← Big Technology Podcast
Big Technology Podcast
Zuckerberg’s Disappointment, OpenAI’s Equity Gamble, Alex Karp’s Rally Cry
00:59:12
Loading summary
Transcript112 lines
- [00:00]
A
Mark Zuckerberg says AI development is moving slower than expected as Meta thinks about selling off its excess compute. OpenAI wants to give its equity to the White House, and Palantir CEO Alex Karp takes on the foundational labs. That's coming up on a Big Technology Podcast Friday edition right after this. In the face of ongoing disruption and opportunity, TMT leaders need to deliver tangible results, not just ideas. When pace and performance matter most, PwC combines market insights and deep sector experience with AI, cloud and emerging tech to accelerate your transformation and drive measurable ROI. From strategy to execution, PwC can help you anticipate what's next, outpace disruption and compete. For more information, visit pwc.com Insurance isn't
- [00:50]
B
one size fits all, and shopping for it shouldn't feel like squeezing into something that just doesn't fit. That's why drivers have enjoyed Progressive's name your price tool for years. With the name your price tool, you tell them what you want to pay and they show you options that fit your budget enough. Hunting for discounts, trying to calculate rates, and tinkering with coverages. Maybe you're picking out your very first policy, or maybe you're just looking for something that works better for you and your family. Either way, they make it simple to see your no guesswork, no surprises. Ready to see how easy and fun shopping for car insurance can be? Visit progressive.com and give the name your price tool a try. Take the stress out of shopping and find coverage that fits your life on your terms. Progressive Casualty Insurance Company and Affiliates Price and Coverage Match limited by State Law
- [01:41]
A
welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool headed and nuanced format. We have a great show for you today. We're going to talk all about Meta, deciding that maybe it's time to sell off some of that very precious compute. As Mark Zuckerberg says, AI agent development is going slower than expected. We're going to talk about what that means not just for Meta, but for the broader AI industry. We'll also talk about OpenAI potentially giving some of its equity to the White House, whether that's advisable or not. And of course Alex Karp, the CEO of Palantir, coming out and declaring effectively declaring war on the foundational labs is really what it is. Joining us as always on Friday to do it is Ranjan Roy of Margins. Ranjan, good to see you.
- [02:20]
C
Great to see you Alex. I'm currently in London in a hotel room with mediocre wifi.
- [02:26]
A
Yeah everybody. The fact that we're doing a show right now is somewhat miraculous. I just want to say thanks to Ranjan for sticking with us. It's pretty late London time, so modern
- [02:36]
C
technology, that's what we do. That's what we do.
- [02:39]
A
So our top story this week is what's going on at Meta or what isn't going on at Meta, and that is impressive AI development. And that was something that Mark Zuckerberg in an internal town hall admitted, effectively, according to Reuters. Here's the headline. Zuckerberg says AI agent development is going slower than expected. Meta Chief executive Mark Zuckerberg told an internal town hall on Thursday that AI agent development over the last four months has not accelerated in the way we expected, according to a recording. Zuckerberg added that the company re the company's reorganization that included major job cuts was not as clean as it could have been, and that the company's bets on a new structure haven't come to fruition yet. Meta is projected to spend as much as 145 billion on AI infrastructure this year, a significant portion of Big tech's more than $700 billion outlay on the technology. And I might as well just share the second story here, which is that Meta is now considering leasing some of its excess compute, same way SpaceX did, because it doesn't seem to have the products that can make use of it. Ranjan, is this an unsettling sign that companies that are not OpenAI and anthropic don't know what to do with their compute and are trying to sell it? I mean, Meta, SpaceX, obviously, Google selling its compute, Microsoft selling its compute, Amazon selling its compute. For all this talk about how compute is such a scarce resources, it's a resource. It seems like there's two companies that need it and a lot of people supplying it.
- [04:06]
C
Let's separate out the two different parts of it selling excess compute. But first of all, what do you think it means that AI agent development over the last four months has not accelerated in the way we expected? Because again, remember, Meta was the kind of poster child of token maxing. Everything we saw over the last few months was really around how AI is going to change everything within the organization. We obviously can get into the mouse tracking software and all of that, but what are the AI agents that have not developed? What do you think he's talking about?
- [04:44]
A
I mean, for the life of me, I don't know. I mean, it's almost an admission that Meta, despite the compute and the engineers, just hasn't been able to put the pieces Together. And you know, I think it also goes to basically our model versus product versus harness discussion. Right? It's like Meta. Andrew Bosworth, the CTO of the company, is going to tell me when we, when we air this interview I had with him on Wednesday, this upcoming Wednesday. They're already licensing models from other companies. So it's not just Meta models. So it's not a Meta model problem. They also have the best model builders in the world, or some of the best that they brought in for superintelligence labs. They also have no lack of compute. So it may just be that the knowledge of how to do this is sort of sitting within the foundational labs and it's. They're struggling and it's the, the other, the competitors are struggling to pick it up. Which is bizarre to me because the guy that built cloud code is a former Medical Meta product manager. Right. Boris Journey. What do you make of it?
