Reactions: Ferrari’s First EV, The Enhanced Games | Diet TBPN

0:00John Coogan:Big news of the day of the week, this has been going on all week, is that ARM, the intellectual property developer that creates intellectual property designs for CPUs, is now getting into the chip game. And they got a big ARM pump in the stock market. So the company is up 15 % over the last few days on the news that they will sell their own chips. This is new for ARM. ARM's a very old company. Fascinating history. I actually made a 25-minute YouTube video all about the history of the company back in 2023. But we'll recap a little bit of it today. They normally just license out their intellectual property.

0:38John Coogan:And that is a phenomenal business. 97 % gross margins. 97 % gross margins. Yeah. Let's give it up. There we go. That's amazing. And, you know, it's a big business. $4 billion in revenue last year, nearly$800 million of net income. This new move, they're expecting to ramp revenue to$15 billion by 2031. So they're expanding the market significantly. Now margins will be different. But the market cap for Arm is now around$166 billion. So it's a big company. It trades at a very high multiple. Yeah,$4 billion last year, currently trading at$165 billion. But very, very high gross margin. This is a big shift in strategy.

1:19John Coogan:Arm's not an AI loser by any means. but it hasn't gotten the attention that other GPU makers have received, like NVIDIA. CPUs are far from dead, though. In fact, we are currently in what seems like a little bit of a CPU crunch. Intel can't make CPUs fast enough. NVIDIA is starting to sell their Grace CPU that goes with their hopper. So you get the H100 and you pair it with the Grace CPU. You get the GPU and the CPU all on one system. Well, now you can buy the CPU by itself if you're CPU constrained. So you don't You don't want to just be GPU rich and CPU poor. You got to be rich in both camps.

1:56John Coogan:And a lot of companies are jumping in to fill this gap. Agents, and a lot of this is because of agents. Agents need CPUs. You need to fill the GPUs constantly with new data and tasks. Also, all of the agents use CPUs to make web queries, search the web, run Python, spin up web servers, interact with anything. uptime is going down for non-GPU accelerated workloads. Like you go to a SaaS product that is basically just a web server that's running on a CPU somewhere in a data center and you're like, oh, this thing isn't loading today. Or like there's downtime. And a lot of that's just because we're writing more software, we're using more software as a society, as a country, and we need more CPUs as well as GPUs, even though GPUs are like the hot thing to talk about.

2:42John Coogan:So ARM has a very interesting backstory, But you can think about it basically as a joint venture between three groups Apple Acorn computer and VLSI And so they needed to design a low-power CPU for mobile devices before phones You know what was hot Jordy PDAs and have you ever seen a PDA Tyler? No, you've never seen a PDA Okay, so before the iPhone, you know everyone had flip phones, but before is not a pager. It's not a pager You don't know what this is either? No. Wow. So there's something called a Palm Pilot. And basically it was iPhone size screen and it has stylus and you could write on it and you could make notes.

3:22John Coogan:And PDA stood for personal digital assistant. And I feel like that brand is just itching to come back in 2026. Everyone is working on a PDA. Yeah. Personal super intelligence assistants. There's a whole bunch of things. We're building PDAs, folks. They just live in the cloud and it's not a physical device. But these PDAs back in the 90s were physical devices. This was post-pager, pre-smartphone, and it was the device that you carried to take notes and do things that you do today in apps, you would do on your PDA. But it wasn't always on, internet connected, none of that. It was a Palm Pilot, and this was like a fantastic business for a while.

4:01John Coogan:But there were a lot of problems with this because for the first time, you needed a CPU that could live within a plastic shell, basically. These were like plastic enclosures. You need to run a battery. You needed to be able to do some things, not crazy compute. But the CPU industry in the 90s was very focused on mainframes, servers, desktops. There were a few laptops popping up, but it was not the mobile phone revolution that's happening now, where you have Apple Silicon chips, and those are, of course, based on ARM architecture. But that's where all this came from. People said, okay, we need a lower power chip that can actually run off a battery, not overheat and melt the plastic, do all of this, and it can power a device that you carry with you as a personal digital assistant, a PDA.

4:48Chicken in the chat says PayPal started on PDAs. That's right.

4:52John Coogan:That's right. So PayPal started, originally the idea was these PDAs had, they didn't have like tap to pay or anything like that, RFID. They had basically the same device that you'd see on a TV remote, so IR. And it would flash a light that could be seen by another sensor. And if you flash the light at a certain rate, you can send a message. You can basically, it's like advanced. What's that SOS thing? Morse. Morse code. It's like a more advanced version of Morse code. And so you could send a specific packet of information from PDA to PDA, and this was the original idea for PayPal. That's correct.

