The Meta Compute Debate | Diet TBPN

0:01What are the best practices of the AI?

0:02John Coogan:Meta is selling computers. They're getting out of the computing business. They said we don't need computers anymore to do what we need to do. We don't need them. We're going to be selling them. Meta Platforms is developing plans for a cloud infrastructure business to sell access to AI computing power and models competing with industry leaders like AWS and GCP. The company is considering selling access to various AI models hosted on its existing AI infrastructure, as well as raw computing capacity. As part of its Meta Compute Initiative, Meta plans to generate revenue from excessive computing power could help return its investment in AI infrastructure, which includes hundreds of billions of dollars spent on data centers and expensive chips.

0:43John Coogan:And so lots of reactions to this. The NeoCloud market is selling off. Oddly enough, Meta has a bunch of NeoCloud contracts. Some of those companies are selling off because now they're a buyer and also a competitor. Lots of different takes about, you know, Meta finding its footing, finding something that justifies the massive CapEx. Of course, Meta is... It's deeply confusing, John. Yeah? Is it? I mean, the whole thing. The whole thing. What's confusing about it? I think that, I think it is practical what they're doing. Yeah. But as somebody that, you know, I would say overall has been a big, you know, cheerleader for Meta, I think it's truly the best.

1:22In my view, it is a perfect business. It doesn't give you a lot of confidence in, like, the strategy overall. They're signing these NeoCloud deals worth tens of billions of dollars. They're built, you know, spending hundreds of billions of dollars. And yeah, they can make the argument that these type of like doing any type of NeoCloud deals themselves is just good business. It's just like it's just the best way to get ROI today. It doesn't give you a lot of confidence that there's near term products on the horizon for Meta that are going to be able to utilize that capacity themselves, which is clearly been their strategy.

2:02Mark and the team have never said we want to be in the cloud business. They've talked about the possibility of it, but the stated goal of MSL is personal superintelligence.

2:17John Coogan:Which I was a fan of, and I think you were a huge fan of. You were like a manis on your phone, going around your social networks. That's my biggest bull case for all of this. There are so many different applications that I can imagine being a daily driver of in the meta family of apps. Oddly, none of that has really been even tried, in my opinion. It feels like a little bit early to call it. Yeah, all we've really seen so far is Muse Spark. Good on benchmarks, like decent, you know. Again, not anything that anyone should really get that excited about. As an API provider. Exactly. They did announce that they were going to release it via an API.

3:00John Coogan:I don't think they have. They might still. It's a good model, sir, but I don't think it will have a lot of demand. Yeah. And then we've seen MetaVibes, which was a mid-journey wrapper. Even if MewSpark is not on the super giga frontier, can it be good enough to get some work done inside MetaFamily of apps? Like, it should. I would imagine, yes. But they just haven't found that killer feature. There are plenty of applications that are AI powered. There are plenty of models out there that have found their footing without being on the super intelligence path or on that particular curve. And it's interesting because yesterday we were talking about the story where Google had been telling Meta like, hey, we don't have the capacity for you.

3:43And here Meta is with plenty of capacity themselves. I don't think we can read too much into this because it's just one article from Bloomberg. I think it will matter a lot who the potential buyers of compute are going to be. If there's a number of companies that I think the market would get excited about. But if they're actually just going and trying to compete as.

4:06John Coogan:Yeah. It's weird that it came as a leak around like a plan to sell compute as opposed to just what SpaceX did where it was just like, boom, huge contract with Anthropic. Lots of excitement going into the IPO. That was such a perfectly massaged story as SpaceX entered the public markets. That would have been great if they just said, hey, we have a frontier lab that's paying us a billion dollars a month now. And it's going to show up in earnings next quarter. Get ready. But the stock market loved it. The stock is way up. And I don't know if it's way up because they see it as a huge growth area for Meta.

4:41John Coogan:Now, is it that crazy that a hyperscaler could have a cloud platform? Where's the ROI going to come from for this hundreds of billions of dollars to spend? And up until now, there's been no obvious place that it's going to come from. Right now, the Manus deal, that's being unwound. They have the deal with Mid Journey and Vibes, that's unclear. It seems very obvious that they're going to be able to integrate AI into their glasses over time. But the glasses that have product market fit today are more of just like the, I think the product market fit is really with the camera, not the intelligence combined with a pair of glasses.

5:21John Coogan:Yeah, which is also surprising that we haven't seen a diffusion model, an image model. Instagram in general is pretty, it's remarkably slop free, at least my feed is. I don't see a lot of viral slop images, but you can imagine AI powered features, background replacement, a lot of that stuff being AI enhanced and people receiving that very positively and enjoying that. I was using Adobe products recently and like in Photoshop, they have an integration with Gemini, Nano Banana, and you can plug in all the different image models and you can actually use the tools in very interesting ways. I keep going back to that idea of like the personal super intelligence?