- [05:57]
C
What's so fascinating to me about this is Meta is the greatest organization in terms of productizing things that are already out in the world, whether it's through acquisition, whether it's through copying, whether through, you know, like they know how to build a product, whether it's internal, whether it's external. So I think that's the most fascinating part for me as well. And I still, I mean, I think one thing to note, note again, working in the area of enterprise AI, but specifically outside of software engineering and in knowledge work, myself at Rider, like that distinction, I think is really important because what was more surprising about this kind of statement to me is everything that I've read and that's been reported, the way Meta still looks at AI agent development, it's in the software engineering realm and maybe it's starting to kind of like bleed over into the other elements of the overall kind of organization. But it's still, I think this is a really, really important moment because what happened in software development, I've said this a lot on the show, what we saw from, call it November to April in terms of that like insane logarithmic growth in terms of software development and coding, everyone is realizing is not the same straight line path with other types of work. And again, what exactly is AI agent development? We don't know. But I think it's important because if Meta is not figuring it out themselves, separate from foundation models and whatever else, this is a powerhouse company that is able to actually just organizationally run things at a very high caliber. So I think I'm pretty surprised that whatever he is referring to exactly, I
- [07:42]
A
Think it's notable that that's a really good point. So I think that when we talked a little bit in the beginning of the year about whether this would be the year of agents and the fact that a couple months later it seems like that's been delivered upon, a lot of that has been coding agents and there's been some cross application in things like work like cloud Cowork for instance. But it is important to say that the level of agent development that meta is taking on is much more ambitious. Right. It's this personal superintelligence where, where the AI knows your context and is helpful to you. And that to me seems like a much more difficult problem, much more open ended problem to solve than coding. And again just to go through everybody that's tried to do it, Amazon, Google, Microsoft, to some degree, which we'll talk about Apple, no one's been, I mean really nobody's been able to build this sort of personal superintelligence, personal general intelligence or even personal intelligence at this point. And so you really wonder where things will go from here.
- [08:47]
C
I actually think, I think that's a really good point because the idea, and I think this is exactly what Apple faced and with LLMs in general, the idea that something is like perfectly executable and predictable with any kind of simple build, I think the entire industry is understanding that it takes a lot more so for meta and the types of things that they wanted to release. Like if you are actually releasing to your billions of users something that. And again have you used Meta AI, the actual app?
- [09:22]
A
Yes, I have, but not a huge fan of it.
- [09:25]
C
Yeah, I mean it's pretty bad. Like it's genuinely like the image stuff was actually kind of good for a little bit or it was on par I feel with OpenAI and Google maybe a year ago. It's kind of like if you have meta Ray Bans you can, you have to use it and that's where you upload your videos through. But it's still like to translate all of the power that they have into anything kind of consumer facing. It's clear they're having a very rough time. But I don't know, it's still, to me the more surprising thing I think that you brought up was the fact that they are readily starting to use external models openly and say this is part of our strategy now. Do you think that's going to be part of the kind of like high level strategy going forward?
- [10:15]
A
Yeah, for sure. I mean definitely. It definitely seems like that is going to be something that Many more companies do. Especially, you know, we talk about the companies that have struggled to do this, and it seems like Apple has had success in a way. Now, I don't want to crown them yet because Siri isn't out and I'm sure there'll be bugs that we'll have to talk about. But. But what they've shown on the new beta version of Siri, at least from, you know, what we've been able to see, has been pretty impressive compared to where they were. And again, that was probably the low floor. But it's easier for Apple because they have an operating system. And Meta, you know, the glasses is an attempt for them to build their own operating system. So it does make me wonder, okay, well, is this replicable?
- [11:02]
C
Well, hold on. But Meta has. I don't know if it's an operating system, but it has the single stickiest user base of any company. They essentially own the entire world in terms of pushing out software. I mean, remember threads? I don't even. What do you think it gets weekly average users, even though it has no cultural influence.
- [11:25]
A
Like, you want the real number or the fake number, right? Is it just people getting people getting their threads notifications pushed on Instagram and clicking to get that number off the Instagram app?
- [11:35]
C
Yeah, but that's what I mean, that they have the greatest distribution mechanism of any company in the world. So if they just do something, okay, it would be everywhere and it'd be very good. So that's actually even more surprising to me that whatever they're trying to do, that they haven't actually been able to do anything reasonable on the AI front, on the consumer side. But I think, I mean, the more we're talking, it really is clear there's no strategy because, like, is it a consumer app experience? Is it the actual foundation model it seems like they're giving up on? Which is actually crazy given how much money they just threw into it. Is it on the compute side? I think we should get into that, but I don't know. What do you think is the bet for Meta?
- [12:23]
A
Okay, well, actually, I'm going to take a different perspective from you and say that, that this isn't a Meta problem, that this is an industry problem. And I do think the fact that, like, they don't. If you have the operating system, you're sort of de facto plugged into a lot of apps and you can use the context side of generative AI to make sense of them, but I think it's more telling. So I think the strategy for Meta and all these companies is to do something similar. And I can't believe I'm saying this to the vision that we saw with Apple Intelligence. It's just that everyone has struggled to ship it. And then of course, like, you know, you go from context to having the thing take action for you. And that's sort of like the holy grail, the personal superintelligence. That's also an action layer on what you're doing. I can't believe I used the words action layer, but sort of seems like it makes sense.
- [13:10]
C
But the context layer, now it's about the action layer.
- [13:13]
A
Okay, don't get on me too much for this, but here's the thing, all right? When you go around the horn and you see all these companies struggling and now you start to see them sell their compute, it, it says something is happening in the AI industry, which is where there once used to be multiple points of failure, it now seems like there's fewer points of failure. It now seems like the AI industry is moving toward two points of failure. OpenAI and anthropic. And those businesses, you know, while we both would agree that those, those products are good, the businesses are being kept afloat by financing and they're losing a lot of money. And so when you see Meta going out. This is from Bloomberg. Meta is developing plans for a cloud infrastructure business that will sell access to AI computing power and models, setting up a new vector of competition with industry leaders like aws, Microsoft Azure and Google Cloud. When you see that happening, when you see it happening with SpaceX, when you see Microsoft, who's effectively gone out and tried to build its own model with opening eyes ip and all they've been able to do has been is be, you know, effectively a reseller and an implementer. This is why I'm leading with this. It makes you wonder, is the health of the broader AI moment somewhat in trouble?