5:30That's a great, great piece of lore, tech lore.

5:33John Coogan:ARM starts to build these. Later, there's Robin Saxby, who was the CEO of ARM at the time. He wanted ARM to become the global standard for CPUs. And so in the 90s, there were lots of different CPU makers, lots of different architectures. There's this whole CISC versus RISC debate. There's the x86 architecture that Intel pioneers. And slight differences in architectures can limit interoperability. You see this with what's going on with Mac, where a whole bunch of applications have needed to be rewritten for Apple Silicon, because previously you would have an Intel Mac, before we got to the M1, M2, M3, M4, M5 chips.

6:13John Coogan:Those are Apple Silicon. Those are ARM-based. And sometimes you'll go to a website and it'll be like, oh, do you want to download this for a Mac? Like, what do you have? Do you have an Intel Mac or do you have an Apple Silicon Mac? Well, it's important because when you write the software that runs on the Mac, you need to use specific instruction sets. Now, there are ways to abstract that and run on either, but there are lots of pieces of software that interact with the CPU at a low enough level that they need to be aware of the instruction set. So ARM sets out to be the global standard for CPUs, and they create the ISA, the instruction set architecture, and that ultimately let Apple design their own chips, but within the architectural guidelines set forth by ARM.

6:53John Coogan:So Apple pays a license to ARM for every chip Apple sells. It's a very small license fee because Apple does a lot of design work. They do the manufacturing, the TSMC fabs it, and there's a million other pieces of the value chain. But for this one little slice, they have to pay Arm. And Arm just takes that and says, thanks. Cool. You used our intellectual property successfully. You know who else says thanks? Who? Masayoshi Son. That's true. Because they own 90%, roughly 90 % of the company. It was a full buyout in 2016. That's right. They bought the entire company for something around$25,$30 billion USD.

7:32Buy the entire company. They still own 90 % today. So their like SoftBank's holdings just in that one company are somewhere in the range of$140 billion. It's great. What I was looking at. It's the second time he's made$100 billion. Well, and they've massively levered up against that position. Yes. They've raised debt against their holdings in ARM. but certainly Mazda is somewhere out in the world seeing it go up 15%, just smashing a gong.

8:08John Coogan:I hope so, I hope so. Although everyone knows Apple, my clickbait for my YouTube video was like, did you know the iPhone is secretly British? I think it's funny because it uses ARM deep inside, but ARM licenses to a ton of different tech companies. So Amazon uses ARM for the Graviton chips. A lot of Android phone makers have ARM-based chips. tons of other tech companies license ARM's technology to build chips. But now ARM is going to be making the chips themselves, and they're going to be working with meta platforms and OpenAI. This means a shift in the economics of the business. So 97 % gross margins for just those licensing ISA contracts.

8:47John Coogan:This will be closer to 50%, but it should be offset by the huge gains in market size and revenue potential. So the company has one of the highest multiples in semiconductors, roughly 90 times forward earnings. So there's a lot to live up to, but there's a lot of value coming from AI agents that have access to plentiful CPU resources. The industry dynamics are also particularly interesting. Ben Thompson pointed this out in Stratechery. So NVIDIA sells an ARM-based CPU, that grace CPU. NVIDIA is sort of competing with ARM. Jensen showed up and gave like a remote talk at this ARM event where they announced this chip and so they're competing but they're also in some ways working together because they're challenging x86 options from Intel and AMD that are still really popular.

9:31John Coogan:If Nvidia sells a bunch of chips for, you know, AI workloads, well then that actually makes ARM more likely to sell their chips as well because whatever software is built for the Nvidia ARM-based chips will probably run on the ARM ARM-based chips. So there's some sort of like integration there. The x86 moat is not as as the CUDA mode, but there's still this dynamic of NVIDIA and ARM are going up against Intel and AMD in the same way that different GPU makers are going up against the CUDA mode. So there's all these different interactions there. But it'll all be interesting to follow. Let's move on.

10:09John Coogan:The big news that drove this was of course Meta, engineering and medicine. Today we're announcing a new partnership with ARM to collaborate on the development of multiple generations of purpose-built CPUs to support compute and AI infrastructure. Arm called it the Arm AGI CPU. What a great name. The Wall Street Journal has another good write-up of this story, talking about how the timing is good. There is a boom in CPUs right now, but the stock is already priced very highly, and everything has to go perfectly because it trades at one of the highest multiples of all the semiconductor companies.

