Welcome everyone to The Information. My name is Akos Pasicha. It is Thursday, October 1st. Happy October to you. Today on the show, we have exclusive reporting about SpaceX's push into the NeoCloud business. We're then going to unpack the risks and dangers of trusting an AI agent with your credit card information. We're also going to be bringing on the CEO of Cynthia, which just announced new features for its AI avatar products today. And to close out the show, we're bringing on the CEO of Circle for a conversation about IPOs, about AI, about its new blockchain. We might even get to prediction markets. It's going to be a great show. So, let's get right on into it. The Information has exclusive reporting about how Elon Musk is pushing further into the NeoCloud business after initially being reluctant. Our Elon Musk reporter Grace Kay published an in-depth story on that topic today. I want to bring her on to share more about what she found. Grace, welcome back to the show. It's great to have you here.
Thanks. Great to be here.
So, Grace, you reported that SpaceX's AI unit is now making billions as a cloud computing provider. This was something that Elon was initially reluctant about. Tell us about how the company ultimately made this transition.
Yeah, it was quite an evolution. So, about a year ago, executives asked him because they were not fully using the capacity of the data centers. They said, you know, why don't we start leasing them and try to make money that way? And Elon Musk, you know, was not happy with this idea. He basically said like cloud computing is not a sophisticated business. It'll reflect poorly on the AI lab. Like we're just pushing forward and we're going to use this compute for ourselves. And then about a year later, earlier this year, he made an announcement to the engineers at XAI and he said, 'Anthropic is like basically begging to use our compute. We have these, you know, older Nvidia GPUs that we're not really using. We could just throw those to them and, you know, see what happens.'
And we're going public, so why not get some numbers on the board?
Yeah. Get some shots now on Google.
Yeah. Now you look at it and they're leasing compute to Google, Reflection AI. There's a few other deals in the works. So, you know, they've fully gone into that space and embraced it now.
Yeah. I mean, well, look, the numbers are staggering. So, Anthropic initially committed to $45 billion. Reuters reported in Anthropic's IPO prospectus it is committing to an additional $40 billion. I mean, how did they go about making this pivot so fast internally at the company?
Yeah. I mean, they just positioned themselves really well for it. You know, while other, you know, tech companies like Microsoft were being a little bit more conservative about their AI buildout, Elon Musk and, you know, the people at XAI were all in on it. They were building capacity even when they weren't fully using all of their capacity. You know, they were building data centers at this record pace. They got 200,000 GPUs online in 122 days, which is, you know, kind of astounding for the industry. And, you know, they were getting all these supplies and bringing all these people in, you know.
What about these gas turbines? You talked a little bit about that in your story.
Yeah. So, they were buying, you know, hundreds of millions of dollars worth of gas turbines, you know, before they even had sites to install them years in advance. One purchase order was for, you know, enough gas turbines to supply 400 megawatts of power, which is quite a lot.
And so this is ultimately to, you know, bring the chips online essentially, right? Because we know that getting the chips is one thing, building the data center is another thing, but then getting the power is in many cases the rate limiting step. So they actually had that capacity for power even ahead of building these data centers. What about the people on the ground? I mean, I'm sort of curious, building a data center and then operating it and having people on site to, you know, make sure that they're operational and service these cloud agreements. That seems like a different skill set. How have they ultimately gone about staffing these data centers?
Yeah, I mean, initially they had a big push and they were working with, you know, Tesla and, you know, XAI social media site X and now they're bringing in all these SpaceX engineers. So they had an all hands during the summer where they basically asked SpaceX engineers to volunteer to come in, you know, for these six to eight week rotations, you know, keep pushing forward with this data center build.
Right. And I can imagine, you know, one thing that I feel like Elon is really good at is galvanizing his troops or his soldiers to, you know, really work towards a common goal. I almost feel like, I don't know this for a fact, but I feel like maybe there might have been some excitement on the ground to say, 'Hey, we want to build this faster than anybody else.' Does it feel like there was a little bit of that amongst employees?
Yeah, I think there's definitely some excitement, you know, and they went about using some new technologies and like they were kind of willing to evolve and explore with things. So, like they brought in these AB robots to the data center.
There are little robots that they were using, you know, to coil the networking cables, which is something usually a human would do. So, this was like a new evolution. They were looking at like modular data center builds. So basically pre-fabricating the data centers, you know, parts of the data center and bringing it in. So they were, you know, experimenting with a lot of different things, doing some different design stuff, with some Tesla engineers that, you know, came over to XAI. So yeah, they've done a lot of exploratory stuff.
