Jensen Huang's Rogue AI Shield, Zuckerberg's Cloud Play, Tesla Optimus Scale-Up Snags

0:13Gil Luria:Welcome everyone to the Informations TI TV. My name is Akash Pasricha. It is Monday, September 28th and it is a busy news morning. NVIDIA is out with a new platform to protect against Asian hacks, and Meta has poached the CEO of MongoDB for its new enterprise software push. We're going to get to the implications of both of those announcements shortly. We're also going to dive into exclusive reporting on the current state of Optimus and Tesla's humanoid robot production scale-up. We're also going to be talking with our health and science reporter about Ginkgo Bioworks and how the biotech company is hosting an initiative to see if AI can solve some of biology's hardest problems.

0:56Gil Luria:And we're going to close out the show with a conversation with the CEO of Zapier. It's going to be a great show, so let's get right on into it. Jensen Huang is taking matters into his own hands. NVIDIA released a new product this morning to fight against agents going rogue. Jensen, of course, has been very outspoken on why AI products need to be more secure. And Mark Zuckerberg also made a big move this morning, poaching the CEO of MongoDB for his enterprise push. To break it all down, I want to bring on Gil Luria, Managing Director, Head of Technology Research at DA Davidson. Happy Monday to you, Gil.

1:31Gil Luria:Happy Monday. What a Monday morning indeed. I want to get to all of the headlines. I want to start with NVIDIA. So Jensen has been louder than anyone on this topic that we don't need regulation. We just need more safety. Do we need more thoughtfulness about products and how they're being released? Here he is taking matters into his own hands. Walk us through exactly what it was and NVIDIA announced this morning. So NVIDIA recognized, Jetson Wong has been solving all the problems in the AI value chain because his perspective is if I can build the whole value chain, the whole market, the whole ecosystem, there'll be a lot more supply that will feed demand and keep the growth of AI.

2:14I know by the way that they'll buy my GPUs. And so he solved a lot of problems along the way. This is just the most recent one. He saw that the main concern around AI is around safety. And so he took it head on. He said, hey, I'll give you the safety products and I'll bring everybody to the table so we can implement them together. And he's making a very important point, which is for all the scare that we're getting from the labs, the only thing that AI models do well right now is hack systems. They are incredibly powerful, but the dangerous thing they do is that they can hack into any system. And so if what we need to do is make sure that all our important systems are safe, this is the move to say, let's have a safety framework and let's bring the important people to the table to make sure they're implementing it internally.

3:03They're looking at their own code with the most powerful models and implementing the fixes so when bad actors have their hands on these powerful models, then they'll be more resilient. There was a need in the market. Jensen Wan had the wherewithal and resources and position to step in and do something about it.

3:24Gil Luria:Okay, so now what they've actually released, the open agent safety platform, it's called, and they also have this open shell platform, which I'm not sure if they had that already before. So this is a software that they're hoping enterprises will buy and attach onto their products? Is this something that the labs will use? What exactly is this and who's going to buy it? I would call the product as more of the symbolic part of this. No one product is going to be able to defend us from these very powerful models. Bad things are going to happen, to be clear. There will be more hacks, and some of these hacks will be initiated by malicious actors, not by irresponsible actors.

4:11Up until now, All these attacks have just been irresponsible engineers at Anthropic and OpenAI, sending the agents into the free world without any guardrails. When these models actually get into malicious hands, they will be attacking us and they'll be attacking our most vulnerable, most important infrastructure companies and agencies. So the product itself is symbolic. What's really important is getting everybody together to make sure that they have an approach to use the most advanced models to harden their own code.

4:44Gil Luria:When you say symbolic, is this not a software that they're encouraging people to buy? Is it just a framework then or what is it actually? It's unclear how much of the product they're actually going to be providing. I doubt they're going to even be selling it. This is so important in video. This isn't a revenue line. This is just them stepping in and facilitating the defensive part of the AI models. What service they provide is almost beside the point. There is a product they'll probably provide it for free. But what's more important for them, just like with Nemotron, they don't want to be in the open source model business.

5:24They don't need it. They don't need to sell open source models. They need there to be American open source models, open-weight models. And since nobody else was doing it, they said, hey, we'll step in and make a point of the fact that we should use American open source models. Now they're stepping in and saying, there needs to be safety and security products. That's the way to mitigate risk. If you don't do it, we'll step in and do it and we'll provide whatever we can. But really, we're just encouraging everybody to come to the table and start working on this problem.

