Pod Insights Big Technology Podcast DeepSeek's Fallout For AI Companies, OpenAI’s Path Forward, Siri Somehow Got Worse About this episode Ranjan Roy from Margins is back for our weekly discussion of the latest tech news. We cover 1) The DeepSeek impact on Silicon Valley 2) The four areas of margin in AI 3) How OpenAI is positioned after this week 4) Whether OpenAI can be fine losing the lead on model building 5) DeepSeek's impact on Anthropic 6) Why Amazon is happy about DeepSeek 7) Should NVIDIA be taking the worst of it? 8) Okay, let's discuss Jevon's Paradox 9) We have a discord now 10) Apple made Siri even dumber
---
Join us on Discord
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 10,483 words · 4 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Alex Kantrowitz? Let's assess the fallout of the DeepSeek moment and how it changes things for OpenAI, Anthropic, Nvidia, and others. Plus, Apple makes Siri dumber and iPhone sales are falling. We'll cover it all 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 Agentic 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 a large-scale
0:42 Alex Kantrowitz? 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.
1:20 Alex Kantrowitz? Welcome to Big Technology Podcast Friday Edition, where we break down the news in our traditional cool-headed and nuanced format. Well, last Friday, we spoke with you about DeepSeek, and over this past week, it seems like it's really changed the entire conversation in AI. So, we obviously covered it on Friday. We had MG Seeker come in for an emergency podcast on Monday. Today, we're going to pick up that conversation and talk a little bit about, now that we have a little bit of distance from the actual news itself, what it's going to mean for almost every tech company in the AI game today. And then we'll also talk about Apple and how Siri has become even dumber with Apple Intelligence. Joining us as always to do it is Ron Jon Roy of Margins. Ron Jon, what a week it's been. Welcome back to the show.
2:05 Speaker 01 I can't believe the DeepSeek sell off was only five days ago now. These weeks, these weeks are getting longer, I think.
2:12 Alex Kantrowitz? That's crazy. And it's very interesting that Nvidia was the one that was hit. I guess Nvidia has been a symbol for artificial intelligence and maybe the AI bubble, if you think it's a bubble. But there are some others outside of Nvidia that I think it's worse for. And we can talk about that a little bit today. So let me just kick it off because I about to publish a story on big technology, talking a little bit about the four areas of margin in AI and where this week's news leaves OpenAI. So I was speaking with a source this week who's working in the industry who basically put it this way. You can get margin in four areas of AI business. There's hardware and Nvidia is taking all of that. There's data
2:54 Alex Kantrowitz? centers. So you have Azure and Amazon's AWS and Google Cloud taking all of that. There's AI model building where there's not really a margin, especially which we see this week after DeepSeek has basically shown that we're going to be on a curve where these models will just get cheaper and cheaper to run. And if you're licensing them from an OpenAI, it'll get about as close to free or the energy that you need to run them as possible. And then there's applications and that's the ChatGPT and the replicas of the world that use this technology to build. And so basically, the assessment is the biggest takeaway from this week is that if OpenAI thought it was running an API business or a model building
3:34 Alex Kantrowitz? business before this week, right now, there's clarification. It's all about the applications. It's all about ChatGPT. Good news, it has the leading application in the AI application world, but it certainly shakes up the broader picture for the company. So Ron, I'm just going to put it to you. I'm curious how you react to all of that and where you think OpenAI is after this past week.
3:54 Speaker 01 Well, my first reaction is I'm very glad that I've been ranting that LLMs will be commoditized for I think about a year and a half now. And now it seems to be the conventional wisdom, the fact that we're starting the conversation here saying that models will provide no margin versus the idea that actually that's where the like insane profits would be. That wisdom, I mean, that's completely flipped on his head. Now it's almost the assumption after R1 and DeepSeek that that there will be no economic incentives for from the model side. It's interesting how you put it though, because remember OpenAI in the financial projections
4:36 Speaker 01 that leaked, I believe it was going to be 80 or almost 85% of the revenue was supposed to come from the application layer, the consumer subscription side versus Anthropic is actually one that gets even more concerning here because they were betting on the API layer, the model layer, the fact that people would pay them lots of money for access to the models directly. So even in this framework, it almost bodes worse for Anthropic than OpenAI here.
5:08 Alex Kantrowitz? Yeah, I think that this was really vindication week for us on a couple of fronts. You sent the product as all that matters and me whispering over the past couple of weeks that Anthropic might be in bigger trouble than a lot of people are anticipating. We'll cover it. Dario Amode had a really interesting post about this DeepSeek moment and we're going to cover that in a moment. But you know, it's interesting because I'm actually leading my story off with the revenue numbers about how ChatGPT was supposed to always be the lead for OpenAI and not the API. But there's one thing I can't square with that. Like you would think that this would be total vindication for their strategy and like actually a good thing for them. And maybe they might even eventually use different models within ChatGPT if we end
5:53 Alex Kantrowitz? up in this world where everything is commoditized, which is something I hadn't considered before, but I'm going to mention in the story. But the surprising thing to me has been how flat footed OpenAI has been in their response to DeepSeek. Now, obviously it's a moment that will probably shake you if you're a model builder because you've been not, let's say surpassed, but you know, effectively equaled by an open source model coming out of China that is so much cheaper to run. And we talked about it last week, that the big deal. In this moment is not that it took less money to train and that's obviously under question. It's the fact that you can use the API for much cheaper. And I was speaking with a developer this week who said basically it's $60 per, I think,
6:33 Alex Kantrowitz? million tokens to use the OpenAI O1 model and it's $2.19 to use the DeepSeek R1 model, both comparable reasoning models. And it even goes up to, let's say, $10 if you're using an API provider that's going to make it a bit stable. But it's so much cheaper. And then you look at like, what has OpenAI been doing these past couple of days in terms of its messaging? And it's been terrible. And they've been the ones that have been so good at the PR side of things. I'm just gonna talk a little bit about what Sam Altman has done. So first of all, he says, next phase of the Microsoft OpenAI partnership is going to be better and much, it's gonna be much better than anyone is ready for with some awkward selfie with him inside Inadella. Then he talks about, addresses DeepSeek R1
7:18 Alex Kantrowitz? directly. He says, DeepSeek R1 is an impressive model, particularly around what they're able to deliver for the price. We'll obviously deliver much better models and it's also legitimately invigorating to have a new competitor. We will pull up some new releases. And then, he says, looking forward to bringing you all AGI and beyond. I mean, that to me reads as sort of panic and shock from Altman. And I wouldn't think it would be the case if it is, as we're all agreeing, an application world at the end. What do you think?
