저자: RanjanRoy · 외국인
요약 보기
젠슨 황의 Dwarkesh 인터뷰 평가: 경쟁(TPU·Trainium의 Anthropic/Google 모델 학습)과 수출통제 질문에 명확한 답을 못 했다는 비판. 엔비디아는 여전히 강한 기업이나 방어 논리(생태계·용량)가 설득력 부족. 올트먼 관련 후계자 논란은 과장으로 보며 122억 달러 펀딩은 긍정적. Allbirds의 GPU 피벗 주가 급등과 붕괴, 브랜드 자산 기반 '좀비 브랜드 롤업' 농담성 아이디어. 개별 종목의 명시적 가격 방향 예측은 없음.
다듬은 글
Jensen Huang struggles to address Nvidia's position on competition and export controls. Where does it leave the company? Sam Altman's conflicts are getting at least some open eye investors a bit nervous. And Allbirds is a GPU company now. That's coming up on a Big Technology Podcast Friday edition right after this.
Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool-headed and nuanced format. We have a lot to cover in a great show for you today. We're going to talk all about Jensen Huang's performance on the Dwarkesh podcast and whether that leaves us with some concerning, I don't know, red or yellow flags about Nvidia's position moving forward.
[clears throat] We're also going to talk about Sam Altman's conflicts of interest that have been laid out in the Wall Street Journal and whether some investors want to potentially replace him as CEO. That's what they're talking about. And then of course we got to get into this Allbirds GPU pivot. Joining us as always on Fridays is Ranjan Roy of Margins.
Ranjan, welcome. Good good to see you, Alex. I'm excited. This Allbirds story I'm definitely excited about, but This is Red Meat Ranjan Roy. Yes. Ranjan Red Meat. Ranjan Red Meat. I think we should both state that we didn't wake up losers today. So we're going to address Jensen Huang uh conversation with Dwarkesh.
You might be a car. We are not a car. All right. So obviously I'm referencing some of these uh very quotable lines that uh Nvidia CEO Jensen Huang had in a conversation [clears throat] with podcaster Dwarkesh Patel this week. Um basically, you know, the interview has gotten like a lot of play because uh Jensen, who's usually such like a calm, cool, and collected interviewee, um really did seem to get agitated.
But to me that wasn't actually the big story coming out of this interview. I thought the big story was that and uh well I'll caveat Nvidia's still like a great company. It's going to do well. It's the world's most valuable company at $4.8 trillion. But uh to me the highlight was Jensen basically struggled to articulate a cohesive story on the two biggest issues facing Nvidia today.
One is competition, and the other is export controls. Um And Ranjan, I think we can just go by go through those arguments one by one. I have some quotes from the show, and sort of be great to get your take on each one. How does that sound? Let's do it. Okay. So Dwarkish basically leads off this conversation asking Jensen, which I think is what [clears throat] I think is the most pressing question for Nvidia, right?
Which is that Nvidia has been the company powering this AI moment. Obviously, OpenAI, which is the one that sort of spearheaded everything, is a heavily Nvidia GPU-reliant company. But, and this is the way that Dwarkish puts it, if you look at the TPU, which is the accelerator that Google has been has put out, and Trainium, which is Amazon's version of it, you know, two of the top three models in the world, Claude and Gemini, were trained on it.
And what does that mean for Nvidia going forward, right? So, you have Google's Gemini, Anthropic's Claude, not trained on the GPU. So, the question is what's the moat for Nvidia moving forward? Couple of responses from Jensen. I'm going to do my best to capture them as well as I can. The first line that he said about Mythos, and this was in response to a China question, but I think this is really telling.
He said, "Mythos was trained on fairly mundane capacity, and a fairly mundane amount of it." He said, "For for um the growth of these competing chips, Anthropic is a unique instance, not a trend. Without Anthropic, why would there be any TPU growth at all? It's 100% Anthropic. Without Anthropic, why would there be Trainium at all?
It's 100% anthropic. I was not impressed by the defense here from Jensen. I mean, I didn't get any clear answer from him on why Nvidia will maintain its lead if these two have as Dwarkesh put it, two out of the three best models have been trained elsewhere. Am I being too harsh? Okay, I'm going to I'm going to make myself a little vulnerabler here, Alex.
I listened to the entire interview. I will say as someone who works in and lives and breathes AI all day, there's a lot of it I wasn't fully understanding So like walk me through and and I'll say the AI infrastructure side of the world, that's where John, who I write margins with, that's his world. Um, so like explain to me what the what their vulnerability is here.
Again, if it's their training on TPUs, around training, I'm like walk through this argument and why that presents a vulnerability for for the normies like us, the business folks. I mean, it's very simple. I and I don't think we have to overthink it. Like we can get into the technicalities like Dwarkesh did, but I don't think we need to.
I think the point is that the GPU is seen as the perfect technology to train AI models on. Right? Just because of the way it's constructed, right? It can do many different um, processes in parallel and that uh, allows you to do this sort of matrix multiplication uh, that you need when you're doing AI.
Um, I think that's about as technical as we should get. But to me, the bottom line is that that you have Google, you have Amazon who have made alterations to CPU chips that have done a good enough job here and helped anthropic and Google train excellent models. So then the question is, what is Nvidia's moat going forward?