- [14:41]
C
I mean, I think it is. I think it's like. But I guess this is something that's been very difficult to kind of square for a long time for me is, and I've said this many times on the show, I genuinely believe in the medium term, maybe long term, there's going to be just everything is going to be identified. There's going to be massive need for compute across the entire economy. In the short term, I think this is exactly what we're seeing that Meta is saying AI agents aren't seeing the way the development and scale that they had thought they would. I think, like we're very quickly seeing that these Companies that like the entire capex, the entire compute trade is definitely not going to be the straight line that everyone thought it was. I mean, even today I think I saw Sandisk Micron today was another down 14%. Yesterday was up at 10%. Like, like the volatility around this trade is terrifying and I think it just shows that no one has any idea exactly how this plays out. But I do believe what, like the, the momentum with which a lot of these stocks have moved is not sustainable. And I think we're starting to see those kind of cracks in a, in a big way.
- [16:06]
A
Not only not sustainable, right. Like more and more dependent on the success of two companies. And if those companies have a hiccup or if they get into further trouble with the US government, who knows where this whole thing goes, right?
- [16:21]
C
Well, yeah, yeah.
- [16:22]
A
I mean, talk about OpenAI anthropic.
- [16:25]
C
Yeah. In terms of the government, I think that's a whole other topic. But how do you see the entire compute trade playing out? Because if you think about it, there's like a couple of originally, so memory is one part. Then you have the actual kind of providers of the compute with which originally it would be we own the compute and we own the model product. But very quickly SpaceX was the first to go. Now that is going as well, saying we'll sell the compute because we have excess compute. Where is the money going to be made? Where do you think it's going to go?
- [17:06]
A
Well, here's where kind of talking about these things with nuance, like we try to do on the show, sort of starts to wipe out your clean narrative. Right? Because on the other side of this, you do have rapidly advancing technology that OpenAI and Anthropic are producing. And the whole case is basically that, that the product continues to improve exponentially and the revenue will continue to grow exponentially. And so that's the other side of what I'm saying is like, yes, there are twos there. There are, it looks more and more like there are two points of failure. But if those companies deliver, then, you know, the everything like the bottlenecks will continue to bottleneck and compute will still continue to be scarce, etc. Etc. So it is possible that they're able to sort of make good on the bet. But right now it kind of feels like the entire economy is like a vc and of course I'm being hyperbolic, but a VC making a bet on one particular vision of one particular technology working out.
- [18:10]
C
I don't think that's hyperbolic. Like I don't know, I've talked to a number of people about this and to me the almost scariest part of this is all of the people making these decisions are effectively in the same circle or is somewhat of a limited circle. So I mean this comes up all the time like from like non tech but investor friends where it's like do you think like why do you think the smartest people in the world believe that this much investment in data centers and compute capacity is required and like in the short term. And my answer a lot of the time really is, and I'm curious what you think about it is everyone is talking to each other in this small group of people. It's only a finite amount of companies that are even in this conversation. A lot of them are probably in the same WhatsApp groups and social circles and dinners and stuff like that. So it's like to me there really feels like there is a very big group think component of this, you know, like separate from any kind of like traditional forecasting other than if you just extrapolated out like a couple of months of activity. Do you think that's, do you think that's hyperbolic? I'll up your hyperbole. Hyperbolism.
- [19:33]
A
Hyperbility.
- [19:35]
C
Hyperbility. All right. Whatever the correct grammar is around hyperbolicness.
- [19:43]
A
All right, I'll, I will take a chance to answer this. I mean here's the thing. The reason why I use the word hyperbolic is because there is evidence that it's working, right? So it's not like this sort of, it's not, we can agree, it's not a mass delusion. We've seen great leaps in this technology's ability and in these companies balance sheets, you know, well, at least their revenue numbers over the past few years. Undeniable progress on that front. And there seems to be the Runway for these things to keep getting better because the way that they're being applied is just starting to, is just starting to flesh out. So this idea that like all right, well these guys are all. It's not okay, I'm sorry for to our crypto friends it's not like NFTs, right? NFTs you could say was just like a big delusional run up. This is real. But the, the magnitude of the bet is sort of where you get, get into, into trouble. And it's almost like, well, you couldn't really see it out unless there was a bet this size. I mean the amount of compute that's going to come online in the next couple years is insane. So it could be one of those, you know, typical bubble, you know, boom and bust cycles where, you know, the raw materials end up getting cheap and then you end up getting the product. Even though like some investors get wiped out. Like that seems to me like, you know, potentially at this point, given where things are trending, the most likely case.
- [21:13]
C
All right, I like that. And also I've confirmed that hyperballism is a word, so I'm going to stick with that one and let's up our hyperbolism in this, in this segment. But yeah, I think definitely like the scale of compute has been massive. The demand will be there, I genuinely believe. But like everything you read and the price of every asset within this entire value chain has been demand is outstripping supply, Demand is outstripping supply. And I think like that this is a pretty important moment. The idea that like suddenly you're seeing, wait, let's like people are slowing down and realizing from a pure compute perspective, I think already people are understanding that it's again, it's not, not just a straight line, the kind of exponential curve, it might be a straight line, but it might not be an exponential curve, which I think matters a good deal. But I mean the vibe shift in terms of going from frontier models for everything, it's going to be expensive, it's going to be massively consumptive to what's the cheapest and most efficient model. We talked about this last week again, but every, like, you know, Brian Armstrong at Coinbase, everyone is now bragging about reducing their per usage, per task compute, while still maybe increasing their overall compute. But everyone's talking about token efficiency right now. And it is crazy to me the speed at with which that vibe shift happened.
