- 9 hours ago
Category
🗞
NewsTranscript
00:00Let's start with this broader idea that you're talking about building the economic infrastructure
00:04for AI. We've spent many, many months, years talking about literal infrastructure for AI.
00:10Your world's a little bit different. I think just explain the basics of what you mean by that.
00:16Absolutely. So I'm talking to you here from Stripe Sessions, where we have 10,000 people
00:22here working on the economic infrastructure for AI. And AI is changing a huge amount of how
00:30businesses sell and trade online. And so that could be everything from on the consumer side of
00:35things. Customers are now, I mean, you've probably doubted, they're researching products within AI
00:41apps. And we just had Google announcing that they're joining us. We work with Google and
00:44Microsoft and OpenAI and Meta for buying within their apps. And so that's one example where
00:51previously you might have Googled around, you might have looked at some reviews online.
00:55Now you're doing that product research and actually completing the transaction within AI apps.
00:59It's also really changing tech companies, how they sell, because previously they might have sold to
01:03a human. If you're a Cloudflare or a Vercel, you might have sold to a human buyer. Now coding agents
01:08are doing so much work. We can delegate the work to the coding agents. And so increasingly tech companies
01:13are selling to coding agents directly and they're being used by coding agents directly as a result of
01:19real people. And so it's completely changing the landscape. It changes how fraud works as well.
01:23We're seeing new fraud patterns that businesses have to prevent against. But yeah, the big theme
01:28here is that the economy is really, really platforming around AI and businesses are totally
01:31different offerings.
01:33I have questions on all of the above, but you're essentially arguing that an AI agent becomes an
01:38economic actor in and of itself, right? The old adage that there are two sides to every
01:44transaction. But in sort of a meaningful way, what are we talking about needing to happen
01:49for a transaction in any medium to be between one agent on one side and an AI agent on the
01:55other side?
01:57Yeah. You know, I think when people talk about this, it seems really far off. And maybe part
02:03of that is, you know, the examples people give almost sounds incredibly far off where it's
02:10like, oh, I'll just have my AI agent plan my entire summer vacation, you know, and transfer.
02:15It's like, no, you love planning the summer vacation. You want to do that yourself. Whereas
02:19if you say, okay, you just give your AI, this is the recipe I'm making tonight, buy the things
02:26I don't have in the fridge already. Like that's the kind of delegation that, you know, people
02:30are much more comfortable with. And so we think people and AI services will kind of work their
02:35way up the trust curve. Where again, it's already the case that the research you're doing, the
02:39research you're doing, you do within an AI app, you're probably still confirming that directly.
02:43But it might get to the case where you're allowing the AI to make small decisions for you. And
02:48especially as it gets to know you. And so again, on the developer side, imagine you're building a
02:53website and you need a hosting provider or you need to buy a domain name for it. Today, it's the
02:58case
02:59that you go and you like find the domain name buying service yourself and you go fill out the
03:02form and everything like that. You just want a domain name. You can get it from multiple providers.
03:06You'd probably be okay with Cloud Code or Codex or Cursor, you know, whatever you're using,
03:11actually choosing how to buy a domain for you. And so it's already the case that people are doing a
03:16lot of interactive buying with AIs where they're improving the final step. We think people are already
03:22starting to have a little bit in a huge amount of what we announced today is those little decisions.
03:26People are delegating them to the AIs. Buying a domain name, again, is the example I would give.
03:31And then as they start to see that it works really well, people will just give them more and more
03:35rope.
03:37I've been reading a lot about actually the history between Stripe and Google. So I think people are
03:42quite accustomed to Stripe being integrated in Google Pay. The next step is AI mode in search
03:50and then Gemini itself. What does that change? Not just for Google, like it's, you know, it's placed
03:56in e-commerce, but also the habits, how people, you know, transact within the parameters of those tools.
04:04I mean, there's a few trends that are overlaid here, I would say. One is checking out is getting
04:11much more convenient. Apple Pay and Google Pay within their respective ecosystems have made that much
04:17more convenient. We also announced we have a product called Link, which we talked a lot about
04:21today. And that has gotten to considerable scale. We have 250 million people. You've probably used it
04:26around the web where, you know, you see a Stripe Link and it just remembers your details. It makes
04:30it more convenient. And so making those purchases, and again, as the volume of purchases goes up,
04:36has gotten significantly more convenient. But then again, like we're saying, the place where the
04:41purchases will actually happen is starting to change. Where previously, you know, you were doing
04:48some research, you find a product you want to buy. I was shopping for a travel adapter recently. It's
04:52like, okay, I want one with like an integrated power brick, blah, blah, blah, blah. Eventually,
04:56you get bounced out to the website. You're having to fill out all these details. You kind of want to
04:59be able to tell your AI app, okay, that's it. Yeah, sounds good. Please go buy it for me. And
05:04that's
05:04what's changing, where the transaction is getting pulled into the AI app, where it's already the case
05:08where many things you do. You want to be able to say if they are, yeah, sounds good. Please go
05:12do
05:12it. You know, again, if you're using a coding agent, it's frequently, you know, yes, go make
05:16no mistakes. And people want to do that in a shopping context. And so the purchase gets pulled
05:21up into there. And we're seeing huge demand from businesses. You know, we're working with all the
05:24biggest brands you can imagine, Best Buy, Kate Spade, all folks like this, who see this new kind of
05:30buying that's happening, this new kind of product discovery, and they want to be ahead of it.
