Pod Insights Big Technology Podcast DeepSeek Rises, Stargate Drama, OpenAI’s Operator Debuts About this episode Ranjan Roy from Margins is back for our weekly discussion of the latest tech news. We cover 1) DeepSeek r1, an efficient, cheap reasoning model from China 2) How DeepSeek stacks up against the state of the art 3) Does DeepSeek invalidate the billions of dollars invested in current foundational models 4) DeepSeek's wild origin story 5) How DeepSeek changes U. S. and China's AI rivalry 6) OpenAI, Softbank, and Oracle team up to invest up to $500 billion in AI infrastructure in a program called Stargate 7) But do they have the money? 8) Will Stargate test the limits of the scaling hypothesis 9) OpenAI's Operator agent debuts 10) Will people want Operator? 11) Analysts downgrade Apple after Apple Intelligence disappoints 12) Wait, TikTok is still here?
---
Enjoying Big Technology Podcast? Please rate us five stars ⭐⭐⭐⭐⭐ in your podcast app of choice.
For weekly updates on the show, sign up for the pod newsletter on LinkedIn: https://www.linkedin.com/newsletters/6901970121829801984/
Want a discount for Big Technology on Substack? Here’s 40% off for the first year: https://tinyurl.com/bigtechnology
Questions? Feedback? Write to: bigtechnologypodcast@gmail.com
Learn more about your ad choices. Visit megaphone.fm/adchoices
Transcript 10,399 words · 3 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Alex Kantrowitz? The AI world is buzzing as a cheap, effective, open source model rivals competitors built for billions. OpenAI, Oracle, and SoftBank put a half trillion dollars in Stargate, but is the money really there? Agents may finally be here, and TikTok is still alive. That's coming up on a Big Technology Podcast Friday edition right after this. This episode is brought to you by Genesis. What does it actually take to put a Gentic AI to work across an entire enterprise? I recently spoke with Genesis chairman and CEO Tony Bates about why orchestration is emerging as a competitive advantage in customer experience and what companies need to get right in the AI era. Then, I sat down with Adam Mitchell, head of enterprise business solutions at Voya Financial, to go inside
0:45 Alex Kantrowitz? a large-scale agentic CX transformation, including the business decisions, trade-offs, and lessons along the way. You can watch both conversations now on my YouTube channel today. It's never been more challenging to get frontier compute. How do you know where to look, who to trust, or how much to pay? Akash is an open cloud that provides easy access, flexibility, and peace of mind. That's how companies like Razer were able to cut their inference cost by 15 times. Visit akash.network/alex for a free consultation. That's akash.network/alex for a free consultation. Welcome to Big Technology Podcast Friday edition, where we break down the news in our
1:27 Alex Kantrowitz? traditional cool-headed and nuanced format. Wow, we have a big show for you today. We're going to cover everything from this new model called DeepSeek, an open-source, cheap model from China that is totally reshuffling the AI world as we know it. We're also going to talk about the big project called Stargate, where OpenAI, Oracle, and SoftBank are putting maybe up to $500 billion into an AI infrastructure project that OpenAI is going to use. OpenAI has also released Operator, a new agentic model, and TikTok's still here. Joining us as always to discuss all this and more is Ron John Roy of Margins. Ron John, welcome to the show.
2:04 Speaker 01 Man, we got real news, some fake news, everything in between. I don't even know where TikTok falls in that, but I'm ready. I'm ready.
2:13 Alex Kantrowitz? This is definitely one of the weeks that I've been looking most forward to recording our Friday show because there is so much stuff to talk about and real important developments that have just come throughout the week. And on that note, I'm sure we have some new listeners here coming in from my Demis Asabas interview, whether you found us on Spotify or YouTube or elsewhere. Just to give you a rundown of how the show works, we do a big interview every Wednesday with someone like Demis. Next week is going to be Reid Hoffman. And then every Friday, Ranjan and I break down the news that's happened over the week. We cover a lot of AI, anything big tech. And so we're so glad that you're here. So first, let's talk about this DeepSeek story, which is a massive, massive story. Basically, what DeepSeek is, is a much more efficient reasoning model. They just released this model called R1.
2:56 Alex Kantrowitz? It's built by a Chinese startup. And the crazy thing is that what it's taken American companies, billions of dollars to do, it's taken them just a few hundred million. And I think that this could completely upend the AI industry as we know it. We're just starting to see the beginnings of this and everybody seems to think that it's legit technology. So Ranjan, I'm curious what you think about DeepSeek. Do you think it's legit and what does it mean for the industry?
3:25 Speaker 01 I'm incredibly excited by this. I think this is one of the biggest stories in a while, exactly for what the reason you're saying, that being able to train and deliver a model that's relatively on par with the expensive ones, with the O1s. I think this is huge and it's what I've been saying for a long time, that the power of the model, the amount of money that's being kind of like claimed, the, the big shows around massive compute structures was all a bit of a show and for actually delivering on something that people want to use. This shows that it can be done in a completely different way. And, and have you been using DeepSeek much?
4:08 Speaker 01 Like everything I've tested, it's relatively, and Joe Weisenthal had a good piece on, you know, how a lot of these like ChatGPT can effectively be commoditized, like a normal chat interaction. With one of these models, you can get similar results. And I already switched between ChatGPT, perplexity Claude on a pretty regular basis. And now DeepSeek has made it to my favorites tab and is up there as well.
4:35 Alex Kantrowitz? That's crazy because it just hit like a couple of days ago. So let me just give a couple of stats that just show the magnitude here. So this is from the New York Times. The Chinese engineers said they needed only about $6 million in raw computing power to build the system. That's about 10 times less than the tech giant Meta spent on building its latest AI technology. For further context, I mean, you could build this model for 6 million. We just saw OpenAI raised the largest venture capital round in history at $6 billion, right? So that level of magnitude is crazy. And then you have it. So how does it rank? How does its performance rank? So I was on Chatbot Arena earlier today, which ranks large language models. And DeepSeek
5:16 Alex Kantrowitz? is in the tie for third place. So here is what the rankings on Chatbot Arena look like right now. You have number one Gemini, number two Gemini, number three tie for third place between ChatGPT 4.0, the latest model, and DeepSeek. And then everything else that we've been talking about, Grok and the Clods of the world are below DeepSeek. So this is crazy. The thing I can't get out of my head is, does this sort of invalidate all the billions that these research houses have been spending on training their models if DeepSeek can do it for peanuts and rank this highly on Chatbot Arena?