10:43Let's talk about this data center moratorium bill. Yes. I wrote in the newsletter today, yesterday, Senator Bernie Sanders and AOC introduced a new bill, the AI Data Center Moratorium Act of 2026, that if enacted, would require all current and planned data centers to halt construction slash production. It would even block upgrading existing data centers. So if you have an asset and you want to make changes to it, as this bill is written today, it would be blocked. they sort of like define data centers based on power demands, cooling capabilities, like how much power you can get to each individual rack.

11:21So they have been fairly specific, but trying to be like... I was thinking I was fine with this.

11:26John Coogan:I was thinking like, I don't need any more AI data centers at this point. We can freeze those. I just want to build AGI data centers and ASI data centers. And so as long as I can just build tons of those, it should be fine. but it is interesting that they seem to have figured out the semantic loopholes that might happen if it's defunding AI. Yeah, the bill would halt all new data center construction and upgrades until more legislation is put in place to guarantee the following, and these will be tough to guarantee. So from Sanders' site, they want AI to be safe and effective, preventing executives in the AI industry from releasing harmful products into the world that threaten the health and well-being of working families, our privacy and civil rights and the future of humanity.

12:08The economic gains of AI and robotics will benefit workers, not just the wealthy owners of big tech. And AI does not increase electricity or utility prices, harm communities, or destroy the environment.

12:20John Coogan:So this stuff seems good. Yeah, all generally good. But no one wants unsafe and ineffective AI. Well, yeah, and the bigger problem is any time you're creating. I don't think we've created a technology ever that didn't have some negative impact. Car crashes, for example. I'm sure this will be rewritten and debated, and obviously it has a long way to go before becoming law. But this set of requirements seems completely impossible to actually achieve. It depends on how you define it. I mean, AI does not increase electricity or utility prices. like the rate payer protection pledge and behind the meter stuff could.

13:00John Coogan:I'm more talking about number one. Number one, safe and effective. It's all in how you define that. Like some of the parental controls are a good example of like how to take that overarching thesis and then boil it down into something tractable. And when I hear that, I think like, oh, like I don't know how we are defining safe. I don't know how we're defining effective. Like this feels like this could be like some sort of very vague thing where like if one particular administration likes this company, they just approve it or not or whatever. But then when we actually talk to lawmakers and you hear something like, oh, yeah, we're going to, you know, require parental controls.

13:39John Coogan:So if your offspring has an AI account, you can say, hey, like they are this age. Don't show them anything that's inappropriate for that age. That seems good. Yeah. So I don't know. Overall, at least this first bullet point preventing executives in the AI industry from releasing harmful products in the world, that feels like you could end up having something like an FDA that's like every product you create needs to go through years of studies in order for the government. And it's like, hey, I just wanted to create like a slightly more AI native version of the SaaS tool. Yeah. Do I really want to make SaaS?

14:19I just want to make sass.

14:21John Coogan:Yeah, at the same time, like the bull case, I mean, I think the FDA model would really slow things down based on how long the FDA takes to approve things. At the same time, what is the definition of harmful here? Is it net harmful products? Because that is the goal of the FDA. They release drugs all the time that have side effects. You take this, it cures you cancer, but it's going to make you throw up or it's going to make you lose your hair. And people are like, yeah, I'll take that trade. And so if you went through the government, you said, okay, yeah, I'm going to give you this tool that can write code, but sometimes it's going to hallucinate and you might get some code that doesn't pass tests.

14:56John Coogan:I'd be like, yeah, okay. It speeds me up. On average, I'm in. That's fine. And having some of those disclosures is the same thing with knowledge retrieval. I do a deep research report. I get something that's 99 % of the way there. Maybe there's something in there that's like, oh, that's actually misattributed. or that number's written. I know that that number's from this report online that was wrong and the model doesn't. And so I need to fact check it. Like, I still see that as like net beneficial, but there are of course, like, like flaws in every system. And so, you know, again, it's like, where, how does this get defined over time?

15:32John Coogan:That's like important. One thing I noticed from the announcement was that they are using AI leaders own statements against them. And it's easy to see how this would resonate with their constituents. This is really powerful. So on Sanders' websites, he included this quote. Okay. In December, Elon Musk, who leads XAI, said he had, quote, a lot of AI nightmares and would, quote, certainly slow down AI and robotics if he could. It's so interesting because Elon doesn't talk about that with, like, the rollout of electric cars or the rollout of space travel. He's not saying, like, oh, yeah, like, you know, 2030 is too soon to get to Mars.