And I didn't get like a complete coherent answer from Jensen. So, at that layer, I mean Jensen and I'll say that I think this is the first time I've actually listened to a full hour or whatever it was with Jensen Huang and it was very interesting. I I now really understand how good of a salesperson he is cuz my god, like every line of his was just so pointed around like we are the greatest God's gift to Earth company and no one can touch us.
He certainly made the ecosystem argument at every level and he made the capacity argument at every level. So, I'm going to if I'm going to take the Jensen side here, I think his argument is yes, you can't like from a pure model training standpoint, that potentially there is some competition, but in terms of actual like being reliable capacity for all of the ecosystem rather than just if you are Google, yes, you have access to your own TPUs.
I think that's his argument. I mean he said CUDA over and over again in terms of like that these models are trained on that's linked directly to Nvidia infrastructure, right? Exactly. So, so he is saying it's the stack. It's no longer the chips alone, it's the stack which I'm curious again, I will say as an I'm not an infrastructure guy.
So, like it sounded convincing. You were certainly I would say this versus the China answer. I at least walked away when he was talking about like you know, like their relationships with suppliers and TSMC over 30 years and always being able to deliver expansion of capacity and like like that was the argument.
That was definitely his take rather than the individual chip. To me, that's what I had walked away from. Do you think that's not not a reasonable moat for an Nvidia? Well, I would say the pushback here is again, I don't think you need to be technical on it. The pushback here is take a look at the bifurcation.
Who is right now, I would say the leader in conditional models is Anthropic. Um then you have Google and OpenAI. Who has spent globs of gobs of money on Nvidia chips? gobs. Globs Globs and gobs of money on Nvidia chips and thinking they could brute force their way into this race. Meta and xAI. Has it worked for them?
No. So, um I I think capacity is though however, like I think that that might be the best answer, which is basically like and if I was Jensen, I might might have put it this way. All right, you want to take Anthropic as an example? What is the one thing that people say about Anthropic when it comes to criticism?
They get rate limited. capacity. Yeah. Lack of capacity. I think that really is that would that would be the the better argument. Um like there were some I I think you could have just made that more clearly. You're right. Like now that you I mean, you just said it. He didn't just say that. He I mean, I think you're explaining Jensen better to me than Jensen did himself cuz like he was kind of going very technical, being a bit of bit standoffish, but didn't make that point super simply and clearly and when it said like that, it actually starts to make a bit more sense.
Yeah, and there's no love lost for Dario. So, I don't understand why why he wouldn't have just come out and say that. And then when he talks like Dwarkesh also brought brought up these other companies that use TPUs and Jensen had a weird answer on this one. If they don't try these other things, how would they know how good ours is?
Sometimes you've got to be reminded of it. Like what? They're spending billions of dollars with other companies just to be reminded of how good Nvidia is. No, clearly these other companies are serving a need to them and Nvidia is also moving in in this direction, right? Remember they acquired the or they they acquire Zision, the top talent from Grok, not Elon's Grok, but this Grok company that's meant for inference and that was one of the big answers for them in in their GTC conference this this this year.
So, I I think honestly the the the better answer, the more honest answer from Jensen would have been, "Look, we have competition now. We acknowledge it, but we have the capacity and you could tell that people that use our models, you know, are are being are able to serve them." You know, but there's there's a counterpoint to that also, which is like, "Well, maybe if people wanted to use Meta's models more, they would also have capacity issues."
I think it's it was I mean, again, as a non-infrastructure person, just listening to the interview, one of the most fascinating parts was exactly that. Like I I agreed. Like, just say, "Yes, of course there's competition. Competition is healthy." Rather than his entire argument is like, "You can't escape us.
We control the stack." And like, "You have to come to Nvidia and there's no other options." And I mean, it was basically like I mean, we are a monopoly for majority of the AI stack and that's just the way it is and good luck. Go try something else, but it's not going to work cuz we own the stack. So, yeah, I think I don't know.
I I I feel Do you think humility would be better for him in these cases or cuz again, or is this just why I mean, you can't argue with what he has done with Nvidia over the last couple of decades, especially the last few years. So, if that is just how Jensen does Jensen, I mean, I guess you should we not expect any humility or would humility be better?
I don't think he has to be I mean that humble, but I think grounded in reality would have been good. And I and I think you especially saw that come out on this export controls conversation. All right, so so this is really where uh and I think you're by the way I think you're highlighting exactly the issue here, which is that I think Dwar was asking good questions.
It got somewhat heated, which was interesting to see. But again, it's not the heated nature of this conversation that I kind of pick out here. It's sort of um this this uh I don't know. Jensen's Jen for for whatever reason Jensen just could not get this coherent message or cohesive message out. Uh and I think a lot of it had maybe it maybe it did have to do with some some lack of humility or an inability to nuance, right?
To go back to our show's theme, there was no nuance from him at all. So so let's go to the export control thing because that really gets to it here. So Dwar made a great point, which is that Anthropic and I'll quote Anthropic announced Mythos preview. They said it has such cyber offensive capabilities that we don't think the world is ready until we make sure those zero days are patched up.
So if Chinese companies and Chinese labs and the Chinese government had access to the AI chips to train a model like Claude Mythos, um wouldn't it be a Now I'll paraphrase. Wouldn't it be a vulnerability to the US as opposed to now where we have more compute and we have this heads up to make our institutions ready?
And over and over again, Dwar asked Jensen to acknowledge if we give, you know, that much more compute capacity to China, isn't there a risk of cyber warfare coming from the Chinese side where and and creating vulnerabilities where we're not we're not um ready. And Jensen did not have a good answer.