5:54 Speaker 01 Yes, yes, and yes. I think the fact that we're seeing Claude Sonnet ranked at 18, DeepSeek tied at number three, I think it completely shows that the way we're thinking about foundation models at the center of the business battle for all these AI companies has been wrong. And I think the products, and this is my favorite, we always get into this every week, that for me, it's not the models, it's the products that are built on top of them and how good they are. Maybe this means that finally, we're going to move to a world where people actually focus on the product side of things and not the idea that the models and just having to build bigger and bigger and more expensive models is the only way that anyone's going to win.
6:39 Alex Kantrowitz? But then what happens to all this billion, these billions of dollars that these companies have invested in building foundational models? Really? You think it's completely gone? I mean, isn't that, isn't that like a destructive event for Silicon Valley in a way that we haven't seen before?
6:54 Speaker 01 I'm gonna go constructive instead of destructive. I think it's gonna, it's gonna force a reckoning. It's gonna really make people try to figure out what's the future, the next two to five years, let's say. And the way, the, the way the battle has been approached over the last one and a half to two years of constant big unveilings and constant, you know, attempts at trying to sell the idea of these more and more powerful and bigger foundation models. Exactly. It's gonna, it's gonna move away from that. And Meta, and we're gonna get more into what they're doing, the, moving into open source, I think was a powerful move from Zuckerberg from the beginning, but now having actual,
7:37 Speaker 01 like genuine competition on the open source side is fascinating. But yeah, I think the, the things that are built are going to be able to have, and it's great for everyone else. It's great for everyone else. It's not great for OpenAI, but being able to readily build and have your choice of what you're gonna build on top of and being able to do it fast and cheap, that's good. That's wonderful, just not for Sam.
8:04 Alex Kantrowitz? Right. And that's the way that Silicon Valley has been reacting to this. So Marc Andreessen called R1 one of the most amazing and impressive breakthroughs I've ever seen. And as open source, a profound gift to the world. Anjay Maida, who's also at Andreessen Horowitz, said, "From Stanford to MIT, DeepSeek R1 has become the model of choice for America's top university researchers basically overnight," which is fascinating because, as we know, research, universities just do not have Nvidia H100 chips, they cannot afford them. And Arvind Srinivas, he also said, "DeepSeek is two orders of magnitude more efficient with capital allocation than OpenAI." By the way, so after Arvind tweeted that, I DM'd him and said, "Hey, let's talk about this. I'm gonna write about this."
8:47 Alex Kantrowitz? And then he deleted the tweet and never responded to my messages. So this is obviously, it sort of goes to show that this is really gonna stir up Silicon Valley and AI in a way that we're just starting to grasp this week.
9:01 Speaker 01 I think the fact that Andreessen Horowitz is and Marc Andreessen is saying this, I think is a big deal because yeah, it's it's really pushing the idea. And again, you think about it from like a more, you know, pure venture capital standpoint, it is exciting because rather than only being stuck with having to get into the late inflated rounds of just a few companies, it really potentially opens up the door for way more startups doing interesting things that can actually explode in growth and solve problems.
9:34 Alex Kantrowitz? Yeah, and we should note that Andreessen did sit out the OpenAI round.
9:37 Speaker 01 Yep. No, I was definitely thinking about that, that, I mean, did they see this coming from a, and then that's why they they sat out not, I remember at the time, it was definitely notable. In my mind, I was like, maybe they've already, you know, seen some pretty solid returns from their early round investments, but maybe they really are believing in open source is the future. And that's really going to hurt than OpenAI is of the world.
10:03 Alex Kantrowitz? Yeah, it's quite possible. And I think that's one thing that we really need to hammer home here, which is not that it just took so much less money to build the model, but it takes so much less money to run the model. The model, the way that it works is super efficient, right? And that is going to allow startups to be able to build on top of this stuff without worrying about what we've been talking about for for months, which is this subprime AI crisis, right? The only thing that could have made this idea, the subprime AI crisis, which is basically that these companies have spent so much money training the models that they're going to have to charge people eventually a lot more than they're they're paying right now to use them. The only thing that could that could threaten
10:44 Alex Kantrowitz? that is if a model showed up that was so much cheaper to run. And that might be this one.
10:49 Speaker 01 Yep, it's rare that I'm agreeing with Chamath, but. He came out and said that like AI model building is a money trap. What you're seeing here is a feature of modern AI models. There's no bounding laws like Moore's law where advances can be predictably expected. And that open source will be the clear winner. And I think it's again, it's clear that we and the most exciting part about this is we've all been debating scaling laws and what it's going to look like. But to see these kind of like step changes in terms of cost and efficiency, just coming out of nowhere, coming from unexpected places. And we can get into, you know, the how this fits into the larger US-China technological
11:33 Speaker 01 battle and overall relationship. That stuff, that's, that's the exciting stuff. That's the rather than just another big fundraise from OpenAI. This is the stuff that I don't makes me happy.
11:45 Alex Kantrowitz? Yeah, same here. I mean, I was just totally blown. I was planning on writing this week about the drama between Elon Musk and Sam Altman about
11:51 Speaker 01 Which will definitely
11:52 Alex Kantrowitz? Stargate, which we're going to get into. But as the news came out about DeepSeek, I said, there's no way. And I had to make this the main story of big technology this week. And I just have been unable to, um, to get out of this news because it is, it's so interesting. And then you think about it from the hardware side. So Nvidia, right? Like what's going to happen with them? Because so many of their chips were needed to run the current models. And so there's a senior researcher who's pretty outspoken from Nvidia. His name is Jim Phan. And he's addressed this in two tweets, by the way. So the first one talked about how, uh, DeepSeek is like the real OpenAI because they're open source and because they're making it available, uh, to so many people. But the second tweet was so interesting because he gets into the finance side of things. And I had like DM'd him earlier and I was like, what's going to happen to your business?
12:37 Alex Kantrowitz? And he just puts out this tweet. Uh, he says, first of all, he says, it's a humbling wake up call to us that open science has no boundary. We need to embrace it one way or another. Then he goes on to talk about the financial aspect. Many tech folks are panicking about how much DeepSeek is able to show with so little compute budget. I see it differently with a huge smile on my face. Why are we not happy to see improvements in the scaling law? DeepSeek is unequivocal proof that one can produce unit intelligence gain at 10x less cost, which means we shall get 10x more powerful AI with the compute we have today and are building tomorrow. It's simple math, the AI timeline just got compressed. And this is the, obviously, it's the most optimistic.