5:58John Coogan:What do I actually want? I ran into an interesting conundrum the other day because I've been sort of disappointed by the lack of AI in meta apps, which I know I'm like the only person that feels that way because the general vibe on Instagram is like extremely anti AI. Meta does have granular data about every reel I post. Some do better than others. I went to meta AI in the Instagram app and I asked, what should I do more of in order to grow my account? This would be useful, sort of a high-powered analytics tool, personalized analytics tools for what I do. But I got this very generic LLM response.

6:38John Coogan:We might be able to pull it up. I think it's in, I shared the image in the timeline. Let me see if I can pull it up here. It said, Buffer recommends focusing on sustainable organic growth instead of quick hacks like follow trains. post reels consistently since they get 36 % more reach than carousels in 125. It's so funny to be referencing a blog post from a social media management SaaS company. Like you would think that if you had, I mean, and clearly they're not focused on this use case, but you would think that there would be an opportunity to give creators personal and super intelligence to just be better at creating on the platform.

7:15John Coogan:Yeah, so use Instagram analytics to see which content converts viewers into followers and double down on it. Like, that's what I asked you to do. I said, what can I do better than my following? Like, you have all the data about what does well. I don't want to go and look at this reel got 5 ,000 views. That one got 50 ,000 views. What's the difference here? I want you to do that. Collaborate with micro creators in your niche for authentic cross promotion rather than paid ads. Engage with responding to comments and testing out trust. It also said to optimize your profile with clear keywords in the bio and maintain a consistent visual brand.

7:49John Coogan:And it's like a lot of that I'm already doing. A lot of that is old. And so I'm just like sort of disappointed that say what you want about Muse Spark and its benchmarks or whatever. Like clearly if they wire that model up to the user's data, they should be able to integrate appropriately. And they just haven't productized it properly. And maybe if I get Manus and I get an API integration, like I could get there. But like it should just live in the Instagram search box, I imagine, since they already have an LLM there. It's just a legacy one. On the product side, it's just not enough to get to, okay, there's crazy demand for AI within the app.

8:25Did you ask it if the shareholder value is three eggs, is the goose value?

8:30John Coogan:That just sounds like a jumble of words. Is that even a correct sentence? I don't know. Sounds correct to me. I mean, honestly, like the best place, it's very ironic, but the best place for Instagram growth hacks these days is just Adam Masseri's front-facing videos. He takes to Instagram. He uses Instagram very well. He goes direct, posts reels about all sorts of things. So today I got surfaced one. If you post something and it doesn't do well, should you delete it and then post it again? And he says no, because the algorithm will give the same result. And your followers who saw the first one didn't like it.

9:03John Coogan:They're going to see it and like it even less the second time they see it. He just answers a bunch of common questions. He does Q &As. And it's actually the best way to communicate and get insights into how the Instagram platform works. But it's not personalized. Like he's just giving generic one size fits all advice. I want the Adam Masseri brain enhanced with Meta's AI tailored to my account. That's what I want. And like a social media co-pilot. But maybe I'm in the minority here. I don't know. Maybe I'm the only one that would want this. I feel like tools for creators would be a great way to launch this.

9:37John Coogan:Even like a small user, everyone almost always cares about, I'd love to get a couple more views on whatever, even if they're just using it in primarily a consumption tool. And then there's also agentic shopping, which we asked Mark Zuckerberg about at MetaConnect last year. This idea that, you know, the MetaRayBan displays, they have the HUD, they have AI. You should be able to look at a pair of shoes and say, order me those. And he sort of like gestured towards that being one potential possible future. But it's crazy to me that we haven't even seen them really try to remove at least one click from the shopping experience, store some more of your data, shorten the funnel, increase conversion rates.

10:20John Coogan:That's good for brands. That's good for companies that advertise on Meta. It's good for Meta. It's good for users. And it just doesn't feel like that's where the energy has been in terms of productizing AI within the family of apps. And so it feels like e-commerce will see agentic shopping happening. We're getting closer. Computer use is getting better. APIs are there, MCP servers, and Shopify has a bunch of tools for this stuff. But it's weird that Meta hasn't even been experimenting there, and we haven't seen, like, oh, yeah, like they launched a thing where if you see an ad, you can click a button, and the agent will try and go check out with you and then just confirm the details within the Meta app, and you don't actually have to open up the Safari window.