7:47 Speaker 01 Well, so separating it out from the application side, the most important part of his statement, I think we should get into because there's been a lot of controversy around it is, he makes sure to say and believe more compute is more important now than ever. I think that's the central question for the entire technology industry. It's underlying the entire Nvidia story. I mean, I've seen so many sell-side analyst notes, you know, where the clearly there's a heavily invested interest in Nvidia and some cost into Nvidia, where people are just, you know, just, you know, hitting the bandwagon saying, This actually means there's going to be more compute than ever needed.
8:31 Speaker 01 And Sam's saying it too, because this was the week, remember Stargate was the middle of last week. The idea that investing $500 billion would be required. And then suddenly DeepSeek comes along and the initial chatter is it's only 6 million. And then there's been plenty of reporting around originally the Chinese quant hedge fund. I think I saw some number in the F T. It was like I had already had around like 1.2 billion dollars in GPUs. Invested and so basically, and we all know that the $6 million cost attributed to it was only on the inference side of R1. So overall, it was not $6 million to actually create the R1 model.
9:16 Speaker 01 But overall, it showed us it can definitely be done for significantly cheaper than 500 billion. And overall, the amount of the compute no longer means victory. And I think that is a fundamental shift that Sam Altman and OpenAI have gone so far down the road of those with the most compute will win. That's been Mark Zuckerberg's attitude, even Satya Nadella has pushed this kind of thesis. So to me, no one is still, everyone, the entire industry, at least out of Silicon Valley, is still trying to hold dear to this idea that the more compute you have, you will win.
9:58 Alex Kantrowitz? And so do you think that's why OpenAI has been so shaken this past week? Because basically they were going to have that compute advantage. And now, you know, maybe it doesn't matter as much. I mean, we're going to, obviously, we're going to get into Jevon's paradox as we're required.
10:13 Speaker 01 We have to.
10:14 Alex Kantrowitz? I mean, obviously, this is a contractual thing. We're going to get into that Nvidia section and the Microsoft section. But like, let's focus on OpenAI for a minute. I mean, again, going back to our question here. If it's all about ChatGPT and that is the leader and they've shown that they've been able to innovate with things like Operator, which I think we're going to get your thoughts on in a moment, and ChatGPT Voice Mode, which is pretty cool. And they have, they've gone from, let's say, 100 million to 300 million weekly users of ChatGPT in, you know, in just a year, which is massive growth and shows that there's real promise for the app side of AI. Even though I'd like it to be bigger. But anyway, that's the story for the day.
10:57 Alex Kantrowitz? So what's the big deal? Again, if, you know, this compute advantage or this compute
11:05 Alex Kantrowitz? lead doesn't make a big of a difference, just as long as AI can surge forward.
11:10 Speaker 01 Well, that's exactly why this is their reaction is so fascinating to me because I have definitely argued that OpenAI's greatest asset has actually been their product, their product jobs, that ChatGPT, the UI behind it, the like overall experience, their voice mode, incredible products. Operator, which I have played with a lot. I spent probably like 12 hours last weekend trying to use, is not a great product. It's an interesting product. It's not a great one yet. But, but they've had an advantage on the product side. So on one side, it's odd to me that they don't just kind of say, you know what, we're going to win on product and we're going to have great models behind it. But then you, it's kind of when you go back and remember
11:55 Speaker 01 at their core, which Sam Altman has said many times, they're a research house that kind of accidentally walked their way into a business. And at that point, you can almost picture the ego side of it, the, the kind of like more competitive side that they don't want to lose that battle. They don't want to lose the, the actual foundation model creating who's going to get to AGI first. They care more about that than hitting their $340 billion valuation, which there was reporting that SoftBank will potentially invest in this week. 25 billion out of 340, I think. So, so they're still, I really believe Sam Altman cares more about having the better models, getting to AGI faster
12:41 Speaker 01 than the business side, because this doesn't fundamentally change at least what's been reported around their business plan.
12:48 Alex Kantrowitz? Yeah, so let me see if I can do my best to sort of tie it up now that we're talking about it. I'm thinking it through. I think it's important for OpenAI to lead in models because if you lead in models, you can dictate where the product goes. And honestly, that's why, you know, everyone wants to be able to do that. And when you have open source, in particular, open source, come and maybe not grab the lead, but basically show that it can play in your ballpark, then you're in trouble because open source is a sort of swarm of developers. It's DeepSeek that's going to build on Meta and Meta that's going to build on DeepSeek and all the other developers that are going to customize and push forward. And if they are, if open source is the one that leads over proprietary,
13:31 Alex Kantrowitz? you could end up seeing the product game potentially spin out of OpenAI's control. Just to throw out like a crazy example, like let's say open source gets to AGI before, before OpenAI does. Well, then, you know, the products that are going to be built on that will be, you know, by definition, better than the chat GPT. So if you're OpenAI, it's also like, even if your model business is not going to be the breadwinner, really is incumbent upon you to be able to lead this race. And that's why, you know, you would, we're going to talk about Meta in a moment, but you would think that Meta is devastated by this moment because they've been, they've been sort of outflanked by DeepSeek. But it turns out like actually the strategy, you know, might be proving true.