13:22 Alex Kantrowitz? It's the best case scenario. And I was thinking to myself when I read that, like, yeah, if everything goes right, maybe that's the case. But I think I'm starting to be swayed by his argument. What do you think about it?
13:34 Speaker 01 Well, Jim, I am happy to see improvements in the scale. I think the, again, trying to actually like directly quantify 10x less the cost, 10x more powerful, obviously, those are nice round numbers. But to me, the most important thing is it completely changes the competitive framing for every company in the space. Suddenly, when you are thinking about, and already, I think I saw OpenAI said that O3 Mini, one of the better models will now be part of the Free Chat GPT plan. Like, it's going to put heavy competitive pressure on everyone, and we will move to a world where the expected cost for a developer is more in line with what DeepSeek's offering
14:17 Speaker 01 with R1 versus what OpenAI and other in Anthropic are offering with their models. And that's good.
14:25 Alex Kantrowitz? By the way, you can use DeepSeek right now. So you could just go in, sign in with your Gmail account and get using it. And if you click the R1 Thinking Model, you can actually see its chain of thought be written out. And it's pretty remarkable, the way that it thinks through problems. And let me, let me put this to you, because I'm curious what you think about this. Is this sort of the beginning or the end? Here's my argument for the end. The AI industry might have hit this wall with the scaling laws. Everyone's talking about how data, you know, there's a data wall, the models aren't getting better, they are starting to see diminishing returns. And then the only way to continue to advance them is with reasoning. And what DeepSeek might have done is just a speedrun through reasoning,
15:09 Alex Kantrowitz? right? Where it's gotten this uptake because it's figured out reasoning before everybody else and made it much more efficient and put that on top of, you know, maybe it's built that on top of old open source models to get these amazing results. Is it possible this is just kind of the terminus for generative AI? Where it's like, okay, it's what we have now, but it's cheaper as opposed to a step towards something more?
15:31 Speaker 01 No, I think it's certainly the former, maybe the latter, but the former, that this is just what we have now, but cheaper. But I actually think that's such a big deal. And you said it, what does that mean for Nvidia? What does that mean for OpenAI? What does that mean for Anthropic and Google? And actually, Microsoft plays a very interesting role in all of this as well, especially the relationship with OpenAI. But it just fundamentally changes the entire competitive landscape. And so whether that truly means, and I'm a bit still, I mean, what is AGI? What is actual true reasoning? Just showing chain of thought processing is a nice UI trick almost.
16:16 Speaker 01 And I like it and it's good. And you should go to DeepSeek and play around with it. I think every listener should. But, but what does that actually mean in terms of does, is this something fundamentally different? I am not convinced of that yet, but the same and significantly cheaper is a much bigger deal to me actually than some new level of reasoning.
16:41 Alex Kantrowitz? So can we just take a moment before we move on to appreciate how DeepSeek came about because the story is actually unbelievable. So it was built out of a quantitative stock trading firm called HighFlyer. This is according to the New York Times. By 2021, it had channeled its profits into acquiring thousands of Nvidia chips, which it used to train its earlier models. There's some, Alexander Wang says they may have 50,000 Nvidia chips, but that this came out of a quant firm. Rajan, I feel like you with some ties to the trading world and some ties to China are uniquely qualified to comment on. Is this like just a typical Chinese stories or trader story? Or is it as remarkable as I'm seeing it?
17:21 Speaker 01 Well, actually, I think that's why it makes it even more exciting for me. It's the idea that The greatest innovations, especially something focused around cost and efficiency, it actually seems more likely that it would come from a quant trading firm than a really kind of traditional AI research house, which many of the other companies essentially are with a business attached to it, that that actually, it does make more sense to me that making things much faster and more efficient, coming out of, again, unexpected places is actually kind of cooler. And, and it also, it changes, like, what could be next? Who's
18:02 Speaker 01 going to come out with something completely unexpected in terms of what's the next big wave in a generative AI? And it just opens up the door in a very cool way.
18:13 Alex Kantrowitz? So just game out what's going to happen because of this. I mean, I saw there were some interesting pieces on Blind, for instance. I don't know if you can trust this stuff. I have a Meta executive coming on in a couple of, a couple of weeks that I'm going to ask him about it. But basically, like, Meta's panicking that, you know, DeepSeek was able to train a state of the art model for less than the salary of like one AI executive. But I'm curious, like, what do you think this does to the valuations and the funding in the space? And does it invigorate AI startups to be able to build in the way they couldn't because it was cost prohibitive?
18:42 Speaker 01 I mean, it's interesting to me how Meta falls into this because, again, Meta investing heavily, but still like on the side of open source within this. So what did actual monetization look like? Or how were they thinking about it is very different than pay me for tokens on an API. But I, but I do think, obviously, taking a breath for a moment, it's definitely come in and really taken the entire industry by storm this week. And let's let it play out a bit. But it's, it changes the overall economics in a good workable way, if this is actually going to work. And if this actually, people are able to start building real things on top of it. And the economics of the way
19:27 Speaker 01 things have developed over the last two years for a lot of these companies never made any sense. We've, everyone in the industry is talking about it, especially OpenAI is the poster child of just burning money and having unworkable economics. This, in a weird way, makes things actually a lot more rational in terms of business.
19:45 Alex Kantrowitz? Right. And another thing that's interesting is it's open source. It has an MIT license. So people have been able to replicate it. So you can't just say that it's like, okay, China subsidizing this stuff, which is interesting. But on that note, I mean, it's interesting that Marc Andreessen is the one that's out there hyping this the most. It is a product of China and it is, you know, if you think about there, there's this race between the US and China to develop the best AI. And now it doesn't look like they have the best, but they certainly are factoring in the race in a way that they haven't before. And of course, you know, Andreessen's sort of been working with Trump on the transition and he's one of like the tech MAGA guys. And curious what you think it means. Do you think, do you think that this is a problem for US competitiveness?
20:29 Alex Kantrowitz? I mean, this is the one thing we haven't touched on before, and then we're going to move on after this. But like, they were able to do this even though there are constraints on China because they can't import the top level chips. And they were able to be creative and work around it. So it's just a fascinating wrinkle in the story to me.