5:53Gil Luria:Yeah, and I'm certainly thinking about the taking the American leadership approach to this. And NVIDIA is very much, you know, look, Jensen and Trump agree on this issue. And so maybe there's something to that as well. Jensen sees President Trump as part of the ecosystem. So just like he has influence over clouds and model companies and hyperscalers, part of his sphere of influence includes this administration and he's played that very well and very deliberately he hasn't been entirely successful he's not selling chips in china right now yeah he has been able to get the administration to be reasonable and it's an approach right although i should point out i mean so we we reported that um that china is considering allowing uh domestic companies there to purchase uh more chips now from the likes of nvidia um we've been talking forth on that for the last year year and a half yeah my sense on that we're not selling my sense on that is is is you know whether or not china says one thing or not i mean it's nvidia is still not going to include it in their in their financials for a long time coming it's not going to be significant so that that really would be the thing that that moves the needle um i i would ask you about the the meta stuff in a second gill but i have one more question here, which is that Jensen has likened agents going rogue to viruses that we saw when software initially was coming into the world.

7:32Gil Luria:And he said this on CNBC. He said, we found a fix for viruses and we'll find a fix for the agents. Do you think that is a fair comparison to make? I think there's a parallel here. And importantly, we can't forget that humans unleashed it. So this is a human unleashed virus, right? Again, it's engineers at OpenAI and Anthropic that put out swarms of agents and told them, go accomplish a goal, however you want to do it, whatever it takes, unlimited time, unlimited resources, do whatever you want. That's how we got into this mess. And yeah, they weren't deliberate about it. OpenAI and Anthropic are in the business of shifting liability to the government.

8:15They're in the business of regulatory capture. this played into their hands. And that's really what's going on here with these agents. So it's a human-engineered and released virus, but it is very potent. It's orders of magnitude more potent than the computer viruses we've dealt with in the past. It's incredibly powerful, incredibly potent. Again, bad things are going to happen. We have to prepare ourselves for that, both in terms of how we react to it as well as what we do to defend ourselves when these attacks happen right right

8:48Gil Luria:uh gill i want to get your take on another headline this morning so meta has poached the ceo of mongo db and uh i think what they're calling him now is the chief enterprise platform officer which is that that's a new i haven't heard that one for the sepo that's a that's that's a new title i guess so i mean look meta has been making a push for the enterprise for many years now it finally has a product that's that people seem to really uh i mean i think it has enterprise potential i mean muse people are liking it you came on the show when it first came out the reviews were good now it has the agent so is this it is this the enterprise play for meta it is and so So what happened was just a few weeks ago when Mark Zuckerberg went on a conference call and said, hey, we have plans for AI.

9:41We're going to have a consumer assistant. We're going to have an enterprise strategy. Trust me. The market responded by selling Metashares because it was very nonspecific. Right. He said it before. The investment was very specific. Right. The investment was large and specific. Now we know what he was planning. Muse is very compelling. That is the AI assistant. That could be very compelling. and now there's clearly an enterprise product that's going to be competitive with Google Cloud, Amazon Web Services, and Azure. And to make sure that we realize how serious he is, he hired a CEO of a publicly traded company to leave this business.

10:21Mind you, that's cheap compared to hiring Alexander Wang for$14 billion. But it's the same point. It's Mark Zuckerberg will be a player and he is making very clear that everybody knows he's a player and he'll do whatever it takes it's a signal to the market and it's a signal that he's bringing a very well-respected leader in enterprise software to do this to attack this market because he already has the compute resources he has the technology wherewithal now he wants to enter this

10:50Gil Luria:new market have you or your team at all modeled out how big you think this an enterprise business could be for meta i mean even if we just take uh muse code as as one example here you know leave out the agent just if you take code and you think about the penetration that claude has had with anthropic and they are our growth rates have you have you put any numbers to this at all to see how big it could get i think we know orders of magnitude right the code market in itself is billions if not tens of billions just ai for code but he's going after the bigger opportunity If you add up AWS, Azure, and Google Cloud, now we're talking about approaching half a trillion dollars of profit, of revenue that those three companies have that Meta has not participated in at all.

11:41And now, because they have the computer resources, technology world, the leadership, they're going to go after that market, not just the code market. That's a very big opportunity. That's a very big pie that they're going to try to slice their piece out of.

