00:00Joining us right now is Ed Ludlow. He is at that conference and he's joined right now by the COO of OpenAI, Brad Lightcap. Ed?
00:09Yeah, and the headline is the ability to use third-party apps within ChatGPT.
00:14And the profound impact on the market was simply those name-checked saw their stocks move in a significant way.
00:20Delighted to speak once again with Brad Lightcap.
00:23It's as simple as that, really. You know, the ability to be within ChatGPT and access Spotify or Figma as two examples.
00:32When you were discussing the idea of that and API access with those partners, was it a tough negotiation to get them on board with the idea?
00:42No, I mean, we've seen enthusiasm for this from the beginning.
00:45You remember way back when we lost something called plugins back in the early days of ChatGPT.
00:50This was one of our first attempts to start to build an ecosystem around ChatGPT so that ChatGPT can start to engage with and interact with the applications that are important to you in your personal life and at work.
01:01And now we really have a much richer surface through MCP and other protocols to be able to bring applications into ChatGPT and for really to allow you to engage with ChatGPT in the kind of work around work, right?
01:13It's the contextual aspect of I'm doing X or I need Y. I'm on a road trip and I want to know what playlists would go well with this in the context of my broader trip planning that allows you now to kind of use ChatGPT to solve that higher level task and then also integrate apps contextually to solve those specific problems that are useful.
01:31So in a place where ChatGPT has become more of an operating system, whether that was your ambition or not, is that where you want to take it, to be an OS and a developer-driven platform?
01:44It's almost like an app store, you know, on paper based on what you announced this afternoon.
01:49Well, we've always thought of ChatGPT as like a super assistant.
01:52We never set out to build a chat bot.
01:54We always wanted to build something that was really true to you and what your preferences are, what your goals are, that could actually help you achieve more.
02:02And so I think part of that is ChatGPT having an appreciation and understanding of the applications in your life that are important.
02:09And I think enabling that kind of connectivity and interoperability makes ChatGPT richer, also enables a lot of pass-through for people to be able to engage with apps they love it as well as new apps.
02:19The pass-through bit is interesting.
02:21Were there concerns with some of those technology companies you're partnering with that it would take traffic away from their own domains?
02:28No, I think mostly people are really focused on building into new interfaces, right?
02:32This is just like mobile in some sense, where you have a new interface, you have a new form factor.
02:37People are going to want to use mobile form factors on the go.
02:41And apps like Spotify in some ways exist almost because they really nail mobile.
02:45And so we think actually there's an opportunity for builders to create entirely new applications that are even native to ChatGPT.
02:52And, of course, for services you love to be able to benefit there, too.
02:55Is there a revenue sharing agreement with those third parties whose apps are accessible through ChatGPT?
03:01So we're going to figure out the economics of this over time.
03:04You know, we're brand new here.
03:06Plugins was the first version of this, and that was even an experiment.
03:09And so, like everything at OpenAI, we take this very experimental mindset to making sure we get it right.
03:14But the idea is we do want to build something that's useful for developers.
03:18And, of course, there's going to have to be some exchange in there of economics and value, and we'll have to figure out how to get that right.
03:25You have hit 800 million active weekly users you announced on stage.
03:29Actually, Greg Brockman told me that at 8 o'clock this morning, and maybe we missed it.
03:34But it's a significant milestone.
03:36All the time I'm asked by all kinds of people, do we have any sense of within that 800 million, how many are base-level free users,
03:44and how many are premium-level paid subscribers?
03:47Yeah, we have a very healthy, you know, funnel of people that choose to pay for ChatGPT.
03:54You know, it's surpassed where my expectations, frankly, were.
03:57People have this kind of conception that consumers tend to not pay for software.
04:01And, you know, similar even to what I was saying before around how do you co-develop the product alongside the business model,
04:08ChatGPT is a great example of that, where the subscription model, I think, has been really a testament to how valuable it is for more users
04:15than I think we expected to be willing to pay for it.
04:18So we don't, I think, disclose the exact number, but it's a healthy amount and more every day.
04:23OpenAI, in the beginning, went after the consumer for users aggressively.
04:29You are now very focused on the enterprise business.
04:32What is the strategy for that, and how do you prioritize your enterprise business?
04:36Yeah, I'm glad you asked about that.
04:38So, really, today's announcements actually, I think, target what are an important set of use cases for the enterprise,
04:44things we've been hearing enterprises ask us about now for some time.
04:48So, specifically, one is we now have an ability for enterprises to build agents in a much more visual, much more intuitive way.
04:56You've heard us say 2025 has been the year of agents.
04:59We think that's true.
05:00Codex has been a great example for us of that.
05:03Our coding agent now available through an API also.
05:07The lead time to make software is a lot shorter.
05:09It's gotten a lot shorter.
05:10I think you saw today we demoed, live demoed, I think, three or four different things that we've built in real time.
05:15We expect that to continue to be the trend.
05:17Things like Agent Builder allow enterprises to be able to build agentic experiences, powerful agentic experiences, on the go, iteratively,
05:26and connected into the tools and sources of information that matter for the business.
05:30The data point that jumped out at me is your API is handling more than 6 billion tokens per minute,
05:35and that helps explain why the AMD deal, which is focused on inference.
05:41You are involved in all of these domains of the company.
05:43I've already asked Greg, but I've got to ask you, how are you going to finance yet another infrastructure project?
05:50Is there going to be some debt here specifically for the AMD capacity, and how do you move quickly to get it online?
05:57Yeah, well, the high-level thing is we are tremendously compute-constrained.
06:02It feels like we're in this kind of recurring theme of being compute-constrained, and I think the reason for that is the answer to the question you asked, which is demand.
06:09Right.
06:09We see there are multiples of demand that are latent and untapped from what we have today, and even today, obviously, by any standard, demand and revenue growth has been torrid in its pace.
06:21And so, really, we have to invest ahead of that, and I think that's going to be the rate limiter for us to be able to go capture demand, whether it's consumer or enterprise,
06:29and for us to be able to build new models, parallelize more experiences, more product experiences, and then enable users specifically to be able to use those products more actively in their daily life, at work, and at home.
06:40So, you know, even things like Sora, the app we just launched, we wish we could invite more people onto it now, but we just need more compute.
06:48So, the AMD deal, we're excited about being, you know, directionally a way for us to do that.
06:55So, I've got to ask about the report that OpenAI closed secondary or the ability for employees to sell shares at a $500 billion valuation.
07:04I already asked you this question, but what is the metric we're supposed to judge your success by?
07:09The $500 billion valuation, the 6 billion tokens per minute.
07:13To you, Brad, what is it?
07:15For me, it's actually kind of a metric that we talked about is tokens.
07:21It's, you mentioned, 6 billion tokens per minute on our API.
07:27That is the purest, for me, the kind of essence of utility is that consumption metric.
07:32And so, we've actively tracked that metric to see how people's consumption of AI is growing over time.
07:38And you see this happen in amazing ways.
07:40So, things like Codex, for example, we've seen grow 10x since August, purely in consumption of tokens around coding.
07:47And you start to see that same pattern emerge across multiple lanes of use and across multiple areas of work.
07:53And that's the metric I look at because if that number is going up, it means people are using us for more things.
Be the first to comment