Wouldn't actually wouldn't answer on this. Well, even and [clears throat] I mean I saw that some tweet around this around like that the the great part of Dwarkish having technical fluency in this interview was Jensen even said, "Well, China has 60% of the world's chips." But recognizing that chips aren't the measure that they're talking about, it's compute and that China only has 10% of the world's compute.
So, they still could be contributing to this. I think it's again, this is why it was so baffling to me that he was that standoffish is it's a pretty nuanced conversation. Like export controls to China, what's already done, what can you prevent, like what are the it's going to happen anyway. And I thought I'm surprised from like a comms perspective and I know I always bring everything back to comms that there wasn't a better answer.
That like this is one of the biggest questions facing Nvidia that there was export controls, then clearly they had pushed back themselves and got them removed and did whatever they had to do. And so, like this is such a sensitive topic and it's going to be on an ongoing basis that like why wouldn't you have a better answer that then I am we are not a car.
Which Right. I I got it after. I saw the the so Explain explain the we are not a car thing. You're not talking to somebody who woke up a loser. And that loser attitude, that loser premise makes no sense to me. We are not We're not a car. We are not a car. That is a loser premise. Like the whole loser thing, Dwarkish never said loser, which was so crazy to me that he just went on this tirade about losers, which I I don't know where it came from or what that triggered.
Okay, sorry. I can explain it. Basically, what what Dwarkish was saying is US companies have a history of getting into China and then losing to China in the battlefield that they they created. And the car example comes from Tesla got into China, right? And whether it was through I'm just going to speculate here corporate espionage, very good mimicry, break down the cars or whatever, China ended up crushing Tesla, right?
The Chinese EV industry right now is the world's best. And so could it be the case where like, you know, even if you get into China, you're eventually you know, you help them with their technology and then they they replace you and all you've done is accelerate China and then lose. And that's where you get the I'm not a loser mentality.
And the car mentality. See that Alex, I think I think you need to do like a Jensen translator app where it goes over and just kind of explains in a much more clear kind of like way. Cuz yeah, okay, I get it that way, but again, he literally starts going like we don't that's a loser premise, I don't wake up a loser, we are not a car, and that's kind of as far as the argument goes.
And it just sounds like it just I mean it's it's just made for a viral clip in not a good way. But do you think like what do you think Nvidia should do? Cuz I also found it a little awkward as the conversation goes on where it's like the we need cooperation, 50% of the world's AI researchers are from China.
They have the world's best AI researchers, like we need back and forth, we need to work together, which sounds great in theory, but is a little kind of over the aspirational. So he goes into this whole like better together story and then at the end he's like, but the US is great and the US is the leader.
And like it was just almost this like you feel like this like the US government over him knowing he has to kind of kiss the ring a little bit and like say, oh by the way, no, but the US is the best, US is the leader. Like it just I don't know, the whole thing there needs to be better answers for prepared.
Yes, okay. So, can I go back back into my um you know, coaching Jensen on coms here? And I want to acknowledge that Jensen is in a tough place. Remember, he's uh he's he's accountable basically. He's trying to balance shareholders, public opinion, US government, Chinese government. Not easy. Um and in fact and sales, right?
He's trying to sell. So, I think one of the things like Dwark you know, one of the weak points even though it it might be true and sort of there's a there's a balance here because you always want executives to be honest, but like but but Jensen's Jensen's core argument is selling these chips everywhere is good for the US.
Uh good for Western values. Uh but then he gets pressed by Dwarkesh where Dwarkesh is like just acknowledge the fact that if you sell this into China who has data centers open and ready to plug them in, you're creating a potential cybersecurity risk for the US. And Jensen goes, I just want you to acknowledge that any marginal sales for the American technology is beneficial.
Which is you know, sort of like you can't it's either I want to make sales or I'm good for the country. Um sometimes those things come in conflict and that's where he was weakest in saying uh and he could he really couldn't hold those two things in balance. Because cuz they're not They're not in balance.
They're not in balance you're right. Sorry, go ahead. Yeah, but this is the this is the most ridiculous part of this and and I I almost like feel bad that he is the one who has to kind of defend something that is impossible to defend because it is an either or situation. Either you are working with the Chinese people and research community and economic ecosystem or you're not.
And like there's good and bad to both, but in reality in a in a normal state of affairs, this would not be the job of the private company. The private company should only be saying, "We will sell to China." And it's the responsibility of the government to make the decisions that are on the geopolitical It is in Nvidia's best interest as a pri- or sorry, public company, but still, you know, like a They are responsible to their shareholders.
I do believe that, and selling to China like they should be doing that. They're They're incentivized to do it. There's good arguments for it. And it should be the government's kind of like role and responsibility to decide what is the best interest of the nation and have the best understanding from a security standpoint and everything else, but everything is just so out of whack right now that we're depending on Jensen to be both CEO of public company and kind of like security and trade minister and all of the above.
So, I'm going to Yeah. I think I'm I might be going back to Jensen's side that it's he's put in an impossible situation there. No, I agree. And And he became visibly angry because this kind of talk has led Nvidia to lose some leadership position in China, which he can't be happy about. Remember, there have been export bans.
They've Many of them have been lifted. But okay, now I want to sort of This is where I was trying to get here. All right. There is a good argument to make, and I think some of Jensen's conf- or balancing act might have limited him in the ability to make it. All right. So, the argument And he made a version of this at the end, and I think this was the strongest part of the whole conversation for him.