- [22:51]
A
But I don't think any of that stuff matters as long as the technology delivers on the promised. The problem is that more and more companies have found that they've sort of put their effort, their best people on it. The resources you're supposed to be able to sort of, you put this resource down, you're supposed to get X amount of return and they're struggling to it. And you know who's an interesting case on this front is also it's Google, right? Like Google made this ultimate hedge. Imagine this. If you think that compute will lead to, you know, just unprecedented gains or gives you the straight shot to AGI, why would you rent that compute out to other labs? But Google's done that with labs like Anthropic and I think there's a deal with OpenAI as well, right? So they Did a hedge where they're like, we're going to build our own stuff, we're going to give the DeepMind team a lot of compute. But just in case we're going to actually speed run the my the the metas and the SpaceX is here and grow our cloud business in parallel. And that's what Google did. Very interesting strategy. Given where we are right now.
- [23:58]
C
I've been wondering, have you been using Gemini much recently or.
- [24:04]
A
I'm going to be honest, zero.
- [24:07]
C
Okay. See, I'm actually traveling right now, but
- [24:10]
A
I thought no travel examples.
- [24:12]
C
Oh, damn it. Oh my God. I literally.
- [24:15]
A
All right, no, no, no, come on, let's hear it
- [24:19]
C
for, for regular listeners. We. I, I always get angry at Alex when everyone for the last two years across the industry always defaults to a travel example and travel booking for what is agendic. And it is one of my biggest pet peeves and I'm now doing exactly that. But I think but I'm gonna go with it. So it's been interesting to me. Gemini got really good maybe eight to ten months ago, maybe a year ago, and suddenly was on par. I'm curious how you remember the cycle. It is really not kept up with ChatGPT and Claude. I feel like for pure consumer usages. And that's why I'm saying in the travel context, like it's still fine and it's doing. It's telling me, you know, what pub is showing the US World cup game at 1am and I watch it from my hotel. I was like, still exploring. Maybe I'm going to go out for it, but. But it's still good for that kind of stuff. But it hasn't really done anything different or improved, which has been kind of surprising to me. And then I think maybe is Google, as you said, are they hedging? Are they deciding? Like we're gonna. We don't know if we're gonna be the be all end all of the actual application layer. So that's why let's hedge and also be the compute seller as well.
- [25:53]
A
Yeah. It's possible that Sundar saw this earlier than everybody, which is that the progress would be a little bit slower than the OpenAI's in anthropics were going to let. We're letting people, you know, believe and said, all right, we'll make this bet where we think this is going to progress somewhat fast, but we're not going to bet the entire farm on it. And I will caveat for a moment. I do use Gemini in AI mode in Google search. That's actually pretty good.
- [26:20]
C
Oh actually I will say, and I honestly feel so basic, like when I enter AI mode and maybe that's kind of like obnoxious being in the AI industry, but like I'm using AI mode way more too. So maybe that is where they're kind of like hedging their bets a lot more for really basic stuff. Again, I'm going to stick with the World cup theme, but like what was the score of a match? And then asking a couple more questions about specific players. I'm in AI mode right now, so maybe they're realizing that that integrated product can be where the farm needs to be.
- [27:01]
A
Yes, but again, I'm going to go back to my central point that I've been beating like a dead horse the entire event, the entire. Sorry, the entire conversation. Google doesn't have this sort of centralized know all assistant, the same one that Meta wants to build, that Apple wants to build, that Microsoft wants to build, that Amazon wants to build, build. And so my, my thought about slow playing is go ahead.
- [27:27]
C
Why would they not try to be building a personalized super intelligence when they have more personalized intelligence on every human being in the world? Maybe Meta is the only competitive one there.
- [27:43]
A
No, no, no, they are, they are trying to build it, but it's also been slower for them as well. That's what I'm trying to say. So I think they might have seen that this has gone, this was going to go slower. And, and this is what I'm saying, there's a systemic issue here that, you know, despite all the promises that like the progress would be fast in this specific area, which everybody thinks is going to be very lucrative, it has not happened. And so you're now in this world where at least in the near term AI advances will be enterprise, which is sort of the same pivot that OpenAI made, which can be a good business, but ultimately is a little bit short on the vision because it's so task specific.
- [28:27]
C
So basically to all our listeners and everyone out there, there's going to be mass white collar knowledge work, unemployment, but you won't have your personalized super intelligence. Is that what you're saying, Alex?
- [28:40]
A
No, definitely not. I. Look, I don't think that this mass unemployment narrative is probably panned out either. Right. And by the way, like, you know, Dario last year in May 2025 said, well might, we might see this white collar bloodbath, you know, starting in a, within a year to five years. Well, we're a year past. It hasn't happened and so we got what, four years left and you know, I, I again like it's weird to try to talk about this comprehensively because we're like kind of, there's a large gap between the hype and what's actually happened. But that doesn't mean what's actually happened is not significant. It is, it just seems like as Zuckerberg said, you know, it's going a little slower than anticipated. It could never meet that hype expectation, I guess.
- [29:28]
C
But to say you can never meet that hype expectation, I mean the valuations require that. Right.
- [29:36]
A
Well that's again that's where we might get into a real economic problem here.
- [29:41]
C
Do you think there's going to be systemic issues or do you think whatever, let's say, I'm curious, let's play out the scale of growth slows, the valuations are not realized. People look at the numbers of OpenAI anthropic things don't look good, which we already kind of know. They're not great. Like do you think that leads to widespread systemic problems?
