The Information's TITV
00:15:01 · Akash Pasricha, Jyoti Mann
[00:15:01] S0: And I'm thinking about the competition here. Optimus, obviously, from Tesla is sort of the loudest humanoid player in all this. We were just talking to Aaron Tilley about Apple's robotics ambitions. How far do you think Meta wants to go with this? Could we see an acquisition coming down the line here to really ramp up their efforts to move into this space? Is this still something that is very R and D focused, something that's happening in the background? What's the level of priority here, you think? [00:15:35] S2: Yeah, I mean, wouldn't rule out an acquisition, but I think with Meta Strategy it's more than just simply buying a robotics company. I mean, they already acquired Assured Robotics Intelligence earlier this year which builds models to train robots. So it's possible that they are just wanting to build the operating system and sell that and then potentially partner with another company in The US to build the hardware. But you have to remember that Meta has enormous resources. Not only do they have the AI talent, they've got experience building consumer platforms, so really they just need the hardware expertise and the real world robotics data to make these robots work. So even if Meta did develop that intelligence in house, it's still gonna have to sort of source a lot of these components from somewhere, and that's where the supply chain becomes really important and the FCC rules could kind of restrict their ability to do that quickly. So China does seem to have a significant advantage, but I could see Meta making acquisitions for the hardware side.
00:15:35 · Jyoti Mann, Akash Pasricha, Stephanie Palazzolo
[00:15:35] S2: Yeah, I mean, wouldn't rule out an acquisition, but I think with Meta Strategy it's more than just simply buying a robotics company. I mean, they already acquired Assured Robotics Intelligence earlier this year which builds models to train robots. So it's possible that they are just wanting to build the operating system and sell that and then potentially partner with another company in The US to build the hardware. But you have to remember that Meta has enormous resources. Not only do they have the AI talent, they've got experience building consumer platforms, so really they just need the hardware expertise and the real world robotics data to make these robots work. So even if Meta did develop that intelligence in house, it's still gonna have to sort of source a lot of these components from somewhere, and that's where the supply chain becomes really important and the FCC rules could kind of restrict their ability to do that quickly. So China does seem to have a significant advantage, but I could see Meta making acquisitions for the hardware side. [00:16:37] S0: Great. Okay. Well, Jyoti, want to thank you for coming on. That is Jyoti Mann, our Meta reporter, here at The Information. As AI Labs look to convince their customers to renew contracts and even upsell them, my colleague Stephanie Palazzolo has reporting that Anthropic and OpenAI have seen some resistance from businesses. I want to bring her on to share more about what she is hearing. Stephanie, welcome back to the show. It's great to have you here. [00:17:04] S3: Thanks. Great to be back.
00:17:04 · Stephanie Palazzolo, Akash Pasricha
[00:17:04] S3: Thanks. Great to be back. [00:17:05] S0: Okay. So tell me, Steph, why are customers hesitant to renew their OpenAI and Anthropic contracts?
00:17:13 · Stephanie Palazzolo
[00:17:13] S3: So there are two dynamics going on here, which, me and my colleagues have kind of heard about just from chatting with a lot of these really large, customers. But essentially the first one is that, customers, you know, we, we kind of live in this age right now where it feels like every other week there are new models coming out that are leapfrogging each other when it comes to really pushing the front, the frontier of performance. And so customers want to have that flexibility to be able to switch model providers whenever they want, just depending on whatever the best model is that's out there today. Especially because now switching models has actually become easier over time. We recently wrote a story where Replit was basically able to switch its entire product from Anthropic to OpenAI models with just a couple lines of code, you know, that took like a couple hours to push through its entire product. So that's the first reason. The second reason is actually something that the labs are doing themselves, which might be hurting, usage of their most advanced models, which is that, companies like Anthropic and OpenAI are releasing cheaper, faster, like quote unquote good enough models. So for OpenAI, these are models like Terra and Luna, and then Anthropic has its, Haiku model. And so especially as the models are getting better and better, these cheaper models are becoming good enough that they can, you know, do most things that customers want them to do. So it becomes very easy for customers to switch, you know, use cases from the most expensive advanced models to cheaper ones that the model developers themselves are offering as, you know, more cost... More... Or more cost effective alternatives.
00:18:55 · Akash Pasricha, Stephanie Palazzolo
[00:18:55] S0: So, like, are the model companies completely cooked then? Like, what's, what's, what are you predicting here? [00:19:02] S3: Well, it's not clear. I would say this is definitely not a death sentence for them. I think there has been this theory throughout the AI boom that actually as models get cheaper, people want to use more of them. Because the models are getting better, people are wanting to use more of them. Actually, maybe overall spending will increase over time. But I will say though that, you know, a lot of companies that I talk to, they are just, just very nervous about, you know, every year they basically have to commit to spending a certain amount with OpenAI and Anthropic. So even if they know that their overall AI spending is going to increase over the next year, they don't know exactly that it's going to increase more on OpenAI or more on Anthropic specifically, or whether that new spending will come from open source or from a third model provider, for instance. So it's just been very difficult, I think, these customers to be able to commit right now to spending, you know, whatever 25 or 50% more on OpenAI for instance, over the next year. That's just been a really tough uphill battle, I think, for the salespeople at the model developers to have to go up against. [00:20:14] S0: Now, on the other hand, you did say that companies operating routers like OpenRouter, they've had a very easy job of getting people to renew their contract sounds like.
00:20:14 · Akash Pasricha, Stephanie Palazzolo
[00:20:14] S0: Now, on the other hand, you did say that companies operating routers like OpenRouter, they've had a very easy job of getting people to renew their contract sounds like. [00:20:26] S3: Yeah. So we've really seen revenue at some of these other companies like OpenRouter explode over the past couple months, or even companies like Snowflake, where essentially they kind of act as like a central hub for a bunch of different models. So, you know, even with like Amazon Web Services, for instance, they have a product where you can go to that and then, spend across different models from both OpenAI and Anthropic. And so like what these kind of platforms offer is, you know, more flexibility and the ability to switch between different model providers very easily. And so this might push more customers towards these kind of like independent, like third party platforms versus buying directly from a model developer where they need to commit upfront to spending, you know, X, Y, Z amount over the next year on their models only.