The argument to be made is that listen, you are dealing with LLMs. This There is real cultural soft power in having your influence on LLMs diffuse through the United States. And there's going to be two different one one of these two options of LLM's that will be used globally. An LLM with and and the truth is there's values baked in and LLM with American values baked in or an LLM with Chinese values baked in.
And if you're being honest, the way that it is going to diffuse is through open source. So you have two options here. One is you influence the way that the open source direction is going to go by having it built on Nvidia's tech stack, which is going to have American values, which is going to open the door for Americans who are used to working on Nvidia to influence and contribute to open source.
The other option is you you shut the door to China. You make two competing ecosystems. One is this closed American ecosystem that is great great technology, but isn't globally adopted like Anthropic and OpenAI. And two is by shutting out the Chinese, what you're incentivizing is a separate ecosystem where Chinese values will be baked in.
They have They have the world's They are They are going to lead open source without Nvidia participating and all of the sudden their values will be diffused into models that are used in Africa, in India, and across the world. And And Jensen, I'll I'll just say this one thing. Jensen does make this argument at the end.
Not quite on the nose as I would have liked, but he does. He says, "The fact of the matter is that we get to benefit. We get the benefit of American technology leadership. We get the benefit of developers working on the American tech stack. We get the benefit as these AI models diffuse into the rest of the world that the American tech stack tech stack is therefore the best for it." That's That's my point.
two questions on that. One, yeah. Walk me through how exactly you think an LLM reflects the kind of soft power and cultural values. And then two, is an American company I think like I saw Chesky and Airbnb kind of talking about how now a much higher percentage of their workload is actually deployed on Qwen from Alibaba cuz it's cheaper.
And I've actually I've seen a number of people talking more about that. And especially as in the last week there's been all this talk around how Anthropic has been subsidizing and the real bill is coming and like cost management is going to become the next frontier in a genetic like moving to like is it okay that for if you're calling Qwen for your for your workload or is that un-American as well?
So those are my two. What's the cultural values and then what's is it can we use Qwen to process our podcast transcripts and still be America first? Uh it's a good question. So it's not it's not black and white, right? There's nuance here. If you're building an enterprise software application or if you're trying to do uh hotel bookings, fine.
But let's say there is a uh a proliferation of AI consumer apps out there. Um and I don't think this is fearmongering. I think this is true, right? What happened when you tried to search for Tiananmen Square in DeepSeek? It was blocked. Now people were able to modify DeepSeek and uh and you know, strip that out, but it was only after modification.
So I think you'd you know, if you're thinking about like and again, this is this is the argument I think Jensen should make. I'm not 100% sold on it. Um but if you're thinking about like all right, uh if these models diffuse through the world and there are going to be applications built on top of them, um who's who's who do you want to set the weights?
You know, the Chinese Communist Party or everybody else? Um I think the you know, the preference would be everyone else. That would be the argument if you're if you're Jensen. I think that's a strong argument. No, I I do think it is interesting to me that the way if the next soft power battle across the globe is is at the LLM level.
I think that does make a lot of sense and we should be. But again, that's not Nvidia's decision to make. Like that that's has Of course it's not his decision, but it is it's a core it's a critical part of its business and it's something that, you know, while maybe unfair, Jensen should be better on when he talks to
I I agree. I mean that I think the the main takeaway on all this is this is going to I mean going into an election year with AI is it's going to be the poster child for everything wrong with the world and then China is the These are the two single biggest like political negatives right now, China and AI within the US.
So, I think maybe not the biggest, but two of the biggest and like you got to be better prepared for that. That's That's the main takeaway. think Dwarkesh had a great point by the way, which is like right now as we speak US institutions, including the US government, which we're about to get to, are evaluating Mythos before it's released to the public.
So, wouldn't you want to be first? And if you give more compute to adversaries who we know are trying to you know, at least hack into your systems uh you're you're putting yourself at risk. And that's that's that's always going to be the compromise. All right. Well, walk me through what's happening on the US government side this week.
Okay. So, so first of all, I think let's just start with this. Last week we had a big discussion, Mythos or marketing, uh Mythos real or marketing. Um the I you know, I don't think you can say 100% conclusively now, but certainly the data points suggest that Mythos is more real than marketing, which is kind of crazy.
Uh this is from the AI Security Institute. I brought this brought this up at the Pentagon this week with the meal Michael, we're going to talk about in a moment. They say we conducted cyber evaluations of Claude Mythos preview and found that is the first model to complete an AI SI cyber range end-to-end.
The range simulates a 32-step corporate network attack from initial reconnaissance to full takeover and we estimate that it would take a human expert 20 hours to complete. So of course there are some flaws in these evaluations but it the fact that Mythos was able to complete this end-to-end is a bit spooky and I think that you're starting to see the seriousness of the Mythos model manifest itself in the way that lots of companies and and especially the US government are dealing with Anthropic now because there seems to be a universal acknowledgement that whether it's Mythos or whether it's the next iteration of this you got to take this stuff seriously.
So I'm going to not say we shouldn't take it seriously. I've had more questions [clears throat] on this in the last week since we spoke last Friday when I walked our listeners through what was clearly a very very coordinated PR stunt about a sandwich in the park from Anthropic and I actually had spoken with a few listeners and like and it was funny because it's that becomes so clearly a marketing tactic.
