저자: Kevin · 외국인
요약 보기
Kevin은 All-In 팟캐스트의 OpenAI CFO Sarah Friar 인터뷰를 비판하며 OpenAI 수익성 및 대규모 컴퓨트 투자가 버블 형성 신호라고 주장. 빅테크 자사주 매입 축소(Meta·Google 1분기 0, Microsoft 감소)와 FINRA 마진 부채 증가로 유동성 부족을 근거로 듦. Anthropic이 코딩 시장에서 OpenAI를 앞서고 있다고 평가. Broadcom 매도는 과매도로 보며 410선 회복을 예상했다고 언급. 엔비디아·Marvell 등 반도체 하드웨어 선호를 표명하나 구체적 목표가는 없음.
다듬은 글
Hey everyone, me Kevin here. I have to say I'm really disappointed by this All-in podcast video because there are a lot of very critical components that were missed in this video. Open AI CFO Sarah Friar. Now, this is like Sam Altman's homie G who goes out there and shills for money and I have to say this video left me with such a bad taste in my mouth for Sarah Friar that if Open AI were public, I'd love to short it after that video.
Now, getting that bias out of the way, I don't have any kind of like creative options or stocks or whatever in this. I don't got any money in this. I just want to share some insights that I was a frustrated by, but also that are useful for us as investors, even if you don't care about Open AI and you're not going anywhere near that Open AI uh you know, funding round IPO when it comes, whatever.
This is also useful for chips, specifically chips like Marvell. Marvell, by the way, potentially getting to be eligible to get a spot in the S&P 500. And boy, oh boy, am I excited for Marvell because this has been one of the stocks that we've been shouting out as a hardware stock to play in the course member livestreams and damn, it went vertical.
Holy smokes, that is beautiful. There are a few things more beautiful than that. Maybe this bat, but not much more. All right, let's get into what actually happened in that video. So, the first thing that absolutely made me want to take my laptop and smash it against the damn wall was the fact that Sarah Friar says that there's actually a lot of available money sitting on the sidelines right now.
And she says the reason for that is because a lot of companies are doing a lot of stock buybacks, returning that capital to investors who could then recycle that elsewhere. And while that might be true of software stocks, if you haven't noticed software stocks have been going into the freaking toilet.
So, maybe it's true of really big mega cap stocks. So, in order to verify if that's true, we should jump into the actual earnings releases of companies like Microsoft and Google uh and uh and Meta. And when we look at those, we might be able to actually see, "Hey, what's really going on here with the numbers?
And is Sara Fryer right?" Because she's essentially arguing, "Hey, we could go out there and raise as much money as we want because the market has so much liquidity." And when I go look at Finra margin statistics, all I see is debt going up. Yeah, we just happened to squeeze a little bit more liquidity out of consumers.
By now, you could day trade without having $25,000 as a minimum in your account value, which is pretty coincidental that is happening right before the SpaceX IPO. Also not coincidental in my opinion that Elon Musk is probably going to delay the IPO to a Friday because usually a stock gets a pop from their set price up.
That also happened to Cerebrus, right? So, the set price is purposely set artificially low so you get a marketing pop. It's kind of like listing a house for sale under market value so you could say you got multiple offers, right? But, the beauty of doing it on a Friday is Elon basically gets to brag about that pop all freaking weekend long.
If there's a pop. But anyway, you go back to Finra margin statistics, we're skyrocketing on debt. So, I don't know about all this money around. So, it's certainly not here because people are skyrocketing the level of debt that they have, and they have less cash available in their accounts. Now, if I jump into some of the results over here, for example, we go to Meta.
We've got a balance sheet, we've got a revenue statement. What What we have over here at cash flows from operations? Now, I actually happen to like Meta. I happen to to think they're a relatively cheap stock, and I happen to think that advertising uh with artificial intelligence will be very powerful and rewarding in the future.
But, where do we sit on actual buybacks? Well, right here. Repurchases of common stock. Okay, great. So, in 2025, we pumped $12.7 billion back into the market. Now, how much money did we pump back? Oh, jeez. It's just a line. That's right. Zero. That would be in the 3 months basically Q1. So, in Q1, Meta did absolutely dog SH1T when it comes to buybacks.
Fine. Well, what about Google? Let's go to things. Google, here we go. Uh under financing activities, that's where you want to look. Repurchases of stock, quarter ending Q1 2025, these are inverted. Uh so, 2025 right here, Google rebought $15 billion of stock last year in Q1. How much did they send back to the stock market this year?
Holy smokes, you got to be freaking kidding me. A big fat zero. No wonder they're raising $85 billion in an upsized follow-on offering. Well, folks, maybe Microsoft will be different. Maybe. Shares repurchased in the 9 months ending 2025, we have purchases, sales, maturities. Where is it? Where's financing?
