원문 보기
Broadcom shares fell about 6% on Friday. AMD ended the day up 6%. By the weekend, the generally accepted explanation was that AMD was taking over Broadcom's business of making specialized chips for Google. Google has begun to abandon Broadcom services. But here's the thing. There 's actually nothing like that in the research note that everyone was referring to.
In today's video, I'll break down what was actually in the Semianalysis report. I was able to read it. Why the paragraph about AMD doesn't mean what everyone thinks. Besides, there's only one event in the next 60 days that really matters for Broadcom, and it's not the earnings report in a few weeks.
By the end of the video, you'll have a better understanding of Broadcom's specialized chip business and why , in my opinion, the company is still best positioned to advance AI. So let's start with the basics. Google has been developing its own specialized chips for data centers for almost a decade. The only other hyperscaler that comes even close to them is Amazon.
Google calls these specialized chips TPUs. While Google has enormous resources and expertise in developing such chips, they lack certain intellectual property and, more importantly, production quotas at Taiwan Semiconductor. Many investors don't understand that chip manufacturing at Taiwan Semiconductor is a kind of "club for the in-laws." This is partly because Taiwanese culture values relationships, loyalty, and trust.
After all, if you have been cooperating with Taiwan Semiconductor for a long time , you get production capacity that exceeds the capabilities of your competitors. In addition, Broadcom owns patents. They have intellectual property in chip design and networking technology that Google simply doesn't have.
So, it's a great partnership that's been going on for almost a decade. And that decade-long partnership is what many thought suddenly ended on Friday. But that's not what the Semianalysis memo was really about. First, Google expects to receive fewer chips from Broadcom in 2026 due to difficulties in increasing the supply of substrates for chips on the wafer.
Chip-on-wafer substrate or CoWoS is a packaging process at Taiwan Semiconductor. And look, it's complicated. We don't have to go into detail. You can search on Google if you want to understand more. I definitely did that. Moreover, Nvidia has been very aggressively purchasing capacity for this Chip-on-wafer process from Taiwan Semiconductor for several years, so there simply is n't enough of it for everyone.
But the facts are these: Broadcom is cutting back on shipments to Google due to supply issues. Broadcom simply can't get enough manufacturing capacity from the only company capable of providing it. No one at Google cancels anything. Now, the SemiAnalysis report also reports that Broadcom's TPU volumes for Google will be reduced next year, in 2027.
This decrease in output is due to a greater allocation of resources in favor of other customers. This is because Broadcom redirects some of its production quotas received from TSMC to other customers like Meta, Anthropic, and OpenAI. So, Broadcom's TPU figure for Google for 2027 has decreased because Broadcom is transferring some of the capacity to other customers.
She doesn't lose any orders. It simply does not have enough capacity to serve all customers at the same time. She would really like it, believe me. So the company made a choice about who to serve first . When I started recommending Broadcom to clients back in 2023, when the stock was $80, one of the biggest criticisms was that they only had one customer—Google.
It was really a big risk. Several years have passed, and it seems downright comical that the market is selling off Broadcom shares because the company is diversifying its customer base away from relying solely on Google. But that's exactly what's happening, and that's a good thing, not a bad thing. Now let's talk about AMD, as this news also affected both stocks on Friday.
I'll read you the exact excerpt from the SemiAnalysis report, and then we'll come back and discuss what it means. Market rumors suggest that Google is working with AMD on a 10th generation TPU project . AMD's participation would be the first real involvement in a dedicated AI ASIC project, despite having a team developing custom chips.
AMD has strong intellectual property, especially in the area of advanced packaging. The additional IP of the processors could be a big selling point for Google as their customers demand TPUs with processor cores on the chassis. Now let's figure out what this memo means for AMD and Broadcom. First, SemiAnalysis refers to “ market rumors.” These are their words, not mine.
In the financial world, this is probably the weakest way to confirm information. This means that someone heard something. Are we in second grade, guys ? Here's what I think about this. Someone wants to influence the stock price of AMD or Broadcom. So they start rumors, knowing that SemiAnalysis will write about it as a fait accompli and will not do any additional verification.
As for using AMD because of their packaging technologies and intellectual property on processors, I think it's actually quite interesting. But keep in mind: Back at the beginning of the year, Google posted on its blog that it had integrated ARM processors directly into its eighth-generation TPU system to eliminate what they called the host “bottleneck.” In simple terms: the processor controls the chip.