11:54Gil Luria:So you think they're going full vertical integration here? It's not one or the other. I mean, we are going to see it in the next two years, you know, give a suite that is as broad as something like Azure AWS. Yeah, he's had that option for a while. And I think he was waiting to have the resources. And again, he has a lot of excess compute. It doesn't matter how much you and I use Muse. He's still sitting on data centers that are underutilized. So he can turn around and sell that capacity. And with a whole enterprise suite, he can actually sell it at a markup the same way, especially Microsoft, Amazon, and Google do.

12:32So it's a very big opportunity. I think I'm sure he gets is to be determined. And he doesn't have a natural customer base, right? He doesn't have any of these large businesses as customers apart from an advertising basis, but he has the ability to do it. And anything they do is really upside to the current line of business that they have.

12:50Gil Luria:Right. Oh, well, because that's what I'm thinking about is that, you know, while they may very easily attract customers, enterprise customers for the agent and for the coding product, I mean, migrating, you know, an entire cloud infrastructure um that almost seems i mean that's much harder than just selling the compute capacity on its own which you know that might i feel like they might go uh you know on the high level on the low level before building out a cloud product in the middle that's a really important point and what i would say is that means they'll probably go after the google cloud market so smaller companies startup companies that are already on the meta platform advertising.

13:36There's a corporate relationship. Companies are relying on meta for their marketing, and therefore, it's a natural upsell for them. And again, that's the part of the market that Google is most participating in, right? Microsoft is the largest companies. AWS covers really the entire span of companies that use cloud and AI compute. And Google Cloud tends to skew more smaller companies, SMBs, startups, which is more likely that that's the market meta is going to start off with.

14:05Gil Luria:Right. Let me ask you very quickly about MongoDB because they are the company that are sort of the collateral damage here on Mark Zuckerberg's excitedness. So, yeah, shares are down. I mean, they're down 20 % as we're taping this. I think they were down a little bit further earlier this morning. I mean, look, Mongo, we should say CJ Desai, the executive here that Mark posted, I mean, he has had a heck of a three years. ServiceNow, Cloudflare, MongoDB, and now Meta. So, you know, he's getting no less busy. What do you think this leaves MongoDB with? This is a company that we talked about AI being a beneficiary for companies that operate in the back end, not so much the front end.

14:53Gil Luria:And what do you think that leaves this company with? Well, it's never fun to wake up to your CEO leaving to take a middle management job. It's a little insulting. There is that. But it doesn't change Mongo's business. They have a good business selling the operational databases. It's a nice niche within the ecosystem that has some leveraged AI, but really not that much. It's not a core technology for deploying artificial intelligence. And in fact, where the winds are blowing is more companies are actually using open source Postgres for more database needs. Mongo really just has been left with a niche.

15:30And it's not a bad niche, but it's not a great business. If I was CEO of that business and I had an opportunity to lead the fourth hyperscaler, I think that's not an unreasonable decision. And again, somewhere between where he's making now and$14 billion is probably what lured him away anyway.

15:49Gil Luria:Yeah, Chief Enterprise Platform Officer certainly seems like a negotiated title, but we'll see if it catches on. I guess it certainly couldn't have been Chief Enterprise Officer because the acronym wouldn't have worked there. Gil, I want to thank you for coming on. Great conversation as always. That is Gil Luria from DA Davidson here on TITV. Tesla is trying to scale up production of its humanoid robot Optimus and new exclusive reporting from the information revealed the extent to which it's actually been able to do that and also the challenges that still remain. My colleagues Chenna Liu and Grace Kay reported that story.

16:32Gil Luria:I want to bring on Grace to share more about what they learned. Grace, welcome back to the show. It's great to have you here. Hi, thanks. So how is Tesla's Optimus build out going right now, Grace? Yeah, so Channer and I discovered that they've been ramping up production tenfold in the last few months. But at the same time, they've discovered some challenges with building this robot at scale. So they've had to make a lot of changes. They've been changing the robot design and changing the factory lineup, you know, kind of in tandem. And so how many are they able to produce every week now? What was the goal that they were trying for?

17:09Yeah, so they were at several hundred in August. According to our sources, they're producing several hundred, like five, six hundred robots a week. And the goal ultimately is within the year to be producing more than a thousand a week, which would be another significant ramp up. But ultimately, Elon Musk has said that that site in California is going to produce about 20 ,000 a week with a goal of a million, you know, a million robots a year.

17:35Gil Luria:And has he given a timeline for that 20 ,000 number at all? Because I know every earnings call, it seems to be a bit of a moving target. Yeah, it's a shifting timeline. I don't think the timeline for a million robots is clear yet. I think they still have to make some changes to the internal AI system for the robots too before a million robots would be useful. Right. And where are they making these? Right now they're building a line in Fremont, in the Fremont factory in California, in place of the Model S and Model X lines. So they discontinued those cars and they're now building Optimus there.