- [30:14]
A
I mean I think you're much better positioned to answer that question than I am having been on a trading floor during the financial crisis. So I'll actually defer to you. What do you think?
- [30:25]
C
Okay. Okay. I guess turn it back on me. I mean I don't think it, I'm not worried about that as much as I've probably been worried about other potential systemic issues in the economy. So maybe it's, I still feel it'll be relatively contained, but I don't know. I think like it's, it is we, we are as economy. It feels like just diving deeper and deeper into this one way trade that if it unwinds, it's like how fast does it unwind? How many people are involved? How many kind of like countries are involved? I mean it's like the entire South Korean stock market right now. So, so I do even, I mean it's a good chunk of American stock market growth as well in the last six months or so. So I think that it'll definitely be painful. But I, I don't think like it's housing crisis 2008 level but I think it'll be painful.
- [31:22]
A
No, no, yeah, I don't think it'll be that either. All right. Meanwhile at Microsoft talking about another company that can't get its act together on AI this is from an interview that Satya gave last week with the, with the Journal. We can't let the AI giants eat the economy. The chief executive of Microsoft is joining a growing effort to take on artificial intelligence giants OpenAI and Anthropic Open. Outlining in an interview his vision for the next wave of the AI boom, one involving cheaper models, more user control and political messaging that wins the public's trust. Nadella offered a blistering critique of critique of how the race for AI supremacy has taken shape with a small group of companies capturing the value of the world changing technology as they make dire predictions about safety risks and job loss. You can't say, hey, all white collar jobs are gone and this could even be a weapon and we will use all the power to build data centers. And it told the Wall Street Journal the public, he predicted, wouldn't tolerate just a few models and companies doing all the learning for the world. I mean this is rich, right? And again, this sort of goes with our conversation today. Microsoft is OpenAI's biggest shareholder. Microsoft has access to all of OpenAI's IP. And what has Microsoft done? I mean it's grown Azure, congratulations. But it's also grown Azure as AWS and Google Cloud have grown and it's taken all that advantage and, and it, it should be the one eating the economy.
- [32:51]
C
So what do you think is, what do you think Satya has tried to do? I think this is actually one of the most interesting like I mean the, as a Wall Street Journal editorial, like this is a big deal again I think like it's been interesting because a lot of these have been, you know, circulating and discussed a bit but as you said, like they, they are one of the critical parts of the overall AI story. Not successfully but still from an investment standpoint and like capital allocation standpoint and obviously even an infrastructure standpoint. I mean the Microsoft is everywhere like to come out and say this and we're going to get more into the Alex Karp from Palantir interview as well. But it's pretty crazy to me that in the same week you have the CEO of Microsoft and then Palantir, which is the most, you know, like one of the most AI forward companies out there, both saying this exact story that like doomerism is kind of, I mean Satya basically said it's a bit of a marketing story with all white collar jobs are going to be gone, but we're going to use this power to build data centers in a facetious way. So yeah, like what, what do you think? Let's start with Satya. What do you think he's trying to do here?
- [34:13]
A
Well, both him and Karp are both trying to sell competing services. So we'll get into carp after the break, but this is the news for Microsoft this week. From the Information Microsoft Memo Details AI App Overhaul to earn the right to exist Microsoft is merging its consumer and enterprise Copilot apps and cutting unwanted features to earn the right to exist in the eyes of customers. In a 1200 word memo, Jacob Andreo, the executive vice president of Microsoft running co Pilot, said the new unified app will also feature AI coding tools and a new AI agent that customers would need to pay extra for. According to a copy of the memo, the new agents here's the great name dubbed Autopilot. Come on. Aim to be always on acting on customers behalf to automate the mundane. Microsoft previously previewed one such Autopilot agent Scout that can organize people's schedules and monitor their inboxes to compile curated digests of received emails. Microsoft said in a separate announcement Thursday that it would put 2.5 billion mind a new AI consultancy arm that could aid Copilot. It's called Microsoft Frontier Company. The new arm will embed 6,000 industry and engineering experts with Microsoft customers to co design, co innovate, deploy and continuously improve AI systems. And by the way, one last thing I'll just say is I think Satya has been speaking approvingly of Chinese models like Deep Seek recently. I mean him and Sam must have really had a falling out here, but basically, you know they're seeing an opening here which is, you know we are, you know we can't, we, we've struggled to build this co pilot thing. I think this is what's happening. But what we can do is take another swing at that and attempt to Compete by integrating AI apps whether it's OpenAI or others within our client base. And that will be our business. I mean effectively consultants with computer.
- [36:14]
C
Well, I mean this is. You have Deploy company from OpenAI I believe it was. And who was it with? You know there, there's been these number of partnerships around the deployment side to actually try to get people to successfully use these products. I don't know. I think the first part though, let's go back to Autopilot. How do you think? Because you did you ever see that, you know, statistic around Microsoft? Microsoft has 78 different products with the name Copilot in it. Like how do you think they're. Yeah, no, no, it is. I, I really wonder how these things must get done. And again remember we used to have like what was it like Bard Google. Bard Google, what was it Duo. You know like these companies have lots of competing products and internal issues in terms of like naming. But do you think Autopilot will be the breakthrough that kind of push agentic behavior to the average Microsoft user?
- [37:22]
A
No, I mean, I'm definitely in the please show me something stage of this. You know, I can't, I cannot get excited about a rebrand to an equally bad name.