Here we go. Uh common stock issued, common stock repurchased. Here we go. Oh, look at that. Microsoft is actually pumping some dollahollas back into the market, which is nice. But, out of these three, it's the only one. And look at that. Common stock repurchased Q1 to Q1, lower than last year. And if we look at their actual free cash flow, we could see $46.6 billion from operations, plant, property, and equipment purchases almost 2 x last year, squeezing their free cash flow down to about $16 billion.
Still very impressive. I actually also happen to like Microsoft. But the point is, you're starting to SEE THE SQUEEZE, BABY. And Sara Fryer is implying a still so much money. And she's bragging about how these three IPOs are basically going to be bigger than anything we saw in 2001. And honestly, that makes me a little bit worried when we start comparing to 2001.
But let's stick to what else she said, or better yet, what she didn't say. She was asked about ChatGPT getting passed up by Anthropic, especially Anthropic filing their S1 before ChatGPT, uh and basically winning the IPO race. And her response was well, the Kazakh language is growing from Kazakhstan, and ChatGPT has become a noun and a verb.
And Codex is at 5 million users. Woohoo! Yeah. Go watch the video. That was her response to that question. And I'm like, wait, you didn't respond to the question at all. Which is crazy to me. But what she did imply is that right now they're trying to serve 50% consumers and 50% enterprise. And what she's really trying to do in a sly way is say, well, Anthropic is going after the corporate customer.
That's what she should have said. Anthropic's going after the corporate customer. We're trying to go after everyone. The problem is everybody's not very profitable. In fact, the very last thing she said on this interview was if she had to allocate tokens for profitability, she would do it for API all day long, which is a fancy way of saying she would rather sell to businesses than sell to consumers, which is probably exactly why they killed Sora, and now we get a lot fewer of those cat videos shooting machine guns and driving tanks.
Anyway, that said, Anthropic is kicking their butt because primarily they are focused on that coding user, which this is fantastic. We can generate so much more code. We're sending so many more commits to GitHub, potentially 14x the level of commits being sent to GitHub on a month month base on a monthly basis compared to last year, which is really insane.
Like we're booming in terms of AI code creation. Obviously, we still need people to review code. We'll talk more about that in a different video. I have a different video on that. But it's very interesting to me because her answer is really dedicated to, "Oh, well, we're trying to help everyone." And so rather than answer the question directly about how Anthropic is kicking their butt, which personally I happen to also think as a user it is, she did give us a little bit of a hint.
She suggested that free users on ChatGPT sent about seven queries a day, which queries could potentially be responses inside of a chat. So that's kind of like using a chat once a day and then having like seven replies or six replies, right? Uh the first paid tier, I think that's about nine bucks a month, about 15 queries.
Chat Plus at 20 dollars a month, about 3x the queries. And pro users about 11x the queries, which is interesting. Then she was asked about compute. Does 1 gigawatt of a data center equal 10 billion dollars of revenue? And I thought she was going to answer the damn question because she's like, "Oh, that's a great question.
Yeah, you know." I thought she was going to say something like, "Ah, that's that's a good approximation. If we have a 1 gigawatt data center, we can produce about 10 billion dollars of revenue." Wouldn't it be freaking great if she actually answered? Instead, her answer was, well, "Use the coupon code link down below over at meetkevin.com and get your exposure to Marvel stock and the other coupon or the other stocks that we're really interested in right now as well as our fundamental analysis onl.
We got a really big uh boost of uh of of updates coming out for the membership next week, as well. I'm not exactly sure what day yet. It should be Monday, might be Tuesday, might be Wednesday. So, we'll see, and we'll expire the coupon code marvel along with that. No, actually, the answer to 1 GW = $10 billion of revenue was actually Well, guys, you know, compute is actually very scarce right now.
Um and uh well, you know, Sam Altman's actually going to go cut the ribbon on a new data center. We're in the community talking to people, making sure that they don't know, or that they know that we're not trying to make their utility prices go up. And uh the economics are coming getting better, but we're really trying to show value to our customers.
Okay. This lady drives me nuts, and I'm pretty sure the people complimenting her in the comments are literally just ChatGPT bots. Anyway, so she didn't answer to the question on that. She did give a shout-out to Elon, saying he nailed it by building more compute. And honestly, he kind of did. Like, with Colossus 1 and 2, and even though Anthropic can rug pull that contract with SpaceX, kind of brilliant that xAI has that compute, and xAI you know, inside of SpaceX has the ability to sell that compute, even though there are a lot of older chips, like H100s, maybe Blackwells in the newer generation.