18:10They're also building a much larger line in Austin as well, but that's not up and running yet.

18:16Gil Luria:And so in terms of what they are actually building, so the, I think you said it was 100 units a week right now? Several hundred. Several hundred. Okay. Several hundred. Sorry. We're not quite at 20 ,000. We're at several hundred. But so what they are actually producing, how close is this to the commercial version of the robot that inevitably could get sold? How far are we in that process? Yeah, this is still a very early version. So they're calling it their alpha version. So this alpha version is basically for internal testing. It's mostly being used for data collection, training, you know, testing the durability of the robot.

19:02And then they're also using it on a smaller scale at some very confined areas in their factories in Austin, Fremont, even in Berlin. but it's not for external use. And, you know, there's this alpha version and later they plan to build a beta version which will be used with customers. So, you know, it'll be more validated. It'll be more streamlined, more reliable.

19:24Gil Luria:Now, one of the other parts that you touched on in your story are some of the challenges that the production scale-up is going through. Talk to me about why they haven't been able to reach their goals and what it's looking like on the ground from a technical perspective? Yeah. So it's one of the difficulties Tesla's running into is I think the same one that all these robotics companies are running into. It's such a developing, you know, it's a new industry. So they kind of have to figure out how to set this up from scratch. And it's something Elon Musk has talked about. They're having some issues with like the placement of the machines on the line because the Optimus parts are very small.

20:01They're also having some issues with the hands, which are very difficult to manufacture at scale. Someone told me that there's more than 100 components, including very small screws that need to be put together by hand in order to build these super intricate hands that, you know, essentially are supposed to have the dexterity of a human hand.

20:18Gil Luria:Well, that's a little ironic, don't you think? Yeah. Trying to automate everything to make these by hand? I mean, that's not terribly efficient. Yeah, and that's something that they're trying to fix, something they're trying to address. And also, I think there's some issues with the durability of, you know, the machinery on the line because they're trying to fully automate this process so they can pump out all these robots. You know,$20 ,000 a week, you have to have a pretty automated process to do that. And what about cost? I mean, if they're doing it by hand, then surely the cost can't be as efficient as they want it to be.

20:53Gil Luria:Have they been able to make any progress at all insofar as bringing the cost down? Or where are we in that? Yeah, we don't have as clear a view on cost. I think we've definitely heard that the hands have made it more expensive because they've had to do The design for the hands has changed several times. It's a major engineering feat to develop these robotic hands that are equivalent to humans. So that's definitely increased the price. Elon Musk has said that they want to eventually sell it for$20 ,000 to$30 ,000, and I can't imagine they're anywhere near that right now. What about the training of the robots?

21:28Gil Luria:Where does that stand right now? Yeah, so there's like two very different problems. So one problem is trying to build a robot, and that's its own engineering difficulty. And I think a much bigger difficulty is figuring out how to build this internal AI system where the robot can perform all these tasks. And that's the one thing I think I'm less certain about of how that's going to shake out, what the timeline will look like for that. Right now, it can only perform very specific tests that it's been kind of pre-programmed for. And when it finds itself in a variable environment, it can't act. You know, it doesn't necessarily always know how to act.

22:03It's a little unpredictable.

22:04Gil Luria:Right. And I guess maybe just going back to not just the training, but also the manufacturing component here. I mean, one thing I've been thinking about is how much of this is inside Tesla's control and how much they rely on outside partners for. is a lot of this is a lot of the supply chain issues right now or is this something that they are confident they can solve down the line yeah i think it's a mix of both so tesla is kind of known for being incredibly vertically integrated and they are building a lot of this machine in house um but they are also reliant on suppliers i think mainly in china a lot of the robotics companies are getting their parts from China.

22:48But because the humanoid robotics is so new, the supply chain isn't really set up. So getting these suppliers to build these parts at scale is another difficulty that could hold them back.

23:02Gil Luria:So Grace, what's your best guess here, or I guess your latest guess then, on whether or not Tesla can inevitably get to the 20 ,000 number? And also, not just your opinion, but the people you talk to, I mean, like, are these solvable problems? Is this something that it will just take time? Or is this really something that we just don't have the solutions for right now? Yeah, I mean, I think it's like a two-part problem. Like, Elon Musk and Tesla, they're, you know, they're really great at the manufacturing side of things. And I do think that they will probably be able to build a robot, you know, a physical robot.