- [37:32]
C
Well, also, why, you know, organize people's schedules and monitor their inboxes to compile a curated digest of received emails that like this stuff. Still, maybe this is worse than travel booking for me. Like, you can do that. Like, even Gemini is pretty good. Like Chad GPT connected to your like whatever inbox and stuff like, these are the most. There's a hundred YC startups that already solve this. Superhuman can do this. Which was bought by Grammarly. Like, why do you think there's. I guess most people just don't actually use these things. So they're still trying to go after this. Like, why do you think they're being so basic about this rather than like, we can do some really cool shit?
- [38:22]
A
I don't know. I'm completely confused by their strategy, to be honest. I've tried to make sense.
- [38:27]
C
No, no, no, come on, come on. You're. You are Satya outline to me the AI strategy.
- [38:35]
A
Okay, let me, let me, I will do it. I'll try, I'll try, I'll try. Yeah. I have completely failed to build competing. Well, I ham. I, I both benefited myself and hamstrung myself with my early deal with OpenAI. That meant that I had a good chunk of the leader of the AI moment and that was great for me, but it also prevented me from developing competing services. Okay, so then I signed some new deals that said I can now compete, you know, build competing services and I have OpenAI's IP, but it's not all, it's not very easy to get that IP out of OpenAI. And my attempt to build a competing service has not done. Has not gone very well. So basically what I'm doing is just like any other AI model customer using the technology to try to bolt it on to my technology. My, my technology. And that's led to a poor result and in copilot. So what are my options? My, my remaining option is I have a lot of compute. I have access to these models and I have a lot of enterprise customers. And to me, you know, instead of trying to build, build, necessarily build my own thing, what I could do is use my technical expertise to make a business out of helping those customers sort of use these foundational models to be better. And when I go and do that, who am I left competing with? Good old open AI. And that's Microsoft strategy right now. Talk down OpenAI and try to win those contracts and try to make the best out of a mess that I've in part, created for myself.
- [40:09]
C
Okay, I like that. I think that could be the Satya strategy. I was going to go a little more extreme, though, on this one.
- [40:17]
A
All right.
- [40:18]
C
I was thinking extreme.
- [40:19]
A
Ranjan, coming in hot.
- [40:22]
C
Well, it's out of your July 4th, 200th anniversary of America, I think.
- [40:30]
A
Ladies and gentlemen, you wanted fireworks. On this podcast, you're getting your fireworks Kaboom.
- [40:35]
C
I think we have joked at times that, like, Google's greatest win in recent times was due to Sundar's McKinsey past and thinking about organizational change and merging Google Brain and DeepMind into actually something that was able to push something forward. Now, Microsoft created its Microsoft AI division, but it's so clear when you have 78 products called Copilot, and it's clear this is just this, like, tangled web right now. And now they're creating Microsoft Frontier Company, which is a whole other thing. Even though I would think deployment should be part of the overall Microsoft AI and cloud offering anyways, I think he should just merge a bunch of these divisions together and say, like, I am ready to. Actually, Microsoft is on every computer of every enterprise customer out there, but we are going to make a drastic change on our internal structure and actually build something that is useful and exciting for customers. So I think he should go full Sundar in the next six months. That's what I think.
- [41:46]
A
I like it. I like it. I mean, you got to take big swings now. My big swing, if I were Satya, might be to just bring back bad Sidney, the evil version of the Bing bot, which I think there would be a lot of appeal for.
- [41:57]
C
Wait, Sydney, wasn't it?
- [41:59]
A
Sydney was the race alternative.
- [42:01]
C
I was the. Ty was the racist Microsoft.
- [42:05]
A
Tay Tay. Tay Tay, Right. That was pre Bing. This is the original Bing that tried to take Kevin Roose's wife away.
- [42:13]
C
Oh, he tried to take Kevin Urus's wife. Yeah, yeah. You know what? Bring all that back.
- [42:19]
A
Satya.
- [42:19]
C
Bring all. Bring Sydney, bring Tay, Bring the races.
- [42:23]
A
Like, just what? Don't bring Tay back. Let Tay be dead in AI graveyard. No, you got to take some. You gotta have a.
- [42:33]
C
You know what? I'm gonna. I'm gonna go out on a limb here and say Taylor just reminds us when a time when AI was pure little There is like your racist grandparents and you know, before LLMs had advanced very much and you know, it's just when it was, it was just an LLM trying to predict the next token and it didn't know what end, didn't know any better. And it just reminds us of a pure time of AI before all this, all this chaos and compute and data centers and everything else. So though I miss Tay, I know
- [43:06]
A
this is a joke, but I, I can't get on board the Tay train. I can't, I can't. But kind of an AI that once you pressure test it enough, gets horny for you and tries to do wrong things. I mean, it's not for me, but I'm saying there could be something there actually.
- [43:25]
C
Maybe Satya, the greatest pivot of all time. Open AI gave up on the adult erotica chatbots.
- [43:32]
A
That's right.
- [43:32]
C
Maybe it's time. That is your opportunity, Satya. Go for it, man. This is it.
- [43:39]
A
We need to take a break. We're gonna. We'll be back right after this.
- [43:42]
B
At Avocado, we know that not all mattresses are created equal. While other beds trap heat, ours sets it free. Made without polyurethane foams and crafted with natural latex, cotton and wool. Breathable, comfortable and supportive. No overheating, just clean, organic sleep that performs. Save 15% on our award winning mattresses. This July 4th sale, avocado celebrate organic shop@avocadomatress.com or leading retailers nationwide. MeTV is America's number one classic TV entertainment network, airing over 60 of the greatest TV series every week. Now MeTV presents the Golden Girls of summer, showcasing the best of the golden girls. Watch Dorothy, Blanche, Rose and Sophia Weeknights on MeTV at 10pm 9 Central. Log on to metv.com now to find out where to watch MeTV. Free over the air and on cable, satellite and select streaming services. MeTV is memorable entertainment television.