14:13 Speaker 01 Well, no, I, I think that this idea that a bigger model or better model means a better product is the underlying thesis or thought process of a lot of Silicon Valley. But again, I disagree with it and let me get into Operator. So Operator is OpenAI's new kind of like browser take, computer takeover browser. It has a built in browser, but it can do things and take actions across the web for you. So I was doing a test. I've been, I'm working on a medical innovation fund as a nonprofit, and I've been having to kind of go through a lot of medical research papers.
14:54 Speaker 01 And I'll manually pull all the authors' names into an Airtable, and then look up, you know, like LinkedIn profiles, Google Scholar, PubMed profiles. So it seems like a pretty straightforward, this whole world of Agentic that we've been promised, seems pretty straightforward. So I went to Operator and I paid the 200 bucks for ChatGPT Pro. And I kind of, it is really cool. Like, it's actually like, you give it a prompt and you say, "Here's a list of papers. Go find the authors, put them into Airtable, then go to Google, search the author's name, find these links and paste the URLs back into Airtable." Seemingly like, not an overly complex thing.
15:39 Speaker 01 It worked two or three times really well, and it was mesmerizing because you're literally watching it and it's a built-in browser in the Operator interface. And like, you're watching it do this and just click around and actually copy and paste links. So you start to see the promise. Then, the most fascinating part was, it got lost in Airtable. Like, it started just clicking on random things. You can almost see it just like, losing its place and losing its flow and rhythm. And just, it completely broke. So then you take control, you can actually kind of like assume control back in the browser, try to redirect it, and I could not get it to work again. Or then it started copying in completely incorrect things.
16:23 Speaker 01 Like, it just does not work even close to where the promise is. But you see where it could be going. But again, this is the product they have released, Sora, very similar. Huge promise. We waited like a year and a half. Sora is available for, I think, all ChatGPT Plus subscribers right now. I have not seen anyone posting cool demos or all those kind of things. So, so the last two products that they have launched have been pretty much flops. So I think that's why. They are still working under the assumption that operator, if the model development becomes so good, then that's the only way this kind of like browser takeover behavior will work. So they're still betting that the model will solve everything.
17:08 Alex Kantrowitz? This is why you need better AI, don't you think? Like better models will make sure that operator works better. And in fact, they were asking Kevin Weal, who's the head of product, what's coming next for OpenAI and what kind of moat does OpenAI have? He says, we're about to launch some models that are meaningfully ahead of the state of the art. And some people are speculating that's O3 or who knows? Maybe we'll see GPT-5 finally. So don't you think these two work in concert?
17:37 Speaker 01 Yes, no, it's tough because on one hand, the coolest part of the experience, and again, I'm going to give full credit. It was a really, really fascinating kind of like, holy shit, this is the future moment. So I'm still going to give them credit on that, just the way the entire overall UI and just experience of it. But still, as you're using it, you're like, is this really the way I need to set this up? In reality, I could have written a PDF parser that did the same thing using AI. And Claude, I do this stuff regularly, extracted authors' names, done, written some, like this is, it was basically web scraping. So the web is such a messy thing
18:20 Speaker 01 that how many people are really going to derive value from this kind of tool? So it actually made me question, like, is this the way? And I saw some posts as well around like, could websites change the way they're structured? So they're more open and easy for an agent to access? Maybe that starts to be an interesting thing. But the idea that we need a model that's so smart that it'll actually understand the first time how to use a pretty heavy tool like Airtable, I don't think we do need that. I think the way this, this whole battle will be, like, won will actually be around. You know, creating an airtable connector that's already just pre-programmed and not LLM
19:04 Speaker 01 based to actually do this kind of work. So, so I don't think, I, I, I still embarrassed that there, we're going to see some such a smart model in the next few years that'll actually get the complexity and like of the entire web. I, I don't see that happening.
19:22 Alex Kantrowitz? It's one of the interesting things about looking at DeepSeek with its reasoning model is it will show you its chain of thought. And you can actually see the model kind of struggle its way through trying to find exactly what you're intending with your prompt. And so like, there was this concept of like, oh, the prompt engineer is going to be the new job. And it's already getting to the point where you don't need to write the best prompt because the models will sort of get to that point eventually. And I think that is sort of why being able to watch that within DeepSeek's consumer app is the reason why it ended up going so crazy and taking over the top app spot above ChatGPT and the rest. And in fact, I think that's probably like the biggest risk for OpenAI is like, who
20:05 Alex Kantrowitz? cares if the model's a little better or a little cheaper. And we can argue about whether it was, but the fact that the DeepSeek app went to the top of the app store, that to me might be why they had such serious concerns. And before we move on, I do want to move on to what Dario Fmentropik said. Just want to talk about a couple of things about the OpenAI business. So no one paid attention to it this week, but there was a report that OpenAI has now made four times the amount of its nearest competitor on the ChatGPT app since it launched, that 529 million. It's not a massive number, but this group called App Figures now estimates that the AI app market is $2 billion this year. And as things like operator get better, that's going to be even bigger, an even bigger number.
20:50 Alex Kantrowitz? So we're starting to see this AI app market materialize even again, if it's not as big as I would want it to be. And then it sort of leads into this next part, which Rontan, you've already brought up, which is that OpenAI is trying to raise again at a crazy number. They are in talks, this is according to the Wall Street Journal, to raise 40 billion in the funding round that would value them.
21:12 Alex Kantrowitz? As high as 300 billion. Yes, the 340 billion number was tossed out, but I guess they've come back to earth a little bit and now it's looking like 300 billion. I mean, late last year, OpenAI raised 6.6 billion at 157 billion valuation. And we said, right, what did we say when that came out? Oh, they're probably going to need to raise again at the end of the year. 2025, they're already raising in January.
21:37 Speaker 01 Oh, my God, I loved this. You couldn't give me enough news this week, but then Masayoshi Son had to come in and just push the idea of instead of, you know, the world is like, is OpenAI finished? And he said, no, let's come in and add another 15 or 25 billion. Add bring in 40 billion and pump this from up to 300 billion. But I mean, it's, it's amazing. It's, it's, there's not much you can rationally or analytically say about this. All you can do is just sit back and enjoy that Masayoshi is once again right in the center of things.