20:46 Speaker 01 Yeah, I think it's definitely going to change. It definitely factors into the overall US-China dynamic in a big way, especially kind of showing the world and, you know, the companies and the executives who are, and we're going to get into Stargate in just a moment. That is the symbol of what the US's leadership on AI looks like. So to kind of just go out and show, oh, we can do it completely differently and everything you're putting up front like that is actually wrong. It's a big deal. I think that the scale of how cheap the actual development and training side of this model, there's definitely a lot of debate around what was built over existing things that America
21:30 Speaker 01 had already developed, whether it's Llama or other models and, you know, what kind of leg up and advantage from a cost perspective did that give DeepSeek in developing this? There's certainly plenty of conspiracy theories around, like, is this a psy opera front for the Chinese government just to kind of shake up and the entire industry that American companies have a great deal of leadership on? But I think overall, it's a reminder that, let's say they did build on Llama and kind of like use existing work that had been heavily expensive and done by other companies. In a way, that's kind of a good sign for innovation overall. It's like,
22:13 Speaker 01 Thank you Meta and others for really putting in the CapEx to to get the entire industry to a place where smaller organizations can actually do something big.
22:25 Alex Kantrowitz? That's right. And this is from Yan Le Kuhn, the chief AI scientist at Meta, who's the one that's going to be on the show in a couple of weeks. Everything goes according to plan. He says, "To the people who see the performance of DeepSeek and think China is surpassing the US in AI, you're reading this wrong. The correct response is open source models are surpassing proprietary ones. DeepSeek profited from open source research. They came up with new ideas and built them on top of other people's work. Because their work is published in open source, everyone can profit from it. That is the power of open source research and open source." So on one hand, you have this huge victory for open sourced and cheap models. And on the other hand, you got Stargate. And Stargate, so this is from the AP, "Trump highlights partnership investing $500 billion in AI.
23:07 Alex Kantrowitz? On Tuesday, Trump talked up a joint venture, investing up to $500 billion for infrastructure tied to artificial intelligence via a new partnership formed by OpenAI, Oracle, and SoftBank. The new entity Stargate will start building out data centers and the electricity generation needed for the further development of fast-evolving AI in Texas. The initial investment is expected to be $100 billion and could reach five times that sum." And so basically, there's like, to me, there's two sides to this equation, right? Two sides to this story. The first side is, open source, cheaper models are starting to really hold their own with the bigger models. And the other side is, we still don't know what the limits are of the scaling properties in OpenAI.
23:50 Alex Kantrowitz? And you have to spend a lot of money to get to the outer limit. Ranjan, I'm curious what you make of this. Is this going to be a big watershed week? Because we have both of these things? Or do you think that this is more PR than anything else?
24:04 Speaker 01 I think we said there's real news and fake news this week, and I think this is going to fall into the latter. One thing, this kind of like triggered in me. So I went back and looked. For people might remember in 2016, almost like identical, like I almost felt like I was having deja vu. So in 2016, Masayoshi Son went to the White House, pledged $50 billion to create 50,000 jobs, said the words, "I couldn't have decided such a thing before this new president." And at the time, that's when, you know, the first Vision Fund of 100 billion was being kind of finished up. And my favorite part of that whole saga was
24:49 Speaker 01 that they actually invested over $50 billion into the United States to create jobs. Majority of that was in WeWork. A bunch of it was in Uber, which, you know, did all right. But overall, he took something that already existed, was already under like the works for a number of years, and then went to the White House and said, "I'm doing this because Trump is president." So this one, it really felt like that because we talked about, there is a time, I think, how much did Sam Altman say they wanted to raise? Seven trillion?
25:24 Alex Kantrowitz? Seven trillion.
25:25 Speaker 01 Seven trillion. Yeah. So now 500 billion.
25:27 Alex Kantrowitz? That was the rumors. I think he joked, "Why not eight?" But there must have been some truth to that.
25:32 Speaker 01 So basically, the FT has reported that this has been under the works. Actually, there's a Washington Post White House reporter as well saying that this idea of 100 billion was already being pitched for a number of months. And then the money is not there. And this is obviously one of the best, juiciest parts of this, is that Elon Musk came out and basically called out this announcement and said that they max have like $10 billion. And then there was plenty of back and forth between Sam Altman and Elon Musk on this. But even the FT reported like that Stargate has not yet secured the funding it requires. It will receive no government financing and will only serve OpenAI once completed. The
26:15 Speaker 01 intent is not to become a data center provider for the world. They have not secured the funding. No one, a lot of people have been throwing out like, what are the potential, you know, compositions of what 100 billion can look like? But the numbers don't even add up. Anyways, so I'm not, it, this was such an egregious PR thing that I'm glad it has given us Musk versus Altman on this because at least we got something out of it.
26:45 Alex Kantrowitz? Absolutely amazing comments from President Trump on this also. They asked him if they have the money. He says, "Well, I don't know if they do, but you know, they're putting up the money. The government's not putting up anything. They're putting up the money. They're very rich people, so I hope they do." It's like, okay, so they don't have the money. So Elon is right on that. And even if, I mean, let's say they do get there, right? The question here is going to be, is this the right way to train, given what we just learned about DeepSeek? Bringing you back to DeepSeek. This is from venture capitalist Jeremy Liu. "If training costs for the new DeepSeek model are even close to correct, it feels like Stargate might be getting ready to fight the last war, like bringing an M1 Abrams MBT to a drone fight."
27:30 Speaker 01 No, I loved that. Yeah. The timing was almost comical, but still impeccable that this is the moment. And obviously, in the Trump context, you don't want to say we have a faster, cheaper, kind of like more low-key way of doing things. It has to be big and audacious. And that number 500 billion, it really falls into the, "Tell me a big number." What's the biggest number you want to think of? 500, let's call it that.
27:59 Alex Kantrowitz? But here's the thing, like, we are going to see much more capital put toward a new model, right? Or new model development. Maybe it's not 100 billion from OpenAI, but it's going to be billions, billions of dollars. Like even 10 billion is double their most recent fundraising round. Then you have Elon Musk, who's going to build this million GPU supercluster. And by the way, hot breaking news right off the press here. Mark Zuckerberg posted this morning. "Meta is going to build a 2 gigawatt-plus data center that is so large it would cover a significant part of Manhattan. We'll bring along about 1 gigawatt of compute in 2025, and we'll end the year with 1.3 million GPUs. We're planning to invest 60 billion to 65 billion."