So what I had spoken about this week was is Mythos like Glass Wing in the way it rolled out and and I again cyber security as well is not where I've spent a lot of my time. So like typically if you have this model that can do these zero-day exploits and can actually start to piece all of this together and like why make such a big deal about Glass Wing like normally what like any any model rollout would have had some kind of security check and maybe you're working with other companies as well, but instead they made this like and and I spoke with someone who works at one of the large companies who's on the cybersecurity team who was kind of like invited to be part of Glasswing.
And the way they were speaking was literally like they got into like were like a VIP invited to some event. Like it it it the marketing is when it's that good, I just I have a hard time believing it, but but what what do you think is actually Do you think there's a step change with Mythos that actually represents some massive new vulnerability and it's actually going to be patched up in the next 3 to 6 months as Glasswing gets together and finds all the potential vulnerabilities?
First of all, I'll say Yeah, sorry, go ahead. No, do you do you think that's what's going to happen aside from the marketing, but then also address the marketing as well? I'm curious. Okay. First of all, I'll say there this is the counterargument and I think it's really good. People have said marketing at every step of the way, you know, and we've talked about it also and the capabilities just get keep getting better.
So at a certain point like when are you going to say, all right, well, maybe this is real or not. I don't know. I I can't tell you. I don't have an answer, but I can tell you that it's working. Oh, it's working. It's working. By the way, I just thought like wouldn't it be Wouldn't this just be a setup of like a great movie where Dario just like takes a heel turn and says, I tried to be nice.
I tried to tell you what my models can do. You didn't believe me. Now I will use the power of Mythos for evil and make a society believe in the technology I've created and he just becomes this black hat cyber hacker, takes down society and bends it to his will. That's what Like guys, for how good the marketing is, that's the picture you are painting in our heads with all of this talk right now.
Like if you really cared, like I get there's value marketing value in like creating this god-like complex around the models that are being released, but it just makes people more uncomfortable with with the industry. Like actually, so what I can't get through is like, okay, let's say you have this in your hand.
And you realize that if we are to release this, this will has a potential to kind of like make every single small, medium, and large company completely vulnerable to hacking, has a potential to kind of like take down the entire fabric of our economy. The digital fabric at least. Like Like what would you do?
Would you do a coordinated press release and have a project glass wing and and then still work on releasing it because you just raised another 30 billion at $380 billion valuation and like still keep raising like wouldn't you take a pause and just be like maybe we should all just try to address what's actually happening here rather than coordinated marketing campaign that leads to another funding round that lead I mean, wouldn't you try to fix it?
Whatever like really fix it, not use it as a a lever to kind of further the economic gains that you continue to accrue. No, because the idea is and we should move on because this we talked about this a lot last week. I think the idea is it's all intertwined. You need to keep growing to keep ahead. So you're able to give this to certain companies to try ahead of time because otherwise, you might have, you know, well, I mean, they would say companies that are less trustworthy, but I think Dwarkish made the point that it could be other other um countries with different values that get ahead and then all and then all of a sudden they start the hacking.
And we saw this by the way um this this the downstream effects of this with with Dario on his way to the White House. He's actually there today. It's from Axios. Anthropic CEO Dario Amodei is scheduled to walk into the West Wing on Friday for a meeting with White House Chief of Staff Susie Wales, a breakthrough in his effort to resolve the company's bitter AI fight with the Pentagon.
The Trump administration recognizes the power of Anthropic's [snorts] new Claude model Mythos and its highly sophisticated and potentially dangerous ability to breach cyber defenses. It would be grossly irresponsible for the US government to deprive itself of the technological leaps that the new model presents.
A source close to the negotiations told us it would be a gift to China. Oh, that sounds like Anthropic coms to me. But what do I know? But but but look, there's now there's now I think this is a good thing. There's now a thawing it seems like between Anthropic and the federal government. So, um I think if you're the federal government, you have to take this seriously and if you're Anthropic, I think you're doing the right thing by getting in there.
And is there a chance it's marketing? Maybe. But ultimately, I think that this is sort of the news here is that this is going to create a at least a thaw between Anthropic and the White House. Well, but again, what do you think would happen if they release Mythos today? Like what do you think would really happen?
And what do you think will happen within the current way they're rolling this out? Like do you think if they were to release this today that cyber the hackers would penetrate every defense for even the largest companies in the world or No, not necessarily. I mean, we spoke about it last week. You know, we don't we don't but clearly there's some there's some incremental I mean, incremental abilities.
I mean, the the institute that I quoted earlier like they are independent as far as I'm aware. I think they're UK-based. I don't think they have any incentive to, you know, fluff this up um with Microsoft. Microsoft's in this uh in this um you know, Project Glass Wing. They're big investors in OpenAI.
So, I don't know. Anyway, I think we should move on from the the the marketing. But, you can get a last word in. One last thought. Rather than your Dario heel turn that sounded kind of like an Avengers movie, I want to see it go the other way where actually Dario and Sam come on stage and OpenAI has been invited to Glass Wing and together they are giving mythos to their biggest competitor.
And everyone is going to make this work hand in hand. That's That's the plot twist I want to see. I love it. Maybe even hands holding hands like they didn't do previously, but Yeah, exactly. But, I think you're right. I think peace is a good is a good place to take this because we could see peace between Anthropic and the Pentagon.