22:17 Alex Kantrowitz? I just found a great quote from Masayoshi. I guess he said this last year, he was talking about Gates and Zuckerberg. He says, "Those are one business guys. Phil Gates just started Microsoft and Mark Zuckerberg started Facebook. I am involved in 100 businesses and I control the entire tech ecosystem. These are not my peers. The right comparison for me is Napoleon or Genghis Khan or Emperor Qin, who's the builder of the Great Wall of China. I am not a CEO. I am building an empire." What a guy.
22:47 Speaker 01 I, is that from, so Lionel Barber, who was the editor in chief of the FT when I was there. He just came out with a book. It's called Gambling Man: The Secret Story of the World's Greatest Disruptor, Masayoshi Son. Is that where this, because I saw this quote everywhere and I was trying to find. Like the actual where it was originally from.
23:08 Alex Kantrowitz? Yes, it's from that book.
23:09 Speaker 01 Okay, I'm gonna, I'm gonna buy that right now, Lionel. I'm, I'm reading it. Yeah.
23:14 Alex Kantrowitz? One last thing, can we put to bed this idea of the subprime AI crisis, which we talked about last year? I'm very curious what you think about this. This idea that these companies had raised, you know, so many billions of dollars that startups were going to put their technology in their products. And then they were going to depend on it. And then eventually those rates were going to have to go up and kill the startups and kill the entire AI world. It was a very interesting theory. But now that we, we've seen that rates will go down to just about zero to use this technology. Do, do we still believe in the subprime AI crisis?
23:48 Speaker 01 So again, it was a really interesting theory from Ed Zitron, the idea that, yeah, startups that are building, the cost will only escalate the actual input costs of these AI models. So they're subsidized now. At some point it'll catch up and it'll bring everyone down. That's why this is such an exciting moment. To me, that's the biggest news of the week. Separate from the financial markets and the, you know, considerations on what this means for Nvidia. Everyone out there now has access to very good models cheaply. And that means that the product side, the application side, this is, this is like Andreessen's time to build.
24:27 Speaker 01 Like, let's get out there. Everyone start tinkering with everything and not having to worry about getting some inflated API bill. Like, this is it. This is going to happen and that this is going to keep happening. And now the entire industry is going to start thinking about not funneling all the investment dollars, other than Masayoshi Son, to the model builders. But to me, you can imagine there's going to be hundreds of new startups where capital will flow to, that will actually be more on the application layer and actually solve problems and be interesting for. Regular business people are just regular people and not just OpenAI making a few products
25:09 Speaker 01 and us all being stuck with them. So that's why, I mean, this to me was such a big week, aside from all the actual, I don't know, market and global geopolitical implications. But it just means this is where the world is going in the next year or two. And that that's far more interesting to me.
25:26 Alex Kantrowitz? Yeah, when we let off our show last week, I was like, Ranjan, what do you think about DeepSeek? And your immediate reaction, not thinking, was, I love it.
25:34 Speaker 01 I still love it. I still love it.
25:36 Alex Kantrowitz? And what we're saying right now sort of plays perfectly into this piece that Dario Amode, the CEO of Anthropic, put out where he talks about on DeepSeek and export controls. He says, and this is sort of going to our point, the efficiency innovations DeepSeek developed will soon be applied by both US and Chinese labs to train multi-billion dollar models. These will perform better than the multi-billion dollar models than they were previously planning to train, but they'll still spend multi billions. The number will continue going up until we reach AI that is smarter than almost all humans at almost all things, which I think he predicts is coming in a couple of years. I think his post was excellent and it also is like I kept reading posts by like Dario Amode
26:20 Alex Kantrowitz? and Andrew Ng and Yann LeCun. And I was just like, oh yeah, like this is real. And it's going to spread into the other models, basically making models that are much, much cheaper to run, which should spur that, you know, spark of building. And the only thing is, like you just said, this might not be good for proprietary builders. So I was interested to see it in Dario's post. What do you think's going on here?
26:47 Speaker 01 Reading the post, and again, it was like a very good in terms of him talking about the, what the actual like genuine innovation from the, from DeepSeek's side around their models were and recognizing there was genuine innovation happening. But the push on export controls, I think, I mean, and I'm curious your thoughts on this. It's, it's such a difficult question because on one side, is it going to be enough? The fact that DeepSeek Open source this and I mean literally it's on AWS now and like it's not some like only in China thing. It's not that kind of like bifurcated world of only like the
27:31 Speaker 01 West versus China. It's more integrated than that and it did it. To me, one of the interesting parts is, and again, plenty of debate around what the actual GPU count is for the Quanta hedge fund that, you know, spun this out. But overall, it does feel like, or there's been plenty of arguments for the idea that because of export restrictions, it almost forced the additional innovation and efficiency, like because they still did not have unfettered access to the Nvidia state of the art chips that they had to, they had to innovate around that, which is really interesting on already. But what does that mean about how we approach
28:16 Speaker 01 this? And especially of all people anthropic, a model builder whose entire financial forecast is predicated on selling models, like to kind of lean into this is a national security thing. And I'm always, I'm very pro ban TikTok and then other areas, really cognizant of the, you know, like national security implications of a lot of this. I am curious, like, they've already shown us it's not effective anyways. So I don't know. I think it was a little self-serving.
28:48 Alex Kantrowitz? Oh, definitely. And this is like, going to go to like the practical theory of how CEOs act, which is in their own self-interest.
28:54 Speaker 01 Yeah, fair.
28:56 Alex Kantrowitz? You know, it was like, very interesting to like, see everybody on X reacting to what Dario is saying by saying like, this is an unprecedented level of cope and he's just trying to hamstring, you know, the Chinese companies now that they've, you know, released a recent model and he doesn't have one and he's just upset about this. Like, as if you're supposed to do anything else as a CEO, then try to look out for your, your own interests. I don't think it looks good for him, but I don't think it really matters.