28:41 Alex Kantrowitz? Capex this year while also growing our AI team significantly and will have the capital to continue to invest in years ahead. This is a massive effort and over the coming years, it will drive our core products and business, unlock historic innovation and extend American technology leadership. Let's go build. I do love how everybody has a note to Trump in their press releases right now. Look, here's the bottom line for me. You have these three research houses, at least, that are building these massive, massive data centers to train the biggest models known to man. And we're going to find out the truth about this scaling hypothesis, whether you can build an AGI just by scaling up. And I don't know, you might be saying like, let's go build the products. To me, it's exciting that this is actually something that they're testing.
29:27 Speaker 01 For me, the exciting part is the shade that's being thrown left and right, just as a observer. So you have, my favorite part of the post was, and we have the capital to continue investing in the years ahead, and we have the capital, definitely, you know, a nice little nudge to the Stargate project. Satya Nadella, in an amazing interview this week, was talking about how Microsoft is gonna, you know, and he's like, okay, that's all nice. We're investing $80 billion in Capex in infrastructure build this year, and I'm good for my 80 billion. So, I mean, Satya and Mark going on strong.
30:08 Alex Kantrowitz? Why does everybody hate Sam? Even Dario Amode, Amode from Anthropic was taking shots at him this week, you know, kind of in a veiled way, calling him a hype man.
30:17 Speaker 01 I mean, because I think the secret of OpenAI's success over the last few years has been delivering very good consumer products and moving fast, but also very, very, they are more heavily invested in the marketing and hype side of this, and that's how they raise money. And it's been, you know, a critical part of them maintaining some level of leadership in this. But the hype and the marketing are a core part of the OpenAI story. And in reality, like the Anthropic of the world get to ride the coattails of that a bit. So maybe they should be more thankful to Sam on this. But, I mean, you figure the Zuckerbergs and Nadellas of the world are like, okay, I'm
30:58 Speaker 01 running my trillion dollar company here that's actually, you know, delivering scaled revenue at unimaginable lengths and profits. And they just look at it and they're like, all right, buddy,
31:10 Alex Kantrowitz? I'm telling you, man, Anthropic is the one I'm most worried about this year. It has, it seems like the least amount of money on hand of the big ones. It's not developing, as far as we know, one of these massive data centers. Claude is like you mentioned, number 18 in Chatbot Arena, getting surpassed by DeepSeek in a crazy way. And it doesn't even connect to the internet right now.
31:32 Speaker 01 But Claude is good. Claude, the product.
31:35 Alex Kantrowitz? It's my favorite of all the products, so they can't discount it.
31:38 Speaker 01 Claude is good. Other than their rate limits, that sometimes I'll be deep into some kind of coding or other analytical project on Claude. And then I hit the rate limit and have to wait a number of hours, even as a paying customer. But fix that one, Claude, come on. But other than that, Claude's a good product. So I think it's, obviously their marketing has actually been pretty bad. They had these kind of comically ridiculous billboards that they tried to push around. And actually, I guess my, it's a great product for people who know about it. But I would agree that in terms of being in trouble, how ChatGPT is obviously the kind of household Kleenex Xerox brand name in the space.
32:18 Alex Kantrowitz? 300 million users.
32:19 Speaker 01 I'm sure most of my normie friends don't even know what Claude is. And are they going to be able to overcome that? I don't know.
32:27 Alex Kantrowitz? Yep. All right. One more wrinkle in this Stargate project, and then we'll move on to OpenAI's operator. And that is, and you know, if you're calling this fake news, maybe I'll take that, take the bone that you're throwing there and say, if it's fake news, why would it be fake news? Like, what would it be in service of? And maybe it is some marketing spin on OpenAI and Microsoft divorcing a little bit on the compute side. So the one thing that happened after this was that we found, we found out that Microsoft and OpenAI, which have been linked together from the very beginning, well, maybe not the very beginning, but linked together very, very strongly for the past few years and something that's benefited both of them. They're changing the way that their compute arrangement is.
33:12 Alex Kantrowitz? So Microsoft and OpenAI announced Tuesday, they've adjusted their partnership so that OpenAI can access competitors' compute. This is from The Verge. The new agreement includes changes to the exclusivity on new capacity, moving to a model where Microsoft has the right of first refusal. And then to further support OpenAI, Microsoft has approved OpenAI's ability to build additional capacity, primarily for research and training of models. So, Microsoft still has this proprietary ability to sell OpenAI's models through Azure. But OpenAI no longer has to go through Microsoft for compute, and that's what this new project is gonna do. And this has been something that's been bothering OpenAI for a long time. It's part of the tension between them and Microsoft, and maybe you sort of dress it up as Project Stargate,
33:57 Alex Kantrowitz? and then all of a sudden, uh, everybody's happy, and you hope that people don't notice. What do you think about that hypothesis?
34:02 Speaker 01 Yeah, I think, I think that's actually, everything about Stargate were things that were already in the work. Even Oracle was building a massive data center, and I think it's like Abilene, Texas, that now OpenAI is going to have an exclusive relationship with as part of Stargate, but they were already working on that. Apparently, the information had reported that Elon Musk was offered access to it, but had turned it down. So now it was open, so Larry Ellison saw the opportunity to move in this direction. And then even there is reporting that within Microsoft, there's increasing skepticism around generative AI in general, and specifically OpenAI, based on the negative reactions and rollouts for Copilot.
34:46 Speaker 01 So suddenly, within Microsoft, they still see the long-term value, but the idea that OpenAI is so magical, that that's gonna be a critical part of the success? And maybe they realize the product's more important than the model, that they start to, that's gonna, I'm gonna get a T-shirt with that, uh, that they start to realize like, okay, we don't need OpenAI, in fact, at all, that we've already been moving away from them. And financially, the potential of any kind of great return from them is dwindling. So why not let it go?
35:18 Alex Kantrowitz? Yep. Okay. Product more important than the model. Well, here we go. We've been talking about agents and assistants and computer use for forever. And OpenAI just released what is probably the best computer use model that exists today. So this is again from The Verge. OpenAI unveils AI agent that can use website on its own. On Thursday, OpenAI unveiled a tool called Operator that can go out onto the internet and perform tasks autonomously, like shopping for groceries or booking a restaurant reservation. It can navigate websites and take actions on websites, much like you and I do, said OpenAI product and engineering lead Yash Kumar. Artificial intelligence researchers call this kind of technology the AI agents.