And even though I think in the long term, companies like Cloud Coreweave, and even this portion at SpaceX, I believe these are going to be long-term bag holders. Even though that is true, presently, it is smart, because Elon can now utilize those Anthropic contracts, providing compute for them to help pump the SpaceX IPO.
Kind of smart. So, we'll see. Still not a big fan of companies like Coreweave. I've been bagging on Coreweave since it was like 180 bucks, or what what whatever it was. I've been bagging on them for a while. Uh not the biggest fan because I do think they will end up being a back holder. Yeah, it was 187.
Uh much much more interested in a company like Marvell. And there's actually another hardware company we added uh to to our tracking list. And um uh and that we're buying. So, you can see that over at mekevin.com. But for now, Sam Altman then uh does go on to say that the uh biggest growth sector they're seeing right now is not actually coding, it's go-to-market marketing.
And I thought that was actually very interesting because it implies that they're selling more to people who are working on the SG&A side of businesses. So, basically the you know, selling the product side, marketing it, rather than the development side, which is really where Anthropic is winning. That's probably why they're giving away Codex credits.
Uh that would make a lot of sense. And the the other thing that you have to remember is Anthropic's kind of killing the game over there. So, I think ChatGPT's trying to pull off the well, you know, we're good at other things. Like this new AI device that is going to bring humanity to devices that we're co-developing with Jony Ive.
And I'm like, woman, you are smoking some crack. Who knows? Maybe that'll be the next big thing that'll track every movement of your life. And Jony Ive designed it. It's the shape of a job Okay, I'm I'm not going to get into it. Anyway, uh One thing that I did think was very interesting, this I thought was very insightful.
Maybe I should have led with this because you know, we probably already lost like 20% of viewers who are like, "Ah, I don't know. I don't want to watch the rest of the video." But that sucks for them because then they miss the nuggets like right now you're about to get. They mentioned that from between ChatGPT 4.0 and 5.4, they've actually seen a negative 97% cost reduction in 2 years.
Which actually means that not only are chips getting more efficient, which we know is happening, but it also means their token costs on like if well, maybe not their token cost, but their token efficiency is getting better. Which is really, really, really important. The problem is she's literally trying to buy compute for 2028, and we're also projecting out to 2031 and 32.
And when she was asked by the All-In podcast crew, which I think they could have been a little more aggressive with her, but I think they wanted to be nice to her. When they were asked when she was asked bluntly, "Hey, well, like how do you project revenue out that far?" And she goes, "Well, I mean, we can project a year out, but after that, we're just sort of expecting that revenue will come in from maybe products that we haven't even developed yet." And I'm like, "Wait, you're projecting your compute expenses.
So, you're trying to secure compute data center compute for 28, 9, 30, 31, and 32 with a concept that maybe you will have some new idea that maybe will generate revenue. And you are using that concept of a concept of revenue to go buy compute. Holy smokes. This is going to be the mother of all fracking bubbles.
Now, in fairness, I'm not of the mindset that the bubble is ready to blow, but we are certainly blowing. We are blow We are inflating the bubble. Uh now, again, we mentioned , we don't actually see a sign of peaking yet. I literally wrote in the earnings call this morning, and this is the kind of fundamental analysis that we do in the course member livestreams.
Which, you know, most people don't do on their own, which is fine because we have busy lives, but that's why a lot of people tune into the course member livestreams, and they're like, "Oh, that's really cool insight, Kevin. Thanks for doing the fundamental analysis." Like Broadcom right here, thank you for pointing out that we expect AI semiconductor revenue growth to continue in fiscal 2028 based on the following initiatives we have with our six core customers.
In other words, no signs of the six core customers at Broadcom peaking out yet. Which, in part, is probably because Sarah Friar is projecting compute usage on a concept of revenue that they actually have no idea that they're going to get. Insane. They do also talk about using multiple chip partners.
It's not just Nvidia, it's trying to get that Broadcom partnership, Cerebras chips, AMD chips. Uh one thing to remember is Google is partnering with Marvell, which is also really exciting for the custom uh chip chips that's Meta is partnering with Broadcom. I personally think that, and I said it this morning, that the Broadcom sell-off was oversold.
I said that in the course member livestream as well. I said I think I'm going to we're going to see this recapture 414. It's just giving up the fat over here. And sure enough, we recaptured 414, which is a very good for Broadcom, very good for the chip market as well. But really an overblown sell-off.
Unfortunately, eventually it will prove to be a massive bubble, because you have morons like Sarah Friar. Sorry, I have nothing good to say after that interview. Uh who are literally demanding and spending money and making memory or chip commitments on concepts of concepts of revenue. And that, in my opinion, is full-on tarding.
Anyway. That's my angry video for the day. I'm now going to go on vacation. So, .
Comments 0