It was too far away from other processes, and that distance cost Google some performance. So, to put it even more simply, Google is already using processor technology. It licenses them from a competitor or another company. In the case of ARM, it is likely that Google wants to diversify and perhaps try out the x86 architecture, which is a technology of AMD and Intel.
But here's my favorite part, and it's really the most important one for you as an investor. AMD is buying Tomahawk network switches from Broadcom to build its own proprietary AI racks . So, the company that is supposedly replacing Broadcom in Google is actually now a Broadcom customer in the networking segment, and AMD will not have the technology to replace Broadcom anytime soon.
So let's be 100% honest about something. Broadcom has a dominant position in specialty semiconductors and, more importantly, in networking technologies. This will not change. Another important point. Google's 10th generation TPU has been reported since June of this year, and the partners named in its production include MediaTek, Samsung, and, of course, Taiwan Semiconductor.
So if AMD joins the project, it will most likely be as an addition to one where there are already several partners. AMD takes no one's place or stake at the negotiating table. That's what I think is really important about this story, and it has nothing to do with who designs the chip. Look at who Broadcom brought on as customers earlier this year .
Google, of course, extended the long-term agreement in April. Meta also in April— first over a gigawatt, and by 2028— three gigawatts. OpenAI—10 gigawatts of specialized accelerators. Anthropic, which has already placed orders worth $21 billion over two quarters. And finally, perhaps the best customer of all—Apple, which needed more than $30 billion worth of specialized semiconductors in July .
That's the truth. This is the largest customer diversification Broadcom has ever had in its business history. But here's the part that almost no one talks about. In June, Broadcom arranged $35 billion in financing from Apollo and Blackstone specifically to fund AI labs. Broadcom will guarantee these lease payments for about 5 years.
So Broadcom doesn't just sell chips to these AI labs. She helps organize the money they need to buy these chips. Nvidia has also signed a bunch of such deals. In the industry, this is called cyclical financing, but we have what we have. AMD has two, but here's the main thing to understand about AMD's funding.
They actually had to give away significant AMD ownership stakes to both OpenAI and Meta to secure these deals. Whereas Nvidia and Broadcom are essentially just guarantors of the loan. I would be much better off in Nvidia and Broadcom's shoes. Now that brings me to the event I'm keeping a close eye on over the next 60 days.
Anthropic confidentially filed for an IPO on June 1. The Financial Times just reported last week that investors are targeting a valuation of around $2 trillion in an IPO expected in October, which would be the largest IPO in history. This means that one of Broadcom's specialized chip manufacturing customers is about to receive a lot of money.
So here's the summary. Anthropic's IPO and the commercial success of its models, which is undeniable, is an event that will matter to Broadcom in the next 60 days. Not a paragraph in a glorified blog that begins with the words "market gossip ." Let's face the truth and be realistic. The supply of AI semiconductors is now so limited that even second-tier chips, several generations old, retain their value and can be sold.
Specialized chips manufactured by in-house AI laboratories together with Broadcom will definitely not be thrown aside in such conditions. The reality for Google and Broadcom is that the market for specialized chips has expanded so much in the past few years that Broadcom doesn't need to rely on Google as much as it used to.
And at the same time, Google has such a high demand that they too cannot simply rely on Broadcom alone to supply what Google needs. Both companies, Google and Broadcom, are doing the smart thing— diversifying and minimizing the risks of their businesses. If SemiAnalysis wrote something like this, it wouldn't get any clicks.
And if your business is selling newsletters, which is exactly their business, then clicks matter much more than real facts that bring you, the investor, money. I understand that we live in a world of headlines, where people see a tweet or read a passage and think they know the whole truth or that they know the situation.
But long-term investing hasn't changed with the advent of social media. Buying stocks with sustainable competitive advantages, growing demand, diverse customers, and excellent management is a great way to grow your capital. The way to lose everything is to trade stocks based on Twitter headlines and posts without digging deeper and doing what is called due diligence.
So I hope this video helped bring some clarity. And I'm sure when Broadcom reports earnings in a few weeks, CEO Hock Tan will get in touch. He will be asked about this , he will laugh a little, as he always does regarding market rumors surrounding the company, joke and say: "No, we have plenty of things to do." The bottom line is that Broadcom's business is booming, and I remain confident that it's one of the best semiconductor stocks to own during all this AI development. That was the video for today. See you soon . Good luck with your investments.
댓글 0