16:07John Coogan:We need to slow down on the race to the moon. Like, let's really figure out the spacesuits first. You know, he's like, let's just go. And then another one. In January, Demis, the head of Google's DeepMind, said he would support an AI pause if he knew other countries and companies also paused development. In February, Dari Amadei, the head of Anthropics, said he was absolutely in favor of trying to slow down AI development if other countries also slowed down. That was Davos, I believe. Both of those in January and February. So continuing, I wrote, but the problem, of course, is that there is zero movement on getting other countries to slow down.

16:42I can imagine some companies that would be like. There's already comments in the chat about like, yeah, let China win. Yeah. I'm not going to name the countries that would be down to slow down. But I think we all know that China, even if they agreed to something like this, wouldn't just automatically do anything about it. But the problem, of course.

17:00John Coogan:Are there any countries that are like, yeah, we definitely should. We're ready to slow down. Like we're France. Well, I think if you're way behind, there's kind of a benefit. Elon had said, I think, in 2023 that he supported like a six-month pause. At that moment, that would have been awesome. Because if I could just have six months to like get my team together. I feel like if you pulled people in the south of France or the Amalfi Coast, like those folks would say we should just slow down generally. Like AI, but also just slow down our lives, enjoy a glass of wine, hit the boats. Yeah, or even like a summer break.

17:32Just a summer break. kind of like four weeks.

17:34John Coogan:Yeah, or even like during the workday, like taking a break, taking a nap, just taking, just slowing down generally. I think there's a lot of people that are just in favor of that. Doesn't that mean that China should be more in favor of slowing down? That is interesting. Well, I'm just saying I can imagine in the same way that like for as many chips as we give them, advanced GPUs as we give them, you can assume they're still going to put an immense amount of pressure to kind of stimulate the local semiconductor industry. In the same way here, I'm sure they would love for the United States to just pause all new data center construction.

18:07I think it's possible that they would generally say, yeah, this seems good, but then what would they actually do? They would just use that as an opportunity to catch up.

18:18John Coogan:So my question is, where does China actually stand on this? I'd be very interested to know. Maybe you could look it up. Has the Chinese Communist Party actually put out any statements about whether they want to accelerate or pause AI development? because I think that they might refrain from taking that stance because it would discourage local indigenous development. If the government is coming out and saying, we want to slow down, then a lot of entrepreneurs are going to be like, okay, I'll go back to e-commerce or I'll go back to manufacturing. I'm not going to work on this because the government doesn't want me to.

18:52John Coogan:And so I feel like there's this tug. Even though I agree with you, it would be in their advantage to say, hey, we want to slow down, everyone should slow down, we're pro-slowdown. If they actually said that, it would have an immediate slowdown effect on the local AI progress. Does that make sense? Yeah, because even if they just take that stance, because it's an authoritarian country, like this, like the Bernie Sanders comment, stands in opposition to other politicians who are saying, like, no, actually, things are going well, and here's how we're going to, you know, like, advance energy and to build more.

19:25John Coogan:But if you don't have that and it comes down as like a dictate, like this is the stance from the government, it's much harder for local entrepreneurs and local AI labs to like push back against that because it feels like they're all of a sudden in opposition to the government. So just to kind of finish my thought, all these quotes must go extremely hard if you're not kind of acknowledging the full picture, which is that, yeah, leaders are saying, yeah, if you get other countries to agree to slow down, we'd be open to it. But that is like the big elephant in the room. They don't mention any even conversations or dialogue with other countries around slowing down, and I don't think there's been any.

20:03So anyways, the act has a long way to go, and it seems like the odds of it getting into law are low, but not zero. It's safe to say that, as written, the requirements in the bill would be an incredible gift to America's adversaries and catastrophic for overall AI progress. The question becomes, if anything like this were to become law, what are the effects of that, right? Space, right? The space data center people are saying, like, space data centers don't seem so silly now. Yeah. Taking that angle. Totally. Although I'm sure they would also be like, you can't put them up there either. We're going to try to come over the top.

20:39John Coogan:There's also just the globalization process that happened based around environmentalism in the 90s WTO ascension for China. The reason that a lot of the mining happened in China is because we said, we don't want that here. It's dirty, it's gross, there's chemicals, there's pollution. And so out of sight, out of mind, let's push it abroad. And we could do that again with data centers. We could just be like, they're all in Canada. or Mexico, or they're all in some Australia, or there could be a receptive country out there that just says, we are an ally now, and we'd love to get all these data centers.