Why Its Custom Silicon Moat Still Supports a Buy Rating
There is one important point investors need to understand about Broadcom.
At its core, the company is effectively two businesses combined into one platform. It is both a leading semiconductor designer and a major enterprise software company.
- Updated
- INSIDER CONVICTION TRADES FY26
The market hasn't noticed yet.
Disclosed by law. Missed by most investors. 129 trades tracked by us.
NO Credit card
- Director buys & sells tracked
- Buy / Hold / Sell verdict on every stock
- Technical analysis included
With Broadcom shares pulling back to around US$395, the recent weakness is starting to look like the beginning of a potential buying opportunity.
We do not want to simply jump on the AI bandwagon. The sector has attracted plenty of hype and euphoria. However, Broadcom is fundamentally different from many other semiconductor names, including Qualcomm.
The company has one of the strongest economic moats in the industry, supported by deep customer relationships, high switching costs and a highly specialised intellectual property portfolio. Broadcom continues to invest heavily in innovation, with around US$11 billion directed toward research and development and more than 18,000 patents across the business.
Even with the share price still trading close to its all-time highs, we maintain a Buy rating on Broadcom.
That said, investors still need to consider the risks. A broader market correction, rising geopolitical tensions or a slowdown in capital expenditure from the Magnificent Seven could all place pressure on the share price.
For that reason, the question is not simply whether Broadcom is a high-quality business. It is how investors can build exposure without leaving themselves vulnerable to a sharp pullback.
The semiconductor side
When we break down Broadcom’s revenue mix, the growth engine becomes clear.
The key driver is XPU and AI networking revenue.
This is Broadcom’s core market for custom accelerators, where the company works with hyperscale customers to design chips tailored to their specific workloads. These chips are known as application-specific integrated circuits, or ASICs.
The best-known example is Google’s Tensor Processing Unit, or TPU. which they also had a first-mover advantage with.
Broadcom has played a central role in the development of Google’s custom AI silicon. In our previous research, we explored this relationship in more detail and explained why Broadcom is increasingly becoming a picks-and-shovels play on the shift toward AI inference.
As AI models move from training into large-scale deployment, companies like Google are focused on running workloads at the lowest possible cost.
This is where custom silicon becomes increasingly important.
Rather than relying solely on general-purpose GPUs, hyperscalers can design chips around their own workloads. That can improve performance, reduce energy consumption and lower the cost of inference.
The commercial incentive is significant.
For Google, investing in custom chips can support better margins as AI usage scales. For Broadcom, it creates a powerful growth opportunity as more hyperscalers look to develop specialised silicon tailored to their own infrastructure.
The networking and connectivity segment
There is another major growth segment emerging that many investors may still be overlooking.
Connectivity.
As data centre operators build increasingly large AI clusters, it is not enough to simply add more chips. Those chips also need to communicate with each other at extremely high speeds.
The goal is to connect thousands of accelerators so they can operate almost like a single computing system. Faster data transfer reduces bottlenecks, improves compute efficiency and allows companies to extract more performance from their AI infrastructure.
This is where Broadcom has built another powerful competitive advantage.
The company commands around 80% of the Ethernet switching market, giving it a critical position in the infrastructure required to connect AI clusters.
Competition is beginning to increase. Marvell has attracted attention with its new Teralynx 10 switch, which delivers 102.4 terabits per second of throughput.
However, Broadcom once again appears to be a step ahead.
Its latest Tomahawk 6 switch delivers 204.8 terabits per second, doubling the throughput of the previous generation.
The easiest way to think about this is that the same principle driving improvements in semiconductors also applies to networking equipment. While Moore’s Law has historically focused on fitting more transistors onto smaller chips, AI networking products are also becoming significantly more powerful with each generation.
The key difference is timing.
While Marvell’s 102.4 terabit-per-second product is still being tested, Broadcom began shipping its next-generation technology in 2025. This suggests Broadcom remains around a year ahead of competitors in a market that is becoming increasingly important as AI infrastructure continues to scale.
Does Broadcom have a sustainable competitive advantage?
Broadcom’s strongest moat sits within the XPU segment of the business.
When a hyperscaler like Google develops a custom accelerator, the switching costs become extremely high. Designing an XPU is not a simple procurement decision. It is a deeply collaborative process that can take two to three years, with Broadcom working closely alongside the customer to optimise the chip architecture for a specific workload.
Once a program reaches production, the relationship becomes difficult to displace.
The chip is embedded into the customer’s infrastructure, the design process has already required years of investment, and the platform typically scales over multiple product generations. This creates a highly sticky, multi-year revenue stream.
Broadcom already commands around 70% of the custom accelerator market, giving it a significant head start as more hyperscalers invest in purpose-built AI silicon.
Although an important nuance is that customers are looking at Marvell to reduce reliance on Broadcom, not in the sense of removing Broadcom, but to keep the company honest, so to speak.
The second moat is Broadcom’s intellectual property portfolio.
As we have discussed, one of the biggest challenges in the data centre buildout is not simply adding more compute. It is moving data efficiently and connecting increasingly large clusters of accelerators without creating bottlenecks.
Broadcom has remained at the forefront of this transition through its networking products, switching technology and continued investment in innovation.
This is why the long-term compounding thesis remains intact.
Broadcom’s valuation and why a buy rating
Broadcom’s five-year average EV/EBITDA multiple sits at around 28x, although the stock reached a peak multiple of roughly 56x in November 2025.
The market is clearly pricing in a substantial amount of growth.
Wall Street estimates currently point to FY26 EBITDA of around US$72 billion, representing an increase of approximately 72% from 2025 levels. Growth is then expected to remain strong, with EBITDA forecast to rise by a further 64% in 2027 and 29% in 2028.
To assess the potential upside and downside, we have applied a range of terminal-year EV/EBITDA multiples.
Our base case assumes a 21x multiple, while the bull case applies 25x and the bear case uses 16x.
The result is a relatively balanced valuation picture.
There is still meaningful upside if Broadcom continues to execute across custom silicon, AI networking and enterprise software. However, the stock is already trading at a premium valuation, which means investors also need to account for the downside risk if AI infrastructure spending slows, hyperscaler capital expenditure moderates or market sentiment turns.
This is why we continue to view Broadcom as a Buy rather than a Strong Buy.
The long-term growth story remains compelling, but the valuation leaves less room for execution risk than it did earlier in the cycle.