05:34But before we move on, just quickly on Google, you know, Google has not historically led in
05:40payments. The partnership that you've been through today and the changes of integrating Stripe into
05:45AI mode and Gemini, does that change their fortunes in that market?
05:51I mean, I think Google has done a phenomenal job in commerce generally. And I mean, again,
05:57Google Pay is very large and successful within the Android ecosystem. And so I don't think that
06:02would be quite a fair characterization. I think they've done a phenomenal job enabling
06:06commerce. And again, this is a new tool in the toolbox for them, where now they're enabling
06:11within Gemini, these amazing agent of commerce experiences, which again, we think just has to
06:16be the future, because it's so much more convenient for consumers. That's never been the case in
06:20technology. When you get a way more convenient way of working for consumers, they pick the less
06:26convenient one. It's never happened.
06:28Has to be the future. I want to ask you about the now.
06:31John, how many days are we into the singularity?
06:35I should know the answer because we were talking about 119.
06:39And for the Bloomberg Tech audience, you did. I think you're right, based on my research,
06:45at least, and trying to track you and Patrick's comments. But help the Bloomberg Tech audience
06:50out, the basics of the singularity and why you and your brother, Patrick, why do you consider
06:56us to be in the singularity?
06:59Sure. So we've started half tongue-in-cheek, half not, talking about, yeah, being into the
07:05singularity. And, you know, we decided January 1st arbitrarily was day one. And so we're now
07:11119 days into it. And what we're getting at here is, I think for people in San Francisco,
07:17it really feels subjectively like something has changed. Something has changed since six
07:22months ago where the power of AI applications has really inflected. And Andre Carpathy actually
07:30had some good, you know, talking points on this that he tweeted where it's very dangerous if you
07:35kind of were playing with stuff six months ago or 12 months ago. You think you know what you're
07:39talking about, but your knowledge is actually woefully out of date because things have moved
07:42so fast. And in particular, what's really worked well is, I would say, firstly, the coding agents
07:47where you can have quad code or codex or something like that do this long-running task for you.
07:54And so you can have it. And it's not even just coding because they're like general productive
07:58agents. You can give it a research task. You know, maybe you're researching something for a piece
08:03and it can go off and search and buy the data it needs and everything like that. And so that's
08:08one
08:08thing. You're also seeing, you know, things like OpenClaw, this hit product that is a very novel
08:13way of interacting with an AI app that gives code execution and persistent memory to kind of consumer
08:21AI apps. And what all this line is up to, and again, why we're talking about the singularity is
08:26the coding models have gotten so good that they are now contributing in a significant way
08:34to the rate of improvement of AI itself. And so the AI is really good and it's getting better
08:41as a result of the quality of AI. And so kind of recursive self-improvements is this industry term
08:46here. But it feels like we have reached some point of recursive self-improvements over the past six
08:51months that, again, there's something different in the area in San Francisco.
08:55The economics and business model of software also is changing. I've been looking at like what Stripe has
09:00been going through today. And so you've introduced what you call streaming payments, but it's the idea
09:06that on a per token basis, you're charged in real time. That isn't that straightforward to understand,
09:14but token becomes the currency. What were you solving for with that?