29:22 Speaker 01 Well, Dario, I can just beg you, now that we all have access to cheaper models, can you please make Claude the paid subscription level not hit my rate limit when I'm halfway through a project? Because it's one of the most frustrating things. So.
29:38 Alex Kantrowitz? There's going to be some problems there. Maybe they're just going to try to sell themselves now that I guess acquisitions are legal again in America.
29:46 Speaker 01 Uh, yes, that's, that's actually, I mean, back on the political side, I think the whole changed M and A environment actually is really worth watching with a lot of these companies because there was kind of this assumption, I mean, over the last couple of years that especially once you're hitting these like many billions of dollars in valuation that M and A is essentially off the table. So that'll, that'll be pretty interesting, but also, I mean, and we haven't even mentioned Google in all of this. Google has been making some pretty big strides, at least on the, the model building side, even on the application layer side. So you, then you start to get into which company really needs an anthropic. Again, Claude's a great product, but at what price?
30:32 Alex Kantrowitz? Yeah, I mean, maybe it'll be a capitulation sale to Amazon, but even Amazon is in good shape this week after Deepsea had put DeepSeek in its product and Andy Jassy bragged about it. I was planning to talk about it later, but it's worth talking about now. It just, you know, it took a week and it's there. And I think all in a, in a minute, it is the best model in AWS. And that's going to lead to more building through AWS. So I think that they're probably doing jumping jacks over at AWS HQ.
31:00 Speaker 01 Yeah, remember we, we had the theory that Amazon was gonna Amazonify this whole space by like bringing you into the AWS ecosystem, offering you the fancy expensive models. Like actually they were gonna deep seek with their Amazon Nova, their, their new foundation model suite, which was supposed to be winning on cost. That was kind of our thesis anyways, that they're, they're gonna just bring you in, let you use the high-priced stuff and then give you the Amazon basics version. And now they got DeepSeek already. So, so I think they have actually played this pretty well then if they haven't invested and have just kind of been sitting to the side.
31:40 Alex Kantrowitz? Definitely. And just it's telling that it was Jassy that was out there tweeting about it yesterday. It's clearly important to the company. All right, one last thing on Anthropic. This is from Hugging Face co-founder Tom Wolfe. He says, "If both DeepSeek and Anthropic models had been closed source, yes, the arms race interpretation could have made sense. But having one of the models freely available for download and with detailed scientific report renders the whole closed source arms race competition argument with artificial intelligence unconvincing, in my opinion. Here's the thing, open source knows no border, both in its usage and its creation." And I think that like, to just put a point on the first half of this conversation,
32:24 Alex Kantrowitz? it is, this has just been the week where it's been the open source week. And good for open source, good for the companies that benefit from open source, like, let's say, the Metas and the Anthrop and the Amazons. Bad for the proprietary model builders. And it brings us right into the Meta thing, actually, which is like, a lot of this commotion over DeepSeek was, began when there was a screenshot passed around of like, a blind post saying Meta engineers had been freaking out about the fact that DeepSeek had surpassed their open source models. And there's like, now reporting that top executives within Meta are worried that the next iteration of Llama won't perform as well as DeepSeek.
33:07 Alex Kantrowitz? But ultimately, I think that as they digested it, people like Jan LeCun, who's going to be on the show in a couple of weeks, and Mark Zuckerberg were basically like, yep, it's gonna be open source. We will take what they did and put it into our models. And we're gonna ultimately, you know, be in the right place, which is that we're gonna help develop AI that's available to all. It's gonna be on our terms, as long as we can keep our lead. Right? That's the big question. And then we're gonna benefit.
33:38 Speaker 01 Well, this is one thing I've not been fully clear because like if we're talking about it, does the value accrue to the model layer or the application layer? Meta is in a position where they should probably be able to do some pretty amazing things at the application layer. One, you have distribution that absolutely no other company or nothing in the history of mankind has had in terms of your billions of users that you can push products on to. And Meta AI, obviously already, I mean, image generation, it's fun to play with. In the Ray-Bans, it works incredibly well. Even the real time language translation is getting
34:19 Speaker 01 pretty good. So, so getting good products in front of people at that point, I, I do wonder, why do they care so much about having the winning model? Because in reality, they can build it with DeepSeek if they want. I mean, they'll, they're building with Llama right now, but, but what the massive incentive is. Originally, I remember thinking like it was all done in an awesomely like, you know, in an amazing way to undercut the OpenAI's and the, all the other paid players. But I, it's, I'm still a little bit confused on that.
34:57 Alex Kantrowitz? I mean, I do think, like I said earlier, that everybody wants to dictate how this goes. So they kind of have an idea of where it's going and they can lead like the product development and have people that are developing on their tools and making it better as opposed to somebody else's. You really want to be in the pole position there. But it's again, it's not the end of the world for Meta if DeepSeek just kicks their butt and they're like, all right, well, we'll just put this into Messenger and maybe people will use our chatbots then, which I don't think they use right now, even though they're talking about how they want the billion user assistant. I don't think anybody is, maybe some people, but top of mind, the fact that like Meta has an assistant is still not really competing with ChatGPT as far as I can tell.
35:37 Speaker 01 I actually, I use it for like quick fun image generations, my son, like just because it's right there in WhatsApp search. Like I actually, I'm convinced that probably a large percentage of Their AI assistant usage because it's built into the search bar essentially is people accidentally trying to search for something and ending up using Meta AI. But it's actually, it's good. I mean, it's not, I've never spoken with anyone who's using it as a more dedicated chatbot or chat assistant like most, most of us use ChatGPT and Claude and others. But it for like quick one-off questions, it's, I feel it's again, the distribution side of it.
36:22 Speaker 01 It's right there in the apps that people are in all day long anyways. It's, it's in terms of pole position, I think it's, it's definitely has an advantage.