- [44:44]
A
Oh, God, I'm laughing. As we come back from break. All right, we're back here on big technology podcast Friday edition. We've gone places we never thought we would go. And let's continue on. We have two, two more stories to hit quickly actually, since, since we were talking about this enterprise software situation, let's, let's. Or this sort of enterprise AI application. Let's get to this Alex Karp interview, Right? So the CEO of Palantir, Alex Karp, was on CNBC this week and he said something has gone completely wrong with how AI is sold. He. He says what the technical customers want is control over their compute, their models, their data stack and their alpha. They want to know how they own the means of production and it's not being transferred to someone else. He's saying they should ask who owns the data, is it being cached? Are the prompts secure? Is this being transferred to you? And basically his point is he is arguing that if I can summarize his words that the AI foundational labs have so lulled their customers into a state of comfort that what they're doing is selling their technology at a loss so they can copy their customer's IP and build it out in their own products. And David Sachs of the White House and all in podcast said this as said as much on X he said look at Figma Anthropic blindsided it's then business partner with the launch of Claude Claude Design figma's founder said Anthropic has not been consistently honest with them. This isn't an isolated example. Anthropic has launched Cloud science, Cloud security Club Legal and of course Claude Code Dario has argued that the open source models over that open source models powerful enough to compete with Anthropic are dangerous. But dangerous to whom? Not to enterprise that want to retain control over their data and workflows. Dangerous to a business model that benefits from customers having few real alternatives at the model layer. So basically what do you make of this? And this was a very popular interview, it made the rounds. What do you make this idea that the models are training off of their customers IP and then just trying to regulatory capture all competitors out of the way so that they can therefore they can therefore make good on their investment and win.
- [47:03]
C
So I watched all 18 minutes of the Karp interview. I'm not like a big Alex Karp fan I would say but he really captured the state of the current market and kind of like current political situation I think really well and the current incentives that a lot of again the Figma example is such kind of a brutal one that we all just kind of gloss over but all of these of connect to our system, feed us all of your data and then suddenly we will launch competing products. And remember that was the thesis behind a lot of the kind of like valuation growth around they're going to subsume more and more of the overall software ecosystem and I think, I don't know like there, there's still this kind of like general assumption that Anthropic at the enterprise level OpenAI are not training on the data that is fed through them
- [48:07]
A
but then really because they say straight Up. They'll train on your data.
- [48:12]
C
No, no, no. Chad. Like, Chad the Enterprise.
- [48:15]
A
On the Enterprise side?
- [48:17]
C
Yeah, on the, on the enterprise level, which I guess like a FIGMA would be. If the average user is connecting FIGMA through Claude, then all that stuff's getting sucked in and like being. I mean, there's no worry there. But, but it is to me overall, like, it's interesting that like, that has not been a real concern for most people yet. And like the fact that David Sacks of all people, also Alex Karp, like, you know, before I feel that would have lived in the conspiracy theory side of the world, but the fact that they are saying this is happening at the enterprise level is pretty nuts. Like, I think it's like, I mean, whether it's just living as kind of accusation and they're just not happy with Dario and whatever kind of like interpersonal conflict there is. I think, like for both of the Frontier labs, I think this actually is a pretty big, interesting, big issue that if we start to realize, like, actually, yes, they are training on your data and the goal is to kind of subsume your business model, which it's clear, non enterprise, a FIGMA plugin receiving all of that data and understanding how people design and then using that to create cloud design. I guess it's crazy. Like, that just happened and it's done. So, yeah, I think. Will everyone be okay creating the next Claude plugin? Like, where do you see this playing out?
- [49:47]
A
Well, it all ties into sort of where I was at the beginning of the show and I sort of haven't deviated from, which is that we are, we may be entering a world where the entire, the entire bet here is not dispersed but concentrated. Right. And when that happens, you are going to start to see some of this stuff come out from companies that would surprise you for speaking. Speaking out about it and speaking out against it. And there is, I mean, there is this possibility that if it does. Okay, so, so that's like, you know, that's one situation. And now if it does go right, right. If it does become a situation, like Sam at the end of the year last year told me that if a company, you know is AI native versus bolt on, then the AI native company is going to be the one that wins. There's a big economy of companies that it would be built on for. And you would imagine that they would sort of, you know, they would come out of the woodwork and start to protest. And by the way, we haven't exactly seen a summer of companies building on these platforms. In fact, most of the value, it seems like most of the value being captured by these platforms, you know, is being captured by them versus companies, companies built on top of them. So this is where the competition starts to heat up. Where when it's these two single point of failures and everybody else, you know, seems like they, they view themselves as potentially picking up the scraps, then it is open season on these businesses. Because I don't necess, I don't think it's anthropic and open AI is sucking up the data of their enterprise customers and building, you know, their applications for them. I think it's different. But I also think that like, you know, sort of you're going to start to see these companies try, try to peck on against these bigger, these bigger foundational labs, even if they're not in the perfect position. And Palantir, by the way, isn't exactly the best actor when it comes to freeing your data that you put in there.
- [51:47]
C
Well, no, I know. I mean that that's what's almost like comical about it. I mean, you said these, like against these big frontier labs, but it is. Let's just take a moment. And when Satya, Alex Karp and David Sacks are all echoing the same message that concentration among OpenAI and anthropic is dangerous, that's pretty remarkable. I think these are very different personalities. These are very different. They have very different vested interests and constituencies and they're all in the same week echoing the same message. And as we kind of go to this summer, it's the group, okay, maybe, maybe this is like the non data center investment group text on one of them. They're all like, gotta buy more compute.
