🔗 Broadcom at the Intersection of Custom AI Chips and Data Center Bottlenecks - 2026.09.15

Hi, it’s Dotori. The key issue for Broadcom this time is not so much a comparison with Nvidia as how broadly custom AI chips are actually spreading across the core of the value chain. Analysis has emerged that Broadcom has joined the custom chip businesses of OpenAI and Google, while further reports say that the expansion of business related to Anthropic is even changing how Google’s TPU business is viewed ^{[1] [2]}. Looking at the bigger picture, the competition for AI computing is expanding from a contest over a single accelerator into a structure that combines data-center-specific custom designs with network infrastructure.

Custom Chips: The Structure Matters More Than the Customer Names 🔗

Global Economic reported that Broadcom is influencing the competitive landscape of the fabless industry by participating in the custom chip businesses of OpenAI and Google ^[1]. What matters here is not simply how many customers there are. As hyperscalers design chips to match their own computing needs, the role of companies that can provide both design collaboration and network connectivity may grow.

Of course, this article discusses the direction of business expansion; it does not confirm the size of contracts or when revenue will be recognized. So there is still a gap between the structural opportunity and actual results. It is possible to say that the gears have begun to mesh, but which gear will accelerate first requires separate confirmation.

The Questions Raised by Anthropic and Google’s TPU ⚙️

finance.yahoo.com analyzed that Broadcom’s expansion of business related to Anthropic is changing the way Google’s TPU business is viewed ^[2]. This content reflects the view that Broadcom’s custom chip business may not be tied to just one customer or one type of accelerator.

However, the material itself presents an analysis and a shift in perspective. It would be difficult to interpret it more broadly as meaning that the company has definitively announced a new business plan or that the financial impact of the related business has been confirmed. Dotori believes that the expansion of customers for custom AI chips remains in the verification phase. An increasing number of names and the creation of a repeatable business structure are not the same thing.

Looking at the Industry as a Whole, Bottlenecks Exist Beyond Chips 🏭

The 815 Money Talk video was filmed on September 14, 2026, and covered arguments for slowing AI momentum, AI data-center power demand, and infrastructure bottlenecks in optical communications, substrates, and power ^[4]. Viewing Broadcom only as a chip design company narrows the perspective. As data centers grow, surrounding infrastructure such as switches, connectivity networks, and optical communications must move together.

If we must divide the value chain by relative beneficiaries, first priority would be custom AI chip design and network chips, second priority would be optical communications and data-center connectivity infrastructure, and third priority would be fields supporting bottlenecks, such as substrates and power. However, this order is an interpretation based on the industry analysis in the video and does not represent Broadcom’s confirmed allocation of profits. Since the entire network must connect like a relay, one fast segment alone does not immediately determine the final result.

Market Expectations and Short-Term Volatility 📚

24/7 Wall St. introduced the view that Broadcom could be better positioned than Nvidia in the next phase of AI ^[3]. On the other hand, in recent trading-day data, Broadcom’s closing price on September 14 moved by -4.74% from the previous day to $344.82, while trading volume was 20,175,809. This shows that optimistic analysis and actual market movements do not always move in the same direction.

Ultimately, there are three key points. First, whether custom chip collaborations related to OpenAI, Google, and Anthropic lead to actual business expansion; second, how bottlenecks in data-center power and connectivity infrastructure fit with Broadcom’s networking business; and third, whether these expectations are confirmed in earnings. In the next scene, will the variable to watch be the specifics of customer expansion, or the persistence of infrastructure demand beyond chips?

📰 Reference News (3)

1

Broadcom Rides OpenAI and Google’s Custom Chips… Shaking Up the Fabless Landscape - Global Economic

글로벌이코노믹 · 2026.09.12

📰 View original

2

Broadcom’s Anthropic Ramp Puts Google’s TPU Business in a Different Light - finance.yahoo.com

finance.yahoo.com · 2026.09.13

📰 View original

📺 Reference Videos (1)

4

YouTube

[#Global Insights] Don’t Be Fooled by the AI Slowdown Theory… AI Data Centers Triple / Why Money Is Pouring into Optical Communications, Substrates, and Power | Roh Geun-chang, CEO of Semicon Research Lab

▶815머니톡·2026.09.14