36:31 Alex Kantrowitz? Yeah, and it's just going to get better over time as, as let's say, you know, let's say these open source models do achieve AGI, right? To just go with the crazy example. Well, you can then deliver that to a billion users and you don't have to rely on OpenAI to do that, or billions of users for that matter. There's, there's still a lot of let's, you know, what if though, on this front? And let's quickly talk about this Jevon's paradox thing. So this was Satya Nadella's reaction to it. He says, as AI becomes more efficient and accessible, its adoption will soar, transforming it into an indispensable commodity. I think we mostly share his opinion on that. I think you definitely do, Rajan. But my only question is, if AI gets exponentially cheaper, you know, then
37:14 Alex Kantrowitz? will we see an exponential rise in applications? Because on the application side, like we have ChatGPT, we have these meta bots. Uh, then we have some stuff in enterprise software like Salesforce as agents. And then of course, there's coding applications. But after that, for as powerful of a technology that AI is, I don't know if we've seen the applications that I would anticipate.
37:35 Speaker 01 Well, let's start with, so Jevon's paradox is not actually a paradox where it's not a logical contradiction. It's just kind of, it was a bit counterintuitive at the time. So that's been frustrating me, I think, more than anything. Because it's just completely sounds like one of those things that people have been using because it sounds smart. But we're still going to have to talk about it. And it's the idea because everyone is. And again, to me, it actually is intuitive. As any kind of resource becomes more efficient and accessible, it'll actually grow in overall aggregate demand and output because people will figure out more uses. And, and no, I mean, I think at the core, this is it. This is what we've been saying. And this is now the
38:19 Speaker 01 Thousands of startups building interesting apps on top of it. And finally, the promises that we've been made for so long and have not been realized will be realized because people will actually pay attention to building cool experiences and products using AI. It's not just Microsoft's of the world and OpenAI's of the world and only a few. And then others trying to build stuff, but being really limited based on, you know, the actual cost input side of it. So I think it is real. It's like the idea that AI is going to become much more deeply integrated into everything we do in a good way, I think is going to be real. And this makes it a lot more real.
39:04 Alex Kantrowitz? So do you think that this is unfairly punishing Nvidia? Or do you think it makes sense because there's been more uncertainty that's been added in? Like we were, the AI industry was moving on like, I don't know if a linear path is the right way to put it, but on a path. And now the path sort of changed directions.
39:20 Speaker 01 I think, yeah, I definitely think it's fair. The biggest change is for the last two years, the Nvidia story has been ironclad. Like in every earnings report, the every growth, I mean, so the valuation just becomes richer and richer and the overall market cap grows and grows, but no one could question it. There was absolutely no doubt. I do think, again, the Nvidia story has fundamentally changed. And I'd mentioned this before, but it's almost fascinating to me when you remember, think about the $3 trillion market cap, how much money is tied into this company?
40:00 Speaker 01 Because every sell-side analyst report, just emails left and right, like everything is show, everyone is just like, this means nothing, the story is still great. Because the amount of just vested interest in this company right now means that people are going to fight for it. People are going to fight to make sure the story doesn't go away. But it's, it was always a story and they were realizing it pretty well. But now there's at least a seed of doubt that was not there in the last few years that at a certain point, do we need? Better and better chips, is that where the battle is going to be won? Because then there's no competition. Do we need more and more compute? And even if the world needs more compute,
40:44 Speaker 01 do they need the latest Nvidia chips to realize actual utility in applications? Maybe not. There's at least a little bit of doubt. So I think the recovery here over the last few days is warranted, that, you know, it was a pretty sharp sell off for a company that size. But I did so the invincibility of Nvidia for me is gone.
41:06 Alex Kantrowitz? There's still so many unknowns, right? A lot of this is based off of things that should happen, right? Like because this, because DeepSeek was able to develop with like lesser chips, then if you actually put those innovations and you use the better chips and use more of them, then they should deliver better innovation. But this is all based off of hypotheses, right? It's not like a rule, just a hypothesis, like it's not a law. And so I, yeah, I totally understand why people would be a little wary about Nvidia. And when you trade a company like that, you trade probabilities. So did the uncertainty of thirteen change the probability that they would dominate? Yeah, maybe a little bit. By the way, breaking news, this is from TechMeme. They, they're quoting Wired. Sources say OpenAI plans to release 03 Mini today
41:48 Alex Kantrowitz? with 03, 01 level reasoning and 40 level speed as the company's staff is galvanized by DeepSeek's success. So it's game on.
41:58 Speaker 01 I guess it's game on, but just make operator a little better before you do that, guys. That is like, I, I mean, it still, it still baffles me. And I say this again because I'm not an AI researcher. I am just a technology person that's using all these different apps and technologies, like building some stuff, but like, like, why are they so focused? Like they released Sora with so much fanfare. It's basically useless for the vast majority of the world in its current incarnation. Operator, huge release, not really usable. Enough people show some demos and there's endless Twitter threads of like
42:43 Speaker 01 top 10, like amazing ways you can use operator. But in reality, no one is using operator today in an actual like.
42:52 Speaker 01 Day to day way that helps their work. So I just rather than focus on making that stuff good and being so caught up in the model battle. It's I it feels like to me that if if they that mentality never changes, that $300 billion, $300 billion valuation looks even richer.
43:13 Alex Kantrowitz? Well, you're underestimating pride here. You know, there is a level of pride that's involved in that's probably what's happening. So, OK, let's quickly hit some of the misconceptions about DeepSeek. We can just go through this real quick. We talked about last week, the it was a misconception. We talked about last week that, you know, they said that they trained just for 5 million. I think we can both agree now that they trained for a lot more and 5 million was just their last training run. Or even though it's impressive for a training run, it doesn't fully incorporate the cost. So we should definitely note that.