11:03John Coogan:I don't know. Maybe that works. At least we're on the experiment. Maybe they have. I don't know. Maybe I just missed it. Yeah, this feels like a wind down of the super intelligence ambitions that Zuck was gesturing towards last year. But I don't know. Maybe this is in the path. Maybe this is just a temporary thing. Yeah, and if you look at the SpaceX deals with Google and Anthropic, they weren't like five-year deals, right? They were shorter-term opportunities for both sides to get out. and I think that Meta could easily do something like that where they could make, again, this is like the practical decision and say like, hey, we actually do have way more capacity than we need.

11:41We plan on being able to utilize it fully over time, but we need to get our products sort of ramped up. And so in the meantime, why not sell that capacity and signal to the market that we're not entirely irrational?

11:53John Coogan:Yeah, yeah. I think Meta can actually wait until like the features that they need to implant are extremely obvious. Apple waited a long time to actually implement these things, even though everyone was basically begging. I don't know if people are begging for AI features in Instagram, and maybe that'll happen in the future, and they can basically just wait while, like, oh, we're not really sure how to implement this stuff. Yeah, you can put it in the search bar or whatever, but it doesn't improve the experience that much. But maybe in a year or two, there's going to be some feature that, like, wow, everyone really wants this in Instagram.

12:21By then, they'll have all the compute necessary. They can, you know, 90 days before they implement it, they can get out of the lease or whatever.

12:29John Coogan:Yeah, yeah. Yeah, it's interesting. I feel like they just need to do more product experimentation. Like the core Instagram team, they launched the, what's it called, glimpses or something? It's like a shorter, even shorter version of stories, like lower. It like sits in like the little sidebar of the chat. Like they are launching different features across the family of apps, but that culture, like it feels like they have a research organization, but they don't have an AI product organization that's actively productizing things as quickly. I mean, MetaVibes, even though it was like this white label of MidJourney, at least it was launched fast.

13:04John Coogan:At least they got the feedback. I don't think they've ever been the company that really innovates on product, right? There's going to be some feature that everyone's like, wow, this is really great for AI. And then they can then just integrate it into all their apps. They have all the capacity to serve it by that point. Yeah, I don't know. I don't know. I don't know where it goes. Amit is investing. Over there shares his perspective, two perspectives on the Meta news. He says, bearish. The bearish take is if Meta has excess compute that they are willing to sell via a new cloud business, doesn't that mean we aren't compute constrained?

13:38John Coogan:Isn't this really bad for neoclouds? Why would Meta give a deal to CoreWeaver Iron if they just sell the compute themselves? Furthermore, wouldn't they cut CapEx because idle compute as the basis for a new business means they don't need as much compute as they bought, which means CapEx should come down. That would be bearish for all semis. The bullish take is if Meta is building a cloud business, even if they are using idle compute, which means they aren't compute constrained, they might end up spending more on CapEx to compete with GCP, AWS, and Azure. Like if they realize that selling computing services on top of a Meta cloud is better than just ads, then wouldn't they end up having to spend in the same way that Google, Microsoft, Amazon do in order to build out a full cloud business.

14:19John Coogan:They do have a lot of capabilities in terms of spinning up data centers quickly, maybe not quite as quickly as SpaceX and AWS, but they're certainly near the frontier of that capability in terms of putting up GPUs in tents. So more CapEx would be good for Semi. So what do people think? Where do we land on this? The market is certainly reacting positively to Meta and negatively to the NeoCloud because there's a new competitor in town. Zephyr and the Cetrini team are going pretty hard. They say LMAO. Zuck finally takes the L. Okay, so if this is a metaverse style side quest and the super intelligence meta trained models are the VR of this cycle for meta's attempt at a new business creation and the meta trained models get sort of mothballed in the way that the meta quest and the VR strategy got mothballed, what is the meta ray-bans of msl like what would what would remain because when they wind down one of these projects sometimes you get a meta ray-bans which is pretty good business and growing and cool and like there's development there but it's a much more narrow focused and i would say image model but they they seem to wouldn't it just be this like neocloud business maybe yeah that's what's going to stay on that's like profitable on his net ray-bans though i want a consumer product it's a consumer company.

15:46Yeah. And it's interesting because you can, you can make the case that Meta could build an amazing inference business, right? They already serve millions of businesses globally. They could make a pretty compelling case for how like, Hey, we help people acquire customers and we help people deliver products and services over here. Yeah.