Of course, I was in uh the Pentagon earlier this week speaking with Emil Michael, the Under Secretary of War, who's the one that banned Anthropic. And I asked him, you know, you've made you've made uh uh peace with um with Google, who was on the outs with the Pentagon for a while. Can you make peace with Anthropic?
And he said, "I think so." He said, "I hope I hope that companies as they get more mature and more of an understanding of what it means to work with the government and understand us better, get to a good spot hopefully sooner than eight years." I think that was interesting. It was the first interview where uh Emil opened up the door to possible reconciliation uh with Anthropic.
And I I think that's a good thing. Do you think it'll happen? Yes, I do. I really do think that that this is that they will. They will find um some common ground and get back to work together. I think the Pentagon likes Anthropic, honestly. They just the technology. They Claude-pilled. They are. All right, uh before we go to break, I do want to address something cuz we definitely had some uh and I don't know if I really should.
I don't want this to come off as defensive, but I think it's like good to like take a moment to sort of explain uh the way that that I you know, I operate on this show. Um and uh and there were some questions about like why did you go into the white to the Pentagon and and ask these questions and um you know, I uh here I'll just read there was one uh review on the Apple Podcasts uh that I think is worth addressing.
Al- Alex displays a clear pattern of showing deference to the rich and the powerful. His interviews with OpenAI and Trump admin leaders have been extremely deferential. His questions to Dario were much more pointed and journalistic. It seems like Alex is truly dedicated to sucking up to the most evil people in America.
The episodes with Ron Jon are cool though. Two stars. Um first of all, I'll just say uh we I I appreciate listener feedback. I I I'm glad to hear from listeners. I'm I it's a privilege to be able to do this show for an audience. So, so thank you. I'll also say the two-star reviews really are they they they're aimed at they they they in practice um have an effect of really harming the show.
So, um I'd much rather these come in through big technology [email protected], which is in the show notes. Um or if you have a criticism, five-star it and write it and we'll take it seriously. Send us hate mail. Send us hate mail. I but I do I I do think it's worth addressing sort of the style that I have of interviews and sort of what happened with a meal.
Hold on. I want to I want to ask you. I'm going to put Alex on the spot here. Like one of the fascinating parts of the whole Dworkin-ish Jensen interview is like everyone's like, "Oh my god, it was heated. It was pointed." In reality, 10 years ago, that would have been any normal interview because like and now we have the podcast circuit and I'm certainly going to exclude Big Technology Podcast from that and I think you do push a lot of the guests in a hard way, but like how does it work?
Like cuz if you rip apart a CEO, does do people want to come on the show? Like how do you balance that tension at a personal level? Like to try to make this a success. So I So this is great setup because this is my my philosophy here. Um I really think the job is to get newsmakers in the room and ask them the hardest questions possible.
I think the reason a lot of these uh you know, TV interviews or these hard interviews and these back and forth, did you order the code red? You know, there there is there is value in asking officials tough questions. I think the reason why people became disenchanted with the media is because it became more about showboating for the journalist and more performative than it became inquisitive trying to find the answers to these tough questions.
Um and you know, I think Jensen became very defensive in this conversation with Dwarkesh, which is why it led to this sort of heated back and forth, but my style in particular, I I try to go in I want to ask the people I'm sitting across from the hardest possible questions I can, but I also want to do it in a respectful way.
I want to have a conversation about it. For me it's not about a performative thing. And I think if you look at the transcript of the conversation with a meal, even though it was conversational, there were there were I believe tough questions there. I mean I I mentioned to him that you know, I'm puzzled by the supply chain risk designation.
I I mentioned to him if you were so close to being willing to work with Anthropic, um then how could they end up being a supply chain risk? On Mythos I mentioned, um wouldn't you want this tool at this disposal? You're putting your Are you Aren't you putting yourself in a corner when you're not taking these capabilities and using the and using the ones you want?
And then on on this idea that Anthropic could shut them off, You know, I think I I mentioned like you have Claude, you could have you are building a system with Claude, Grok, and OpenAI in there or you're on your way. Um and if Claude has an update you don't like, you wouldn't you'd just be able to run OpenAI on it.
So, why shut off the relationship? For me, I don't think I need to yell or sort of be like, you know, did why did you do this? Uh I think it there there is like basically, the what I want to do is to to ask what I'm curious about, get the toughest questions in front of these people, but also do it in the context of a conversation.
And if there's there I'm open to all forms of criticism. If there's like this was too conversational or the pushback wasn't good enough, I'm open it. But the, you know, I would say don't mistake the conversational nature for this idea that these people are getting a free ride because if you listen to any of the more recent uh interviews with Emil, uh I those questions were not asked.
So, that's just my piece and I thought it was worth saying. No, I I not to not to glaze Alex here a bit for listeners, but I think that is yet I think that's what I find valuable. And I do like that that like you can be it doesn't have to be confrontational. Like and any leader at that level, I feel should and I that's probably why the guests you're getting they're just getting kind of crazier and crazier good.
Like I mean, like if you're that level of success, you should want tough questions. You should be ready like that's I don't know like maybe it's being a former debater myself, but uh Yeah. You should want to. tough questions and most do. And if they they it like makes them better and smarter. And yeah, I think uh there's not enough of that out in the world.
So, so even for this listener I uh just please they like stars. shows. They like the Friday shows. Rundown, you're saving this thing from We got to disagree when we come back on Sunday. okay. on Allbirds. Um Allbirds. You got to say it's good. Yeah, but I don't want my my point here to be mistaken as please don't criticize me.