36:03 Alex Kantrowitz? And this is some examples that Kumar showed. Booking a San Francisco restaurant reservation using OpenTable and then also buying a list of groceries through Instacart in the debut video they showed. They showed the thing going through, basically looking at a paper grocery list and then going out and buying the groceries. And then I also found this really fun one from this guy, Peter Wallinder, who is from Sweden, and he says he likes saunas. And he has Operator go through TripAdvisor and then basically search through all the hotels in Stockholm and try to find the best hotel sauna. And you just see the thing going through the websites and, you know,
36:47 Alex Kantrowitz? searching the dates and searching the hotels and then searching for the sauna and the reviews and then coming to a conclusion. And this is usually something that would take a human a lot of time to do. And maybe AI can do it for us. Real product news. Ranjan, your reaction?
37:04 Speaker 01 Real product news, and I'm actually genuinely excited because when Claude or Anthropic had released their Take Over Your Computer product, I think it was very limited in terms of ability to use and very buggy, people found. OpenAI, from at least the demos I've seen, looks pretty slick, looks pretty interesting. But I actually think the problem with this rollout, they need to do, say, agents. They need to say Agentic. That's just part of 2025. If you want to compete, you got to say agents. But The use cases in this, I just don't see what is going to actually achieve any consumer
37:45 Speaker 01 level scale with kind of computer takeover workflows and processes. Cause again, everyone is showing, okay, I can find and maybe make an open table reservation or buy. I saw one demo where someone, you know, find me a leather jacket in my size that can arrive in two days. And it technically worked. I don't see anyone changing their behavior to actually allow an agent to do this for your own personal life. And I think like the companies that are actually what they need to be focusing on. Again, this is something where I think Agentic wins in the enterprise, not in consumer.
38:23 Alex Kantrowitz? Well, to further that point, first of all, I thought you'd be more excited, but true to your intellectual honesty, you're calling it as it is. And to further that point, Casey Newton of Platformer paid the 200 bucks to upgrade to pro and played around with operator. And he did two things. He tried to get it to create a list of walking tours in London for him, which had canvassed the web and found, and then also a curriculum teaching the Great Gatsby, which had been something that the professor Ethan Mollick from Worden, who's been on the show before, I guess both of those folks are friends of the show. He's, he's done that before and used that as a benchmark. And it was able to do it sufficiently well.
39:03 Alex Kantrowitz? But then here's the crazy thing. Casey just asked ChatGPT the same questions that it had the operator do. And without having to do the fancy like go through the whole internet, it produced better answers faster. So is this a farce?
39:19 Speaker 01 Yeah, I think the interesting part of this should be that last mile of actually booking the walking tour. Like the research side of an agent is not that interesting. It's the can you actually enter my credit card data and actually click buy, which is pretty amazing and it should be able to do that. But again, from a consumer standpoint, the average person is so far away from letting an autonomous agent to actually spend your money or do things or make decisions around your time. So I think that's one of the most important parts. I think. Like when Apple Intelligence were gonna be discussing, like their view of Agentic,
40:00 Speaker 01 at least as they presented it, is more real for regular people around find my flight time, book me an Uber to get me to the airport in time and make that all one action. If it actually worked, that would be amazing. But I think like computer takeover Agentic AI for the average person is just so far like thinking about what would I use it for, trusting it to use it for that and having it more necessary than, as you said, just a traditional LLM search or query. I think we we we're just so far away from that. But I do like, I'm glad. OpenAI makes great products that actually work pretty well. So
40:43 Speaker 01 so there I'm glad they're still pushing this forward.
40:46 Alex Kantrowitz? Yeah, I thought, should I pay? Because right now it's only available to these pro users. You pay $200 a month, you get OpenAI's ChatGPT Unlimited, and you could also get this operator. And I said, should I pay this 200? And maybe I should, you know, for the love of the game, since I'm I am a reporter covering this stuff. But I just thought, nah, not right now.
41:04 Speaker 01 I mean, I do think I will recognize it is a big deal. Like, if you start rolling out this ability, maybe we'll start to get the genuinely creative, potentially repeatable actions that really and again, not research where you actually take action on different websites and define those actions. I think it's going to start to get pretty interesting. But a lot of this, again, you could do with a script. Like, you could do it with a script and it's basically creating that script like behavior for you, which is cool and which all empowers people who cannot code scripts. And that's a good thing. And the people will ideally start coming up with creative use cases, but.
41:47 Speaker 01 It's uh, I thi- I think they should actually put it, put it as part of ChatGPT plus guys. And I'll play with it. Maybe, maybe I'll be impressed. From the demos, it did not look that exciting.
41:58 Alex Kantrowitz? Maybe DeepSeek comes up with their own operator model than OpenAI.
42:01 Speaker 01 Would you let, would you let DeepSeek take over your computer?
42:04 Alex Kantrowitz? No, of course not.
42:07 Speaker 01 That's actually one thing, I mean, just on that, because, because there were some very on deeps, going back to DeepSeek, like very egregious demonstrations of censorship around sensitive topics in China that were very clearly baked into the model. And it is interesting because like on one hand, we are talking about how this could be completely revolutionary. But on the other, there's a lot of, I think on the trust side, on the security side, there's still going to be some questions.
42:36 Alex Kantrowitz? No doubt.
42:37 Speaker 01 Again, taking over, letting even OpenAI take over your computer. I feel behaviorally, the average person is so far away from trusting this stuff.
42:48 Alex Kantrowitz? Yeah, I, I totally agree. And it's something that I've been talking about on the show with everybody who's come on, whether it's Demis this past week or Eugenia from Replica the week before. There's going to be a real reticence to let people, to let computers take over and, and even to try, I mean, still to trust AI. You know, sixty percent of people don't really trust AI. You know, we're in like a little bubble where, where like, okay, yeah, we trust this to, to a certain extent. But most people are like, screw that. I'm not touching that stuff.
43:12 Speaker 01 I guess you just did mention the CEO of Replica. And you know what? Maybe people that won't trust a computer to actually browse the web for them, but they will trust in AI to be their lifelong companion and love interest and romantic partner. So.
43:27 Alex Kantrowitz? Yeah.
43:27 Speaker 01 That's where people are right now.