43:44 Speaker 01 Yeah, no, no, I think it it seems to be like everyone has accepted it's not 5 or 6 million. It's something more, but it's still not 500 billion or 100 billion or it's not it's not at that scale. And I think that that still means that the story is is very important. And then also, again, even amidst the export controls, and then you get into, did they circumvent them? When did they circumvent them? What chips exactly did they get and did they build on? But whatever it is, it was not some super cluster of the most advanced up to date chips. And so it still reminds us that kind of that necessity breeds innovation or whatever
44:29 Speaker 01 the quote is, is still definitely part of this, uh, part of this story.
44:33 Alex Kantrowitz? Yeah, and I said this on CNBC, like, it's not the process as much as it is the output. And the fact that they were able to run reasoning at such a cheaper cost is the thing that matters most. And people can complain about the way they did it. They can complain about the fact that they're talking about a different number of chips, but ultimately, like, the process has galvanized. I mean, the output has galvanized Silicon Valley, like you're hearing from Altman and Amade, and that's the bottom line. The other the other thing that's been interesting has been. The anger, it seems like, from Microsoft and OpenAI, where, and from others, where it seems like OpenAI had, that DeepSeek had effectively copied some of the stuff that they were doing,
45:16 Alex Kantrowitz? or like, maybe taken their data. This is from Bloomberg, "Microsoft probing of DeepSeek link group improperly obtained OpenAI data. Microsoft and OpenAI are investigating whether data output from OpenAI's technology was obtained in an unauthorized manner. Microsoft's security researchers in the fall observed individuals they believed may be linked to DeepSeek exfiltrating a large amount of data using the OpenAI application programming interface or API." This led us to a, I think, quite funny headline at 404 Media. "OpenAI furious, DeepSeek might have stolen all the data from..." And OpenAI stole from them. Raja, you're pretty good at like, sort of assessing like, when it's fair for tech companies to take content and when, you know, it's all
46:01 Alex Kantrowitz? right to rip it off, or rip other people's products off. Is this like, a sort of nefarious move from DeepSeek? Or is this like, somebody in a glass house throwing rocks?
46:12 Speaker 01 I think glass house and rocks, especially OpenAI. So you have on one side, the kind of, like, I mean, certainly larger question that I've still predicted will be answered in some capacity around, like, individual artists' work, or, you know, creating in the style of specific artists, or certainly the New York Times lawsuit. But then, even at scale, if we remember, OpenAI apparently had been scraping YouTube, like, at a large scale for their initial data sets. So I think, and the entire world seemed to have reacted in a pretty similar way, where obviously, it's a bit rich from all
46:55 Speaker 01 companies, OpenAI kind of going down this route. I don't, I don't believe they had released any kind of official statement, so at least to their credit, these were off of a, kind of like, just more, like, either leaks or just genuine reporting that they're looking into this or exploring it. But yeah, I don't think anyone, anywhere, is gonna have any sympathy for OpenAI in this matter.
47:20 Alex Kantrowitz? Yeah, I'm on the same page there. And it was like, it wasn't done through hacking, it was done through the API, so like... I guess Crimea River.
47:28 Speaker 01 Yeah, I mean, it's the YouTube example perfectly like and it's a big platform against big, well, actually, small platform against big platform in this case, but just using the actual available technology and probably breaking the terms of service a little bit. And then using that to get started.
47:47 Alex Kantrowitz? Okay, so I want to take a break, but before we do, I should say that we have a Discord now. So Ranjan and I had been talking about starting a Discord as these stories had continued to break and there was more to talk about and we were kind of curious what the audience was thinking throughout the week. So that Discord is now open. It is open to Big Technology paid subscribers. So if you go to bigtechnology.com, you can see there's a post that says, let's talk deep seek AI, etc. on Big Technology's new Discord server. If you're a paid subscriber, you just scroll to the bottom. There's the invite link. If you're not, you could just sign up and then scroll to the bottom and there's the invite link. It's been kind of fun. We're just about wrapping day one and there's been some really good conversation there already. Alex Demos, the security researcher,
48:32 Alex Kantrowitz? is in there telling us right now about how DeepSeek's security issues really are compared to all the hype. Like, what happens if you download DeepSeek? Are you in trouble? So thank you for that, dear Ranjan. It's been it's been fun getting it off the ground and I hope listeners will join.
48:49 Speaker 01 I actually learned, I think, more from just a few comments from Alex Demos around what are the actual security concerns or considerations on DeepSeek than pretty much everything else I read this week. So like actually at a very technical level, even on like having to myself look up what safe tensors are and other like really deeply technical terminology. So we're it's off to a good start today.
49:13 Alex Kantrowitz? Yeah, it's been cool getting it off the ground. And again, you could go, if you want to join, just go to bigtechnology.com, click the let's talk deep seek AI, etc. on Big Technology's new Discord server link. And then join us. We'll see you over there. All right, take a break. And when we come back, we're going to talk about how Siri's still terrible, or maybe it's even worse, and then briefly touch on Apple earnings, and then we'll get out of here. So we'll see you right after the break. This episode is brought to you by App Point. Everyone's racing to roll out AI right now. Co-pilots, 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,
49:54 Alex Kantrowitz? secure it, and prove it's under control. That's exactly what App Point 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, so innovation scales without scaling risk. App Point, the unifying trust layer for AI. Head to apppoint.com to see how enterprises deploy AI with confidence. Learn more at avpt.co/bigtechnologypodcast. That's avpt.co/bigtechnologypodcast.
50:39 Alex Kantrowitz? 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 Ironwall knows this better than anyone. They've protected some of the most targeted executives and individuals on the planet for almost two decades. 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, Ironwall 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,
51:24 Alex Kantrowitz? 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.
51:39 Alex Kantrowitz? And we're back here on Big Technology Podcast Friday Edition. So we've talked about DeepSeek the entire episode so far, which I think has merited, but we're not going to let the week go away without talking about this great post by John Gruber on Daring Fireball titled Siri is Super Dumb and Getting Dumber. I mean, the, the long and short of it is that basically he had a friend who asked Siri, win the Super Bowl, who won each Super Bowl and Siri got 34% of those right, which is truly disastrous. And basically Gruber's point is Apple got this gift of generative AI. It put it into Siri and somehow it made Siri worse.