Now, you and I have talked about this before on the show, but I mean, speed has not always been the best way to do things. There have been some reliability issues with these data centers as well.
Yeah, there's definitely trade-offs. You know, when you get something online that fast, one of the things they were doing was not having as many backup systems. So, we've seen some issues with that, especially in the early days. And then more recently, since SpaceX has started helping, they had an outage, you know, a few weeks ago, actually, that brought, you know, a lot of their systems down.
Now, one of the interesting bits of news that you reported in the story is that SpaceX AI at one point earlier this summer was in talks with Microsoft perhaps about an agreement. You know, we don't know the current state of that deal and if that's going to go through or not, but my question is, I mean, if they were able to sign on Microsoft, if that deal did go through, I mean, what do you think a deal like that would say about the data center story right now or about SpaceX's story? I mean, walk me through the implications of that.
Yeah, I mean, I think it kind of proves to the point of, you know, Elon Musk kind of had an early vision for what they were going to do with these data center builds and they're really capitalizing on that. They went in really aggressively, more aggressively than most of the other tech companies. And you know, now they have this compute and they know how to build it very fast and other companies are trying to catch up and you know, now they have to pay XAI maybe for some of that compute.
Right. And what questions do you still have on your mind that you're hoping to get answers to as you watch the story develop?
Yeah, I think I have kind of two big things I'm looking at. One is what other deals are going to come about as a result of this? What other AI labs or companies are, you know, going to start buying from XAI and what is that going to look like? And then also, you know, as they continue to make all this money from selling compute, what does that mean for, you know, the engineers building their AI models, you know, and Elon Musk initially kind of, you know, said that this was a temporary thing, you know, that they would sell to Anthropic, but now they're continuing to build out, you know, this compute business and finding new customers. So, what does that mean for, you know, the AI lab and Grok?
Right. And I'm also curious about the structure of these deals and the length of these deals because as you said, if it's just a temporary measure and if the ultimate goal is to get this compute to service their own coding products, Grokbot etc. I mean, I can imagine, I mean, look, at the end of the day people want compute as fast as they can but also they want predictability. So, you know, do I really want this for three months if I know it's going to end or if I don't know when this capacity is going to be available to me down the road? I mean, I don't know. It's a question for some of these big hyperscalers.
Certainly. Yeah, it'll be interesting to see, you know, if they actually end up being temporary agreements or if it's something XAI continues to build out. Right now, it seems like it's something they're, you know, really going full force with.
Yeah. Well, I mean, certainly their shareholders would be happy about it because again it means more points on the board. Grace, I want to thank you for coming on. That is Grace Kay, our Elon Musk reporter here at The Information.
Should you trust your AI agent with your credit card information? Today The Information published a special report looking at that exact question. It's one that I personally have been thinking about a lot. My colleague Yui Yang wrote about the risks and dangers of trusting agents with so much information. I want to bring her on for more about her column. Yui, welcome back to the show. It's great to have you here.
So, what do you think? Is it safe? Should we be trusting these agents with our credit card information? What's the verdict?
I think it depends on your risk tolerance. And for me, I'm personally not going to put my credit card information.
Now you're sounding like a crypto investor which is certainly the world that you come from.
That's true because these things are at such early stage. The Meta Muse and Instinct are just starting to take off. People are still playing with it and experimenting with it. But putting your credit card information and handing it over to agents I think is a big step and as I report in my story, it really forces all the parties involved from merchants to payment processors to AI companies and consumers to confront new questions over security and data privacy.
So I want to go through each of those stakeholders one by one. Merchants, payment providers, AI companies and consumers. Let's start maybe at the end here. Consumers. Do we have any data right now as to whether or not consumers do trust agents with their credit card information?
So there's a survey put out by ACI and YouGov which showed that only 7% of fashion shoppers currently are willing to trust agents to make purchases completely autonomously without involving the human to sign off approval and that's quite a small number. I think for the most part people are willing to use AI for the search function. They wanted recommendations from AI for products that they're looking for. But then when it comes to actually making the purchase and pushing the buy button, I think people for the most part still wanted to be able to have control over that.
But you can do it to be clear, right? I mean Muse and Instinct, they do allow you to hand over your credit card information. Basically they just say put it in.