I'm open to criticism. I want to hear it. We have the email address bigtechnology [email protected] and I take the comments seriously because again like this is this is only possible because we have an audience and uh and your if you're listening or at home or watching your your opinion does really matter.
All right, let's take a break and come back and talk a little bit about Sam Altman. We'll be back right after this. And we're back here on Big Technology Podcast Friday edition. Wow, today is flown by. Let's talk about this story in the Wall Street Journal about Sam Altman's side hustles which is interesting because the headline was Sam Altman's side hustles blur the line between OpenAI's interest and his own.
That was not the story. Uh they they Wall Street Journal in this story about Sam Altman's conflicts which he he basically you know I'll just give it at a high level. He doesn't really take a story from salary from OpenAI but he has these investments in like the OpenAI startup fund and in other startups and he's tried to get other startups to OpenAI to fund some of these other startups like Helion which is a fusion company and Stoke Space which is a SpaceX challenger.
This is the thing that really stunned me in the article and I don't know why this wasn't the headline. OpenAI leaders and largest OpenAI's leaders and largest investors say they support Altman and crediting him with the company's success. Yet some shareholders have begun to privately question whether he should lead OpenAI through the turbulence of going public and have floated board chair and former Salesforce CEO Bret Taylor as a potential successor said people familiar with the matter.
Hold on, you're you're going to bury this notion that there are people there are investors on there are OpenAI investors who are so uncomfortable with Sam that they are already floating a specific name of a person to replace him. That to me is the story. Well, I think I mean, in a way though, it makes sense.
Like I think for all the problems that or all the challenges they're facing and they're not necessarily problems yet that have been realized like I feel like I mean, Bret Taylor, I mean, he's on the board of Open AI, CEO of Sierra. Yes. Like he's uh I mean, that's kind of like operator that Open AI still lacks and Fiji Simo was supposed to come in and be that operator.
Tony Streeter might be that operator more so, but like they still haven't found I mean, I hate the everyone Remember everyone used to say like their Sheryl Sandberg to Mark Zuckerberg like uh they're still looking for it. So, I think uh Yeah, it's it's not unreasonable. I think you're going to be seeing more of these kind of leaks at some point cuz there has to be investors, there's a lot of money in it, there's a lot of kind of lack of clear strategy still even after they're supposed to kind of bring focus to the company.
So, I think you're going to be seeing a lot more of this as we get into this IPO race and battle against Anthropic. Yes, I first of all, I just want to say I am mostly against this you need an adult in the room type of uh operator CEO. a Sheryl Sandberg. I think they just they take out the imagination of companies and I much prefer I think founder-led companies are where it's at and that sort of
No, founder-led with with the adult in the room I feel like Snapchat like they that's the one thing that always they never had. I think like uh overly founder-led. I'm trying to think Then you have like an Nvidia where founder-led but years of maturity and development and then they hit their stride in an insanely successful way.
So, yeah. I think Yeah. And a dominant room, isn't it? You're right, especially as as an IPO approaches, right? And by the way, Bret Taylor, a combination of Sierra and OpenAI would make so much sense, especially as OpenAI goes after goes after enterprises. It would be a perfect match. I would be shocked be stunned if they haven't talked about that within OpenAI.
Well, that's money that they have to invest. But hold on hold on Oh, yeah, you go ahead go ahead. Wait, but he's on the Isn't he the chair of the board? Yes. Yeah, how does that I can't even get it but I conflict of interest conflict of interest has never been a problem for her. There's no You know, we've seen the much weirder happen in this industry.
that That was the most [clears throat] naive thing I've ever said on this show. head out of 1995. We're in 2026. Okay, but but you're right. It's going to It is going to It is going to you know, sort of come to a head as this company goes to an IPO. And let me read on in the Wall Street Journal story.
Am I excited to be a public company CEO? 0%. Altman said on a podcast in December. Am I excited for OpenAI to be a public company? In some ways I am and in some ways I think it would be really annoying. He said it on a podcast in December. Hmm, which podcast was that Wall Street Journal? But yeah, it's an issue for for I mean, if you think broadly about OpenAI, it is an issue, right?
They They Let me just set the scene. I mean, they have they They are chasing Anthropic into some degree or even a large degree, right? They're they're pivoting towards this Anthropic model of the super app, the Fiji thing, you know, of course it was for medical reasons, but it's a pretty rough time for that for her to to be away.
And and Sam is is doing his thing. He's obviously raising a money a lot of money. That's critical, but but but it's not good to have questions about It's not good for your investors to be talking about a potential successor while you're with less than a year away from an initial public offering. I see I'm going to I'm going to push back on that.
Good. Do it. Do it. People have been talking about potential successor for a long time. Like again, Fiji gets announced, potential successor, is that the real move here? Like Brett to So So that part doesn't worry me as much. And again, as you said, wait. How much did they raise in the last round? 122 billion.
That's stupid. mean, if you're the CEO and your job is to get funds, he's doing an okay job, wouldn't you wouldn't you say? Wait. So So this was while I was out on vacation. And I was in Utah and we got some snow and I was able to ski and I actually didn't look at my phone too much and I like saw the headline.
I'm like, how but it it it kind of was just like a blip in the news cycle. Like 122 billion at 852 billion-dollar valuation. Like I don't know. How It's It's amazing. And they are chasing Anthropic strategically. They're going after enterprise again, like Mythos gets released and then 54 cyber gets released right after.
