing?

25:04For sure.

25:05Josh Gruenstein:They may not pursue this. They may not see this as a market worth tackling right now or in the future. So it may be canceled. So it's kind of interesting, though, because, I mean, the fact that they're considering this at all very much signals the moment that we're in. Why do you think they're getting into this market? I mean, we had you on the show a couple of weeks ago talking about their business and enterprise focused event. I think that they hosted. Apple has not traditionally been, you know, a B2B company necessarily. early, you know, put aside outfitting all these offices with computers and stuff.

25:49So what's the strategy here, do you think?

25:53Josh Gruenstein:So they've really seen this insane unexpected surge of demand for their Mac computers, their Mac minis and their Mac studios. These are like contained box-like computers that people are really buying up like nothing else for AI applications. And they're like AI labs are buying them, startups are buying them, big enterprises are buying them to run AI workloads. So I think it's the opportunity they see for these chips to be useful in all sorts of contexts outside of the consumer. It's really this demand that they're seeing over the past year. And how does NVIDIA fit into this story? How are they working with them on this?

26:39Josh Gruenstein:So yeah, as part of the story I wrote today, there's also a sort of warming of relationships between the relationship between the two companies after sort of decades of really kind of arms-length relationship. And Apple has been looking at NVIDIA's networking technology called NVLink Fusion. This is a technology that connects traditionally, historically connected NVIDIA chips, so as if it were to work connecting multiple GPUs as if it worked as a one chip, one system. And so recently, NVIDIA has been selling this technology outside of its chips to other companies, for example, Amazon and their chip making endeavors.