- [52:37]
A
Zuck is. We'll have the same message.
- [52:40]
C
Oh, I like that. Okay, you heard it here. In the next week or two, does Zuck come out and say something similar about the danger of concentration across two frontier labs? That will be something.
- [52:54]
A
Yep. And that's like the perfect lead into our final story, which is this new story that's come out that Sam Altman is from the Financial Times. Sam Altman has suggested that the White House potentially take a 5% equity stake of OpenAI or the US government take a 5% equity equity stake of OpenAI. And that would also. He also suggested. This is again, and it's not clear that it was a formal proposal, but he also suggested Anthropic and Meta and Google also give this 5% to the government. And that's like, I mean, talk about coming, you know, you sort of, you're in this position, you want to entrench yourself as the winner, right? Just say, oh, government, take, take my, you know, a chunk of me. And by the way, you know, wink, wink, when I have a model ready for approval and wink, wink, when I have a competitor ready for approval, their model ready for approval. You know, it's like you think the government won't want to look out for its own investments? Of course it will. So that's sort of an interesting coda to this entire situation.
- [53:54]
C
I mean, I think this is where it is getting kind of out of hand in terms of like we talked about systemic risk earlier and. Well, actually, you know what, I'll try to be nuanced on this one. It's like on one side, I think this is absolutely ridiculous, absurd and like, you know, like the idea, if it's actually being pitched actively from the OpenAI side, I think that's like a bit depressing. But also on the other hand, again, you have the US Government currently under the administration buying intel and whatever else. But also, like, I think, like one thing we haven't talked about is the, we're going to be heading into midterm season very soon. I mean, we. Very, very soon. We know data centers, we know anti AI sentiment is going to be the center of a lot of this. Like, do you think some of this is these companies trying to at least not be all in on one thing the same direction in trade and try to be like, at least create some, I don't want to say diversity, but some kind of, you know, like it's going to be a massive issue and it feels like people are starting to not all speak the exact same tone on this right now.
- [55:12]
A
Yes. No, there's definitely a certain degree of desperation that's involved there, which is just like these companies being like, please, like us, please approve our models, please don't, you know, become our enemies in the public. Because it's going to be tempting.
- [55:28]
C
I won David Sacks on a highly produced Rick Rubin Poly market commercial, cross legged and thinking. And then it's going to air during the World cup final. I'm not happy until that happens.
- [55:42]
A
All right, I want to end with this because we do some serious journalism on this show. You know, despite the laughs, we definitely need to get into the most important things influencing our world. And there has been some confusion about a major event happening this week that I think we need to dig into. This is from Cosmo. There's growing evidence that MSG is a ruse and Taylor Swift and Travis Kelce's Wedding is elsewhere at this point. Even the New York Times is reporting that Taylor Swift and Travis Kelce's wedding is taking place at Madison Square Garden tomorrow, July 3rd. So this will happen on Friday. I mean, the Thursday story. And there's a lot of evidence, like, you know, the fact that a ton of decor is being loaded into the arena as we speak. But a small corner of the Internet has remained convinced that MSG is an elaborate roost to keep everyone's attention off of Taylor and Travis's real wedding location. And honestly, there's growing evidence to support that theory. Now, Ranjan, we don't have enough time to go through all the evidence here, but basically, I want to get. Want to get this on record. Do you think this big MSG wedding has been a head fake all along and that in time it will be revealed that. That Taylor and Travis just got married somewhere else, despite all the stuff they've shipped into msg?
- [56:56]
C
No, I think this is real. I think. I think Taylor recognizes the kind of like, you know, perilous state of the overall AI trade, the overall kind of, you know, American political infrastructure right now. And she recognizes on this July 4th, she needs to do something. Travis Kelce's long for the ride, and I think she is going to bring this message of unity, peace, and love to. To the world. I'm excited for this Madison Square Gard Garden Garden wedding. Do you not think it's for real?
- [57:32]
A
I'm not. I think they will. They will. I think they will tell the world that they got married at MSG and never show up there. And meanwhile, maybe they'll send some body doubles and the entire thing will happen on Rhode Island. That's what I would do.
- [57:47]
C
No, no, no, no, no, no, people.
- [57:49]
A
That is what I would.
- [57:49]
C
You can't piss off 20,000. You can't piss off 20,000. Friends, fans, whatever you want to call them. Now, Taylor, she knows how.
- [57:59]
A
They only have a thousand people going to that thing.
- [58:03]
C
It's an msg. What do you mean, that's. How do you only have a thousand people?
- [58:09]
A
Well, you just. You don't have to fill the stadium just because you've rented it. You're the richest people on the planet. Well, not the richest, but close to it.
- [58:18]
C
Trillionaire Elon. Then he would. He would fill the stadium. But I don't know. I think. I still think. I think it's for real. I'm gonna. We'll find out in a day or two. But I definitely think Taylor and Travis at msg, they recognized after the next one. That. That is a magical place. Right now. They want to be in New York. And I think. I think I'm excited for it. I think it's a good thing for America.
- [58:44]
A
Okay. Well, Ron, John, thank you again for bringing the fireworks this week. You know, we. We went big, we went small, and then we got to the serious stuff at the end, so it's great. Always great speaking with you. Thanks for making the effort to do this all the way from London. Appreciate it.
- [58:59]
B
All right.
- [58:59]
C
See you next week.
- [59:00]
A
All right, everybody, thank you so much for listening. And we'll see you next time on big Technology Podcast.