43:28 Alex Kantrowitz? It's all fun and game till your AI lover asks for your credit card.
43:32 Speaker 01 Yeah, I'll have an AI lover, but I will not give them my credit card info.
43:37 Alex Kantrowitz? Yeah, that's what you say is right now, but just, just wait till that relationship progresses. And next thing you know, you're buying a digital clothing and you're broke.
43:47 Speaker 01 That will happen and the New York Times will do like a 6,000 word reflective deep piece with large photographs of someone walking along and looking forlorn.
43:59 Alex Kantrowitz? He fell in love with an AI. She took his money.
44:02 Speaker 01 She did. But she spent his money on a shopping spree, in a Gentix shopping spree.
44:08 Alex Kantrowitz? Oh, my God. It's great business if you could run it. Okay, so, so, you know, you hinted at at Apple earlier and the current state of Apple intelligence is that like, I don't know, I saw Gen Moji ship, which is kind of interesting, where you can like prompt and get your own emoji or maybe it shipped just in the test version that I'm using. But by and large, Apple intelligence, you know, speaking of AI product, an AI product for consumer is not working. It's not working now. And who knows if it will work in the future. And it's nothing close to the presentation that the company showed at WWDC last year. And as a result, analysts are starting to say, hmm,
44:51 Alex Kantrowitz? we don't want to own this stock. So this is from Yahoo Finance, "Apple stock hit with downgrades, two downgrades on weak iPhone sales and AI outlook. So Jefferies analyst Edison Lee downgraded the investment bank's rating on Apple stock to underperform and decreased its price target by 13% to 275. On Monday, Loop Capital also downgraded Apple stock from buy to hold and revised its price target to 230, down from 275. They believe that iPhone sales in China are going to fall between 15 and 20% year over year in the fourth quarter. By the way, Apple earnings coming up next week. There's also been other downgrades. Moffett Nathanson downgraded a couple of weeks ago. And then everybody believes that that Apple intelligence
45:36 Alex Kantrowitz? is so underwhelming and has not driven a super cycle or any sort of cycle really around the iPhone 16 and are doubtful we'll do it over the next devices. And this is something we kind of previewed when we said, hey, we think Apple intelligence kind of sucks and it's going to lead to some bad sales numbers. And here it is. So is it victory lap time, Ron John?
45:58 Speaker 01 I, I'm gonna, I'm gonna take a moment here. As I, as I have to try to start to explain how bad Apple Intelligence is. And for newer listeners, I've had a long time, a long time hate hate relationship with Siri. And I say this as someone who has HomePods across their house, AirPods in my ears, an iPhone, a MacBook. So I'm still heavily locked into the ecosystem. But all I want is Apple to just deliver basic level generative AI in their products. Apple Intelligence, now I've been using it, I mean, it's been on my phone for a number of weeks now.
46:40 Speaker 01 And it misses at every single level. The more I've been thinking about it, the notification summaries are so useless. Like they, the, the way they will group in different notifications. If you get multiple ones from the same app and it will try to summarize it. It actually, it makes it more difficult than reading for notifications. Because there's the, in terms of how they present the information. Even, I was thinking about a Genmoji I played with. It was kind of fun. The UI side of it is very nice. Like actually, probably the only thing that's cool is now when you ask for Siri,
47:20 Speaker 01 your phone lights up around the borders of the, of the entire screen in kind of a pretty way. But I was thinking about it. It's the most tone deaf thing because the whole value of emojis are that it's kind of like a shared language. It's kind of a common thing where the emoji
47:38 Alex Kantrowitz? Oh, yeah.
47:38 Speaker 01 connotes something specific and we all have that shared understanding. So from a language standpoint, no one wants to create new emoji. I'm, I'm very curious about the utilization of it, but that misses the point. Siri working and with ChatGPT.
47:56 Speaker 01 Is okay, but it's so much worse than ChatGPT Voice or Gemini Voice or any actual other voice mode assistant for any generative AI product. So it's genuinely shocking to me. Like, a friend of mine was trying to pitch, you know, Apple should buy Perplexity. And I genuinely don't, I hope they don't, because I think they would ruin it in integrating it into the overall product and ecosystem. I don't, what do you think is happening over there?
48:24 Alex Kantrowitz? I've always said it's a cultural thing, and they just aren't set up for, to culturally build AI. They're too secretive. Their product groups don't work with each other. You need that, and you need to be risk-taking a little bit. And Apple is none of those. And therefore, I don't think it's going to realize this AI vision until it really is kicked in the face and been, and made to. And so we've seen these stock downgrades, and obviously, Apple cares a lot about its share price. And now Nvidia is the most valuable company in the world again, and Apple is, is not. And maybe something will, will spark within the company. But I don't think Apple today is going to be able to build an AI product that people are going to believe in.
49:03 Speaker 01 But, but do you really think it's cultural how bad it is? And I say that, that we, we are both very early adopters of, and test out every generative AI offering. Like Gemini Voice is really good. ChatGPT Voice Mode is pretty good. I was actually like, Gemini Voice, I've been incredibly impressed. I almost want to get an Android just so I can interact with it rather than Siri. I still am so locked into Apple that I have not yet. But the fact that I'm thinking about that is, is something that is very different than anything over the last decade. Like, do you think they can actually, even, let's say we see a bunch of downgrades, stock
49:47 Speaker 01 drops 15%. Do you think they shake things up and are able to actually deliver something just on par with what's already out there?
49:56 Alex Kantrowitz? No, I think they're more likely to just give up on AI. That's my perspective.
50:01 Speaker 01 Just no more, I mean, maybe, maybe.
50:03 Alex Kantrowitz? You can use ChatGPT on an iPhone. So it's not, it's gonna be a problem for them. They're gonna need, the thing, here's the real issue, is that so many iPhones are just kind of the same. Like, I have the iPhone 15. You could tell me today, Alex, you actually have the iPhone 14, or you have the iPhone 16, and I would believe you. Or you got the iPhone 13. I would be like, alright, yeah, that sounds about right, because there's not much difference. And the phones you can hold on to for three, four years. So like, those, those upgrade cycles are gonna get more spaced out, and there's no real draw to bring people in. So I think that like, it's less of a like, people are gonna switch to Android thing, and more of a like, the upgrades are gonna come in much slower.