16:10John Coogan:We've talked about that before. I don't know that I buy that, that the fact that they have like a, you know, every single mobile gaming company and D2C e-commerce business on is actually flows over to, well, now get your tokens from us. Yeah. I just, I just, I just think that like people have it in their head, like, okay, meta is a consumer business, but they have incredibly massive sales teams, account management teams. They know how to get in front of customers. They've got to the point where, you know, again, And they're not operating a commodity business, right? Yeah. Having their, you know, social networks.

16:46Cloud is closer to one. But anyways, I'm interested to see. I imagine they'll have to come out with their own kind of news around this pretty quickly. So that they're not sitting in limbo with just one kind of rumored article floating out there. And people are speculating.

17:05John Coogan:The Meta Ray-Ban display is a good example of like, or Meta Ray-Bans. A good example of like there's probably a piece of consumer hardware that's AI enabled, like the ring or something, or just getting really good at voice models or just getting really good at image models. And if they're a little bit more narrow and constrained, they could probably completely dominate that. But trying to do super intelligence and coding agents, and it's a little scattered potentially. Jay Yoon says, we are still massively short compute meta and XAI are selling compute because there's no inference demand for their models.

17:39John Coogan:it's a compute allocation problem too much compute in the hands of players with no internal use for it not a compute surplus problem more meta news what is that apparently according to bobby allen over at npr oh yeah consider buying calci before it's developing its own prediction market app that is uh sort of a classic meta playbook this sort of puts to bed your theory that uh they might They just making a clout-based prediction markets where you compete for your ability to see the future. I'm the old Manifold, is that right? Yeah. So it's like a pretty big platform. Yes, but I was saying there's a chance right now based on the reporting that it could be the Manifold strategy or it could be the Polymarket Kalshi strategy.

18:31John Coogan:And the fact that they didn't try and acquire Manifold, they tried to acquire Kalshi sort of signals like, hey, they're probably going the financially incentivized route, which I think fits with your thesis that, you know, it's in consumer, it's profitable and growing very fast. And also Tarek from Kalshi was taking shots at Instagram saying it's brain rot, saying that like every minute you spend on Kalshi is a minute that you're not spending brain rotting on Instagram, which is like, okay. A lot of people would say that these are like equivalent or maybe one is worse than the other. Is the potential profit pool worth the risk of all the attention you're going to get from lawmakers globally by integrating like betting into the product that is already under attack on like a million different fronts, right?

19:19It feels like -

19:20John Coogan:With the movie coming out and stuff, it's like you're jumping straight to the front of the It feels like you have a golden goose, right? And the goose is getting valued. The goose is getting valued. I'm going to just keep going back. I love these slides. The goose is valued. It's producing golden eggs, and you see another golden egg, but it's almost like a poison golden egg. And if you bring it over to the farm, it might kill the goose. It might kill your main goose. Potentially. And so it feels risky. Google AI overviews decreased outbound organic clicks by 40%. Eric Sufert is sharing. A new paper by researchers at Carnegie Mellon in the Indian School of Business finds that AI overviews were triggered in roughly 41 % of observed Google searches, and when triggered, reduced outbound organic clicks by about 40%.

20:05John Coogan:That seems like one to one. The presence of AI overviews increased the likelihood of a zero-click search by roughly 35%. Everyone is going Google zero at this point. And the founder, Josh Marshall, of TalkingPoint's memo wrote about Google AI oligarchy and the end of the open web. It's an interesting read, but we'll go through it another time. December 2010. Demis was scraping together a couple million bucks for a small company called DeepMind. This is for a founder that just, like, you know, had their – started their career in the last few years. This is, like, inconceivable. They're like, wait, you mean two on 50, right?

20:53It's like, no. sold half the company at 5 million posts as one of the most elite technologists in the entire world.

21:03John Coogan:If it stayed independent this whole time. I don't know. Maybe there'd be another path or something. You're saying they sold too early? It seemed like they sold too early. I don't know. Hard to say. Hard to say. Paper hands? Paper hands, Pete? No, no, no, no. I think all the VCs didn't want to sell. Especially because the whole Google thing. That was like the whole story was don't sell to Google. Google's like the bad one, which is funny because Google's been very responsible and, you know, great, great company. I was playing around with VO3 or VO4. Where are we in the VO models? So good and yet still not indistinguishable from reality.

21:46John Coogan:We're so, so close and yet so far if you want to generate a video these days. It is cool you can do it on your phone, though. I like it. And, Jordy, people were asking you to do a little bit of a spin for everyone. Show off the new merch you got there. I'll tell everyone about Codex on the arms. Codex is a powerful workspace for getting work done with AI agents, whether you're writing code, analyzing data, creating content, or automating business workflows, or just trying to look good. Codex. We just got our first shipment of, I think, around 200 of one of our new merch products. Oh, okay. I know the one you're talking about.