09:21So the old way of anyone charging for anything was often per seat or like on some, you know,
09:30a Salesforce or a Workday, you know, was charged on a per seat basis. That could be, you know,
09:34Bloomberg, you know, I'm a happy Bloomberg subscriber. And, you know, that is your whatever
09:40per month. The Bloomberg terminal, I don't have one of those, but, you know, similarly, you know,
09:4525K a year or so. Everything is like a per person per month per year. That's the old way of
09:50doing
09:50things. With AI products, there's very significant inference costs that goes into them. And there's huge
09:56variability in how much of them you can use. And so you can have your coding agent do a quick
10:00job
10:00for you, or you can have it off burning GPU time on some, you know, huge task. And so all
10:06of the AI
10:07products that anyone has need to be charged on a per usage basis. It's just like no other way that
10:13makes sense. And so we work with all the top AI companies that you would think of, you know,
10:16the OpenAI and Tropic and Cursor and Grok and everyone like that. And they all, you know, once you get
10:22beyond some level of usage, it's on a per API request per token basis, like that is the new
10:27model of charging for any kind of AI powered product. It's tricky from an infrastructure
10:33point of view, because you have this like very complex metering challenge to do. And so we really
10:38have two products there. The first is we acquired a company called Metronome, which is the leader in
10:42usage-based billing. And so you can just whack events into that product. And then it will keep track
10:48of how much customers owe and everything like that. Really gnarly technical product. And so that's why
10:53that's used by all the top AI companies. But the other thing that's interesting is stablecoins actually
10:59now come full circle. And you have a really interesting opportunity for, you know, people have talked
11:04with crypto, the opportunity for micropayments for the longest time. And micropayments are always
11:09challenging because of the fatigue. Like, you don't want to pay 20 cents per article for Bloomberg, because
11:14just you don't want to be making that decision every time. You don't want to pay per song with Spotify,
11:18because your finger would always be hovering over the play button deciding whether it's, you know,
11:22whether that extra play really means that much to you. But that's not an issue with the agents. So
11:27for any kind of an agentic product, micropayments and micropayments via stablecoins makes total sense.
11:33And so we've actually incubated a blockchain called Tempo, which is specifically for the payments use
11:39case. And we've been doing a lot of work around that. And so as you think about streaming payments,
11:44it actually could be a really interesting application for crypto and stablecoins.
11:48Because it is very much about real time. But in this kind of shift from money to tokens,
11:54fraud or threat actors also are taking the same shift, right? Stripe is prepared for that.
12:02Okay, so this is really interesting. All of the AI companies we work with are seeing very interesting
12:09new threat vectors and fraud vectors, because again, they're selling...
12:13And that includes, by the way, John, just, I meant to say earlier, but open AI,
12:17anthropic, perplexity, meta, that you, they all work with Stripe.
12:21Correct. And it's just like, if you have a shop, if you have things of value on the shelf,
12:27no matter what kind of shop you're running, you're going to have shrinkage issues. You're
12:30going to need to think about theft and how to prevent it. Similarly, AI companies are selling
12:34something of value. And so as a result, they see ne'er-do-wells trying to come and, you know,
12:41abuse free trials or steal inference or anything like that. And so Stripe radar has become much more
12:47critical because of those attempts on, you know, people stealing. And there's kind of much more
12:54significant attempts on people stealing the products than we saw with kind of software
12:57before or something like that. So Stripe radar in the past month alone blocked, I think it was more
13:01than 3 million attempted, you know, thefts from, you know, attempted bad transactions from just the
13:09top AI companies we work with. It's really a big industry. It's one of the big things that AI companies
13:15have to think about. And we're using the power of the Stripe network, the fact that we see we have
13:20this kind of bird's eye view across the internet economy to help these companies prevent us.
13:25John, final question really is about the future of the company, Stripe. Bloomberg had reported in
13:30late February that Stripe was looking at acquiring a part of or all of PayPal. It's the first opportunity
13:37I've had at least to ask you to comment on that and the rationale behind it, if it's accurate.
13:44We never comment on M&A speculation. We haven't for the longest time. I won't start here,
13:49but I would love to, if we ever do anything in any acquisition, I'd love to chat to you about
13:53it.
13:54The way that I think about it, let's put the PayPal part to one side, is the economic
13:58infrastructure of AI, the relationship with Google through AI mode and Gemini.
14:04What's the consumer story for Stripe, right? If you think about PayPal, and I appreciate your answer
14:09just then, that they have 400 million consumer wallets. Stripe has grown incredibly as a private
14:17company, really integrated with some of the biggest enterprises and AI names you just went through.
14:22Is there a future where Stripe is more directly related to everyday people around the world?
14:30Yeah, you know, I think people see day-to-day the effects of Stripe, but not directly as a consumer
14:35brand. You know, you're not necessarily seeing, you know, a ton of instances of the Stripe name in
14:39your day-to-day, but we try to have that effect felt in less friction and everything being easier.
14:44And so people buy from Stripe-powered businesses day in, day out, and we are just looking to make it
14:49easier for them to do business with them. We're looking to help the businesses launch better products.
14:52I think people feel that as a second-order effect. You know, people see a little bit of Stripe
14:58consumer branding, like again with Link. You probably see it, you know, remembering your
15:02payment details around the web as you go. But again, we're just looking to lower friction.
15:06We're looking to grow the internet economy. And however we do that, you know, often that's at the
15:10infrastructure level as opposed to, you know, a big shiny kind of consumer app or something like that.
15:15And that's, that's what gets us excited.
Comments