It's like everything they're doing is right after Anthropic and following them. Yet they freaking raised 122 billion and it's just become so par for the course now that that wasn't even the biggest story in the world or at least in the tech world. Yeah. No, I mean, you're right. I mean, and that's all Altman.
So uh I don't know if others can do it. He's doing his job. Yeah. Okay. We We are We are running up on time and we need to at least leave 7 minutes to talk about Allbirds. Okay. Can I Can I begin but just by I mean, this is one of those stories where you you see the news come in and then you just wait for Matt Levine at Bloomberg to weigh in.
So if if you will, allow me to just read Matt Levine. He says, I feel like if you ran a big technology company and you were looking to expand your artificial intelligence capabilities, and needed to rent access to GPUs, and you said uh and an a and a GPU as a service {slash} AI cloud company came to pitch you, and you said, "So, tell me a bit more about your company." And the company said, "Well, 2 weeks ago we were a sneaker company, but I've since pivoted to AI." You might say, "Huh, but thanks, but no thanks.
We're going to go with someone with a bit more AI expertise uh and actually a data center." Um but maybe that's wrong. Maybe the sneaker guys are great at AI, but you might worry. But, if you said, "So, tell me a bit about your company." And the company said, "Well, 2 weeks ago we were a sneaker company called Allbirds, but we have pivoted to AI." Your reaction might be different.
Because in this hypothetical, you run a big technology company, and you probably spent years wearing Allbirds. "I used to love Allbirds." you might say. High five. And then you might sign a long-term cloud hosting agreement with the former Allbirds because like many tech executives, you have a nostalgic fondness for their brand.
Um it's not likely, but it's worth a shot. This all sets up the fact that Allbirds uh pivoted to an AI computing infrastructure company this week. Their stock went up 582% on the day of, and somehow uh they are this is uh this is a real thing that happened. I mean, Pranjam, what the hell is this? This is ridiculous.
so I know we only have a few minutes left, so I'm going to try to be uh a little concise with this story. No, good. You you go. Go. You have 5 minutes. It's all yours. Allbirds, I have a very long relationship with the company in the sense that So, I worked for a number of years at Adore Me, a direct-to-consumer intimate apparel company.
In 2021, we were we were at like 250 million revenue. We were like looking at going public. Um and that was as Allbirds was going to market. This was There was actually 2021 summer into the fall was this kind of like DTC Rent the Runway. There's a number of other DDC companies that went public. And I loved this Matt Levine part because the bankers we spoke with like all wore Allbirds.
And they were the they were like the ones, the only ones. The VCs were the only ones there. And it actually you could tell affected Allbirds at kind of a similar slightly larger revenue went out to market at 2 billion on the first day popped to a 4 billion dollar val- valuation. It was like 20X 18X price to sales, which is insane.
Is insane. But the momentum of the people with money who do the investing loving the brand actually being the ones to push it. And the stock just collapsed. And it the reason this was so funny is like it became a penny stock. They essentially went bankrupt. They got bought out for I think 39 million dollars in the end.
Um I'm sure most people don't see anyone wearing Allbirds around. They shut down all their stores. But but I kind of like this ridiculous pivot. It's just It's like you have this asset. And that I I This is why Matt Levine is Matt Levine. He he got it. That is the asset. This nostalgic like it's almost like you're buying a marketing asset and a brand asset rather than you're just a nameless face- faceless GPU AI cloud as a service company or whatever.
And it obviously like the stock market it's almost bad that the stock popped 600% on the first day and then collapsed I think 580% the next day like No, no. It collapsed some like 80-something percent, but it's back it's back up. It's up 191% on the week. Yeah, but it's at a 105 million dollar market cap right now.
I just checked. I mean, but again, if your company is essentially worth zero, that's still a pretty good market cap. But I kind of like like buy up I I I I This is my new private equity idea for everyone. If anyone wants to run with it, go with it. Go buy any dead brand that investors and VCs and tech executives and bankers loved at some point.
Just take their name and turn it into GPU company as a service company. Whatever Take the most boring infrastructure business you can imagine. Go buy some What What do you think some good 2010s tech brands are that are that have left us? I'm trying to think what what else could be in Allbirds. But, that's my idea.
There's got to be a portfolio. Go get BuzzFeed. Go get BuzzFeed. BuzzFeed would be good. We work We work We work GPU company. Take all those 2010s and cuz I actually saw there is something around like like dead startups are actually able to still sell their Slack threads and email lists and stuff for training data for AI.
Such as horrifying. No, no, but let's take all of the all the graveyard of 2010s and bring it back. And just they're all going to be GPU as a service companies. But, that industry is so unsexy and there's so few names available right now that are memorable. And there's Slack on, WeWork, BuzzFeed, Allbirds, whatever else.
And that's the idea. Start your roll-up company. That That will turn the tide on AI. It's you take these beloved consumer brands, and you turn them into GPU companies, and all of a sudden you see the AI ratings in the polls just go through the roof, and Jensen starts nailing his his interviews, and the Pentagon and Anthropic and Anthropic and OpenAI make peace, and we all live in a better world.
That's peace. That's where we're going to end up if you follow this idea. We can only hope. We can only hope. All right. Let's end on this message. Ranjan, thank you again for coming on the show. Appreciate it. World peace is a good place to end. World peace. Thank you. To you, the listeners, we'll see you next time on Big Technology Podcast.
Comments 0