Yes. So you can ask them for a recommendation for an item and then you can check out through their app with the payment providers that they have integrated with such as Stripe and that's how you can make a purchase through the app.
Okay. So Stripe is on board with this. What about the other payment companies, the credit card providers? What's their perspective on this?
So for the most part, these purchases that are made through AI apps, they're being done through credit cards and so for credit card companies like Visa and Mastercard they have a big role to play. And when I talk to them it seems to me that one of the most important issues for them to figure out is how to deal with chargeback requests and that's what happens when a human user wants to dispute a purchase that their agent made on behalf of them. And they wanted to get the money back, they will raise such a request with their bank and that's a new area of development where these companies have to figure out how to handle it.
And through your reporting, did you find examples of customers that have had to dispute these charges or have had agents make purchases without their consent?
I think we've seen reports of customers waking up to be surprised by the purchases that their agents made. But when I talk to the credit card companies, it seems to me that for now, for example for Mastercard, they recently developed this process called verifiable intent where they will literally try to capture and record your chat history with the agent to figure out what exactly was the instruction that you have given to your agent and whether the agent had been able to follow your instructions and make purchases that's based on your orders and that's the kind of evidence that they will record and then present to parties involved when a dispute happens. And that's incorporated into their regular chargeback process. So that's one way for them to figure out how to deal with agent-based shopping transactions.
So let's go to the merchants then. How do the merchants feel about all this? I mean, you could make the argument any revenue is good revenue, but of course, we all know that we don't often go to, you know, when we ask for a refund for things. I mean, disputing or filing a dispute with the credit card company is kind of the last resort. It's not the first thing. You actually end up going to the merchant first and saying, 'Hey, I had this issue.' So, do they feel good about this? Is this more business? Is it more headache? Where are they at right now?
I think they feel conflicted about it. Because on one hand for the past 30 years of the emergence of e-commerce, merchants have been told that bots are bad and their whole payment infrastructure is built to block bots to reduce any fraudulent attempts at making purchases at their website. But then all of a sudden in the past year or so there are all these agents that are popping up trying to make purchases and they're essentially bots but bots that are verified and have delegation powers from their human users. So merchants are trying to wrestle with that. Amazon for example banned Meta's Muse from accessing their site and one of Amazon's concerns was that these are unverified undisclosed third parties that are accessing customer information, their purchase history etc. without the knowledge or consent from Amazon and that raises concerns for them. So for merchants they also do not want to lose the shopping data that they can capture when a human shopper is visiting their website and browsing different kinds of products. And that's another concern that they have. But then at the same time merchants wanted to be able to show up in the search result of the AI chat that you use. So they're also trying to incorporate.
They want it both ways. They want their cake and they want to eat it too. They want it both ways.
And most importantly, there's this new trend where merchants want you to go to their website and use their own agent to make a purchase for you as opposed to use a third party agent because in that way they can still retain their relationship with you. They have better control over your agent because it's literally an agent hosted on their own website. Then they don't lose out the data that can be produced from your browsing history.
Although, I mean, in this scenario though, when we say merchant here, I mean it would really we'd have to be talking about the large e-commerce companies like Amazon or something like that, right? Because if I'm going directly to a brand website, I mean I've done all the hard work already. I don't need the shirt company's agent to make this purchase for me. So I was wondering, crypto, I mean, this is where you started your reporting and it's sort of extended into payments now with these agents. Are we seeing any overlap with crypto companies and these agents doing payments? Do you expect to see any of that?
So crypto companies would love to be able to be part of the conversation. They wanted agents to use crypto and set up crypto wallets so that they can make payments. However when I talk to payments industry experts, their observation is that merchants for the most part are not prioritizing accepting crypto payments yet. And that's simply because human shoppers are not yet asking to be able to pay with crypto and for the most part people still are paying with their credit card. So for them there are all these new areas and new issues that they need to figure out as they accept agent payments and it seems to me that crypto is not top priority for them yet.
Great. Well, Yui, I want to thank you for coming on. That is Yui Yang, our crypto reporter here at The Information.
Cynthia is one of the fastest growing AI video companies making a lot of traction lately with its avatars product. The company released another feature today that will let customers use their avatars for interviews, coaching sessions, and sales reps calls. I want to bring on co-founder and CEO Victor Ripperbelli for more on this. Victor, welcome to the show. It's great to have you here.
Thanks, man. Great to be back.
