MRVL 全部观点

Marll is coming on very strong. Broadcom is coming on very strong. He's comp Let's take Marll. Marll is building custom AS6 for customer. Amazon's their biggest customer. So Amazon's building tranium chips which are application specific integrated circuits. They do not have as broad of a use as a GPU. They're not as powerful. They're not clustered the same way. you know, you take a take a take a cluster of 10,000 chips and and put them in. It's not the way that's being done. But Amazon its tranium line of chips, Alphabet's making chips, Microsoft is making chips, Meta, Apple is making its own chips. They're they're utilizing application specific integrated circuits. And that custom chip making for the hyperscalers is coming out of Marll. This doesn't mean Amazon's not buying GPUs. doesn't mean uh Gemini is running without GPUs, but it is new competition for the buildout of compute. Why would they want to use AS6? Well, if they build them themselves, they can customize them for exactly the uses within their data centers that they see as being important. Okay, so that's what Marll is doing.

And a lot of people either aren't doing this long enough to have learned that or forgot about it. But I do think there's a gravity that is pushing down on the multiples of Marll, of Nvidia, of Broadcom. They understand that these companies are going to go into a phase where they rip each other's throats out for the next upgrade cycle and not everybody can win and that is why they're not selling at 30 or 40 times earnings the way that they were three years ago.

But I do think there's a gravity that is pushing down on the multiples of Marll, of Nvidia, of Broadcom. They understand that these companies are going to go into a phase where they rip each other's throats out for the next upgrade cycle and not everybody can win and that is why they're not selling at 30 or 40 times earnings the way that they were three years ago.

Remember, when humans are faced with a challenge or shortfall or whatever, they fix it because money's on the line and that's what's called capitalism. Capitalism solves problems, hard problems. These guys are doing that, too. So, I don't expect that to to slow down and also remember Jensen Huang said this is the next trillion-dollar company. It's currently trading a market cap of about 200 billion dollars, so that's an easy 5x for the traders out there.

Just like I mentioned before with Marvel, they're capitalists. They're figuring out ways to solve problems like drought shortage.

So what does that mean? Well, what that means is you might end up seeing companies preferring A6 from Broadcom, Google TPUs, which are expecting to scale next year, uh, or Marll A6, uh, or Nvidia 5090s or 6000s over the bleeding edge Vera Rubin chips.

Another set of companies to consider looking at is going to be Broadcom and Marvell. And in fact, Marvell also reports this week, later in the week. These companies are more nuanced because Broadcom and Marvell benefit from AI spending, but it also represents one of Nvidia's biggest long-term threats. Why? Custom silicon. Google, Meta, OpenAI, other hyperscalers, these companies increasingly want AI accelerators designed specifically for their own workloads. Broadcom and Marvell help enable those chips. So, if Nvidia says overall AI infrastructure spending continues exploding, that's great for both Broadcom and Marvell. But if Jensen starts discussing competitive pressures from custom accelerators, that's another important positive signal for both of these companies. This is why I own exposure to both sides of the equation. I own Nvidia, AMD, and then Broadcom and Marvell, but position sizing is different for all of them. Nvidia benefits from general-purpose accelerated computing. Broadcom and Marvell benefits as hyperscalers increasingly design specialized silicon. I don't need to perfectly predict which architecture wins.

This is why I own exposure to both sides of the equation. I own Nvidia, AMD, and then Broadcom and Marvell, but position sizing is different for all of them. Nvidia benefits from general-purpose accelerated computing. Broadcom and Marvell benefits as hyperscalers increasingly design specialized silicon. I don't need to perfectly predict which architecture wins.

So, then we move on to Credo. And really, you could throw Marvell into this group as well. This is where investors need to understand something. You cannot build enormous AI clusters by simply adding more GPUs. Those GPUs have to communicate. And as clusters scale from thousands to tens of thousands to potentially hundreds of thousands of accelerators, networking becomes exponentially more important. If Nvidia continues talking about larger AI factories, rack-scale systems, and enormous clusters, that's bullish for the networking demand. Marvell reports the very next day after Nvidia, while Broadcom follows the next week. So, Nvidia's commentary could effectively set the tone for both reports. That's why I'll be listening closely to Jensen's discussion around networking.

Marvell reports the very next day after Nvidia, while Broadcom follows the next week. So, Nvidia's commentary could effectively set the tone for both reports. That's why I'll be listening closely to Jensen's discussion around networking.

So, with that, there's really two scenarios that could take place after the report on Wednesday night. Scenario one, Nvidia crushes it. Revenue beats, EPS beats, margins hold, Q3 guidance comes in well above 104 billion. Ruben remains on schedule, and Jensen says demand continues exceeding supply. If we get that combination, I think you could see a significant relief rally across AI, Nvidia, AMD, Taiwan Semi, Broadcom, Marvell, Credo, Micron, potentially even AI infrastructure and data center names because the market gets confirmation the AI spending cycle isn't over. And to be clear, this is the scenario I'm not only hoping for, but also it's my expectation.

So, in conclusion, Nvidia reports Wednesday, and this is probably the most important earnings report remaining in the season. I'm watching 92 plus billion in revenue, $2.09 EPS, hoping for higher, 75% gross margins, and Q3 guidance versus roughly $104 billion. What's the take on rumor demand? Is there any sales heading into China? And most importantly, Jensen Huang's commentary around infrastructure spending. Because if Nvidia tells us demand remains extraordinary, that doesn't just strengthen my Nvidia thesis, it strengthens my conviction across much of the AI investment stack. AMD, Taiwan Semi, Broadcom, Marvell, Credo, Micron, and potentially many more. But if we start seeing cracks, I'm not going to ignore them simply because I'm bullish on AI. That's the entire purpose of following earnings. Our thesis should follow the fundamentals, not the other way around. And Wednesday night, we're going to get one of the most important updates of those fundamentals we receive all year.

Well, Marvell is jumping to the top of the list of Google as they signed a chip deal including issuing warrants to the company. We've seen this over at AMD. They had issue a huge chunk of the company to open AI and Meta. Marvell doing the same with Google. This gives Google obviously a huge incentive to deal with Marvell because they essentially are allowed the right to buy shares of Marvell at just $206 per share. I think don't quote me on this, but I think it was about 6% of the company. So if Marvell stock continues to rise and currently it's at 237, well Google of course would exercise those rights cuz they're getting the stock at a tremendous discount. Also obviously incentivizes them to buy from Marvell to juice their revenue, to juice the stock price, and just make those warrants worth even more. You do have to factor that in as a shareholder as those rights essentially dilute you as a shareholder and in some ways can be a very expensive way to sell GPUs in the hottest chip market that we've ever seen.

Marvell actually has some interesting technology for chips that are believed to have co-packaged optics by 2027, probably more like 2028. And obviously having Google on board building custom versions of that could give them the volumes and the quantities that they need.