50:48 Alex Kantrowitz? Now, people are still gonna be using iPhones, so they're gonna be able to buy their services, and they'll come in and they'll get the MacBooks every now and again. But ultimately, I don't think that they have what it takes to deliver AI, and I could be wrong. And I'll admit it if I'm wrong, but so far, I've been right. You could read it, it's in, it's in my book. I've always wanted to say that, but really, in the Apple chapter, I specify why they're gonna have a hard time with AI, and it's all playing out.
51:12 Speaker 01 Okay, I think, I guess, I, I almost would bet that they will win smart glasses as a category if they launch a product, or like, that's the area I could actually see them having more success in building, or like, launching like a genuinely revolutionary new product. But yeah, I, I think I agree. You don't leave the Apple ecosystem, you just don't buy the new iPhone. I actually think that would be a power move, though. If they just said, you know what, no more Apple Intelligence. We're not baking in any AI, like, at the device level. Use all the other apps, that's great, but we're over it.
51:48 Alex Kantrowitz? Yeah, they're gonna make their money through that, too.
51:50 Speaker 01 And we're gonna, we're gonna launch some good new devices.
51:53 Alex Kantrowitz? Yeah, Siri will suck and you will like it. I mean, that's gonna be their new motto. This episode is brought to you by AVPoint. Everyone's racing to roll out AI right now. Copilots, chatbots, agents doing real work. But here's the part nobody loves talking about. All that AI runs on your data and most teams have no single way to see it, secure it, and prove it's under control. That's exactly what AVPoint does. For 25 years, they've been the trusted layer beneath the world's most demanding data, now extended across your entire AI estate, your data, your cloud, and the agents acting on your behalf. It's how more than 28,000 organizations deploy AI with confidence.
52:35 Alex Kantrowitz? So innovation scales without scaling risk. AVPoint, the unifying trust layer for AI. Head to AVPoint.com to see how enterprises deploy AI with confidence. Learn more at AVPT.co/BigTechnologyPodcast. That's AVPT.co/BigTechnologyPodcast. Violent threats against executives are growing at an alarming rate. When a company experiences backlash, executives are the first people blamed. When their address and other personal details are sitting online, this backlash can lead right to their front door. The team at Ironwell knows this better than anyone. They've protected some of the most targeted executives and individuals on the planet for almost two decades.
53:18 Alex Kantrowitz? As a white glove enterprise security service, they protect your people with continuous personal data removal, proactive prevention tools, and human-led emergency support. So when someone goes looking for your executives, Ironwell ensures they hit a dead end. So here's what to do. Go to ironwall.com/BigTechnology, fill in the quick form, and request your free risk assessment. Their team will show you just how exposed your executives are and how to lock it down before a threat reaches their front door. That's ironwall.com/BigTechnology. Learn how to stop online threats before they become real-world attacks. The link's in the show notes.
53:57 Alex Kantrowitz? Okay, before we go, we should really talk just for a couple of minutes about what happened with TikTok because at last week's episode, I said I would be stunned if we got to this week's episode and TikTok was gone and TikTok is indeed still standing. However, I don't think you can find it in the App Store and every company that, every company that has TikTok is supporting TikTok, the oracles and the Akamai's of the world, they risk fines, daily fines and potentially really go into lots of money and millions or billions of dollars over time for keeping this thing running despite the fact that Trump has quote unquote "saved" the app with an executive order. Just very quickly, what's your perspective on what happens to TikTok from here?
54:42 Speaker 01 You know what? It's Friday, January 24th. This has been such a heavy week of news that I almost forgot that whole TikTok ban thing happened on Sunday, I believe, late Saturday night, where everyone got a message saying TikTok is shut down and President Trump hopefully will save us. And then you got a message, I think eight hours later, President Trump says "save the app", which was just the wildest thing imaginable. Like that happened. That happened six days-
55:15 Alex Kantrowitz? Some girl allegedly lit a congressperson's office on fire in that 12 hours because she was mad about TikTok.
55:23 Speaker 01 Like, it was interesting. I was asking a bunch of younger folks who are deep into TikTok and it was interesting because a lot of people's reaction was genuinely like, "Do they really think we're this stupid?" And I'm like, "Yes, yes, I do think the executive leadership team at TikTok does." But yeah, in terms of what's going to happen going forward, this one is going to remain one of the most wildly unpredictable things and very excited to watch what happens because again- Finding an American buyer and divesting TikTok was all from ByteDance was always
56:04 Speaker 01 the necessary request. Now we still have to find one. It's always been questionable about who is going to buy at what valuation because let's say we find people willing to buy, still negotiating because it's still at a very, very high lofty valuation. People, that negotiation is going to be interesting. And then obviously the twist that somehow 50% of it has to go to the US government, I don't think is, makes any sense, or I just don't quite understand exactly what that part of it means. But this one, I have no idea what's going to happen, but I'm very excited to watch.
56:45 Alex Kantrowitz? Ron John, can I just say that this week I expected us to be talking about TikTok ban and all the big tech leaders standing behind or sitting behind Trump.
56:55 Speaker 01 Oh, wait, that was Monday, wasn't it?
56:56 Alex Kantrowitz? That was Monday. Whole script has flipped. There's been so much news, even Stargate, which I thought was going to be the lead story, ended up being pushed down, down the list of stuff to talk about because of DeepSeek. And then of course, we got operators. So, and the Apple downgrades. Fascinating week of news. Appreciate you all for sticking with us and coming back week after week. We break down the news here, yeah, on Fridays. And Ron John, thanks for so much for coming on the show.
57:21 Speaker 01 Who knows where we'll be in seven days. I'll see you next week.
57:24 Alex Kantrowitz? I could deal with a little bit less unpredictability, but anyway, we will be here next Friday, by any means necessary. We will be here. All right, everybody. Thanks for listening and we'll see you next time on Big Technology Podcast.
58:08 Speaker 02 Every mile matters. Whether you're driving your dream car or your daily commuter, the MobileOne team knows that. It's why we created MobileOne Advanced Clean Motor Oil. It cleans your engine in one oil change and helps extend engine life, delivering performance, protection, and cleanliness for vehicles of all ages. All to help keep your engine running like new. Whatever you're driving. Visit mobileone.com/advancedclean to learn more.
Previous Google DeepMind CEO Demis Hassabis: The Path To AGI, Deceptive AIs, Building a Virtual Cell Next Bonus: The DeepSeek Reckoning in Silicon Valley Page updated October 12, 2026 · transcript and summary are machine-generated