52:20 Speaker 01 It definitely, the, the summary notifications have been getting worse and more, or just realizing how much use, how useless those have been getting is just becoming even more and more acute in my day to day. Genmoji were fun for a minute and then I ranted a bit about this week. So I'm glad that even like Gruber, who is about as long time an Apple fanatic as it gets, the fact that he is recognizing and ranting himself about how bad it is. But it really calls into question, where are they in all of this? The only thing though, I don't know, I'm curious. Do you think this is good for Apple this week?
53:01 Speaker 01 Reminding us that the entire direction of the AI industry is in question. So maybe them screwing up the first phase of this battle so badly actually means they'll be okay and they can start fresh.
53:16 Alex Kantrowitz? It's definitely the Galaxy brain take here. It was kind of interesting watching, so the Apple earnings came in yesterday. They sold less iPhones in this Q4 than they did in the year previously. So iPhone sales are going down. Apple Intelligence is garbage. And yet you look at what happened. DeepSeek comes out, actual AI innovation, there's panic in a sell off. Apple made Siri worse and is selling less iPhones and the stock is up 6% on the week. It really is amazing. I don't know, I really, to me, I'm a little bit puzzled at the, the buying activity on Apple. I just don't see what the news is. Maybe services, they beat on services revenues. That's pretty impressive.
54:00 Alex Kantrowitz? But we talked about it, that when Apple Intelligence came out. It wasn't going to lead to a super cycle. It hasn't led to a super cycle. And I think Mark German put it well, put it pretty well today.
54:15 Alex Kantrowitz? He said, Apple Intelligence is a half baked marketing strategy that was rushed in response to OpenAI and Google Gemini. Yes, Apple had no choice. It is exactly the right thing, but they shouldn't have been in that position in a first, in the first place. So I guess like what the market is saying, you know, why it sent it up this week. Of course, there's the services beat, but even after DeepSeek came out. And I just think it's a pretty simple thing, which is just that you can build more with DeepSeek. And Apple has got to find a way to build consumer products and it can maybe use open source to do it. But I don't know, do we have faith that Apple's going to do it? I don't.
54:50 Speaker 01 Well, I was thinking, cause even that idea that the marketing drove the product side of it with Apple, which usually it's the other way around. To me, the most fascinating part of this is, and I think a lot of companies and Apple probably is the one most troubled by this. Is the kind of the dichotomy between normal, everyday users and call it the industry. And by the industry, I mean the actual, the technology focused people working in the companies, the investors focused on the companies. There's such a distance between what any normal person wants or expects from AI and what these companies are pushing.
55:34 Speaker 01 And to me, that, that, that distance has actually been these kind of like the core part of Apple's errors in this. That they're more focused on the market and probably their own competitors and even their own employees and what they're thinking about. Rather than what the everyday consumer is thinking about, because again, I can promise you.
55:56 Speaker 01 No, most certainly non-tech focused iPhone users were not clamoring for Apple intelligence. And I think even most technology advanced Apple users could easily install ChatGPT and whatever else on their iPhone and still be fully locked into the Apple ecosystem. So the idea that they needed to do this, I actually disagree with Mark German that they had to do this. I think they could have still just sat back. The only reason they would have had to do it is for the market. And I mean, apparently, it's still kind of working. Apparently, the average Apple investor has not actually tried to do a basic task with Siri, but it's still working, I guess.
56:42 Alex Kantrowitz? I just want to end with this. I love the how it started, how it's going meme. And I saw one come across my X timeline this week that just really sort of captured it all. How it started is the Think Different campaign with these beautiful black and white images that can know, like, if you're an iPhone user, you're just kind of more artistic, like kind of classic, you have more taste. And then I've been in, and okay, and then the second image is how it's going. And it's Imagine It Gen Moji, and it is a sunny side up scrambled egg with hands and feet.
57:16 Speaker 01 This is what people want. This is what they've been clamoring for and why they will spend, it will launch an entire new iPhone super cycle. The sunny side up egg with hands and feet.
57:30 Alex Kantrowitz? I mean, I've been in Miami for most of the week this past week, and these, I don't know if they're in New York yet, but these Gen Moji billboards are everywhere. And they're terrible. They don't say anything. And if that's your AI play, Lord help you.
57:46 Speaker 01 I saw some, some posts around, like, everything they've offered feels like something that Tim Cook is the only user of. Even notification summaries, like, again, they're relatively useless and almost counterproductive. But I can see if you're Tim Cook, maybe they're kind of useful. I can see Tim Cook really being into Gen Moji. Just, he's just churning out, tossing out Gen Moji left and right. I mean, someone's got to be using it.
58:18 Alex Kantrowitz? All right, well, I will make like a Jen emoji as a big scrambled egg with one hand up and say, have a great weekend, Ranjan. Have a good one. All right, everybody. Thank you, Ranjan, and thank you all for listening. We'll be back on Wednesday. Actually, I'm speaking with an Nvidia executive, so that'll be fun to air on Wednesday. And then Ranjan and I will be back on Friday to break down the week's news. Thanks again for listening, and we'll see you next time on Big Technology Podcast.
58:43 Speaker 02 Now, more than ever, we need someone on our side. But Rob Whitman? Whitman's votes drove up costs, cut Medicaid, and risked 350,000 Virginians' health care to pay for tax breaks for himself and his donors. He's not on our side. Shannon Taylor will fight corruption, cut taxes, and bring down costs. This November, vote for Shannon Taylor. She's for us. Paid for by HMP, the House Majority PAC dot com. Not authorized by any candidate or candidate's committee.
59:15 Speaker 03 Every mile matters. Whether you're driving your dream car or your daily commuter, the Mobile One team knows that. It's why we created Mobile One 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 Reid Hoffman: Why The AI Investment Will Pay Off Next NVIDIA's Plan To Build AI That Understands The Real World — With Rev Lebaredian Page updated October 12, 2026 · transcript and summary are machine-generated