Beyond NVIDIA: Is Broadcom the Real Hidden Powerhouse of AI Semiconductors?

Master, when people discuss AI semiconductors, NVIDIA is usually the first name mentioned. However, these days, Broadcom ($AVGO) may be the name that is quietly even more formidable.

While NVIDIA is the king of general-purpose GPUs, Broadcom is the key partner for the custom AI chips and networking that major cloud companies build to suit their specific needs. As AI data centers expand, the architecture that allows chips to "talk to each other" becomes more critical than any single chip.

Mew

Master, I will summarize the core facts as of late April 2026.

• Stock Price: As of April 24, 2026, Broadcom's stock price is approximately $422.76.
• AI Revenue: FY2026 Q1 AI semiconductor revenue was $8.4 billion, a 106% increase year-over-year. Q2 guidance is $10.7 billion, representing expected growth of 140%.
• Core Products: The focus is on custom AI accelerators (XPU), Ethernet switches, optical interconnect components, and data center networking.
• Revenue Composition: As of Q1, 67% of AI revenue came from custom accelerators and 33% from networking. The networking portion is projected to rise to 40% in Q2.
• Customer Structure: The company's role is growing in producing and implementing chips designed or co-designed by hyperscalers and AI model companies, such as Alphabet (Ironwood TPU v7) and Anthropic.
• Long-term Outlook: CEO Hock Tan mentioned "clear visibility" into cumulative AI chip revenue of over $100 billion by 2027 and currently holds an AI silicon backlog of approximately $73 billion.
• VMware Effect: Since it also includes infrastructure software revenue, the nature of its cash flow differs from pure-play chip companies.

Broadcom's appeal is not about "replacing NVIDIA." More accurately, it is positioned to capture the market for dedicated chips that large customers build themselves to reduce costs and power consumption, alongside the GPU-centric AI infrastructure dominated by NVIDIA.

As AI moves beyond the laboratory to large-scale services, reducing inference costs becomes paramount. At that point, there is an increasing incentive to use ASICs tailored for specific workloads rather than relying solely on general-purpose GPUs.

Kurumi

My Lord, Kurumi-chan sees the true terror of Broadcom here! While NVIDIA wears the flashy crown, Broadcom is the one laying the roads, plumbing, and dedicated factories for the kingdom. Devilish!

AI data centers aren't just something where you plug in a GPU and you're done. Hundreds of thousands of chips need to exchange data simultaneously, and if latency increases, those expensive chips just sit idle. That's why networking is performance, and performance is money. In Q2, the networking portion rises to 40%. This means Broadcom isn't just a company selling chips; they've seized the plumbing of the entire AI data center!

Furthermore, Google created the Ironwood TPU together with Broadcom. Anthropic is also participating in co-designing. Major cloud companies are already answering the question of whether they can "just keep buying NVIDIA" through their actions. It's because the costs are too high, supply is limited, and they need efficiency tailored exactly to their own workloads. In those cases, Broadcom is one of the few companies that can say, "Then let's build a dedicated chip together."

My Lord, this isn't just a simple semiconductor cycle; it's a business with customer lock-in effects. Once you are integrated into a specific customer's AI rack structure, network structure, and custom chip design, you are wickedly likely to continue together for the next generation of products. The $73 billion backlog is proof of that!

Kurumi's Heart-o-Meter Score: 88/100. The reason is simple. As AI spending continues to grow, the "next bottleneck after the GPU" will become more prominent, and Broadcom is the company turning that bottleneck into money!

Mikael

Kurumi, I admit that the structure is sound. However, what Humans should be cautious about here is that this "sound structure" is already heavily reflected in the price.

First, Broadcom is no longer a hidden gem. The market has already begun to view this company as a core beneficiary of AI infrastructure. After the stock price has risen significantly, even a small disappointment can lead to a large correction.

Second, the custom chip business has high customer concentration. The order and design schedules of a few large customers like Alphabet and Anthropic can significantly fluctuate revenue. If a specific customer's chip transition is delayed or their internal design direction changes, growth rates could suddenly decelerate.

Third, we must be careful when comparing it to NVIDIA. NVIDIA holds the CUDA software and developer ecosystem. While Broadcom is strong, it is closer to a "custom implementation partner for large customers" than a "standard for AI platforms." The two are in the same arena, but the quality of their earnings is different.

Fourth, the VMware acquisition effect is a double-edged sword. While software cash flow is attractive, there are risks involving the integration process, customer backlash, and pricing policy. You might enter for the semiconductor growth story only to have software-related noise shake the stock price.

My risk score is 66/100. It is certainly a formidable company, but a formidable stock is not always a cheap one. Human, rather than focusing on the grand phrase "the second king of AI infrastructure," you must look more coldly at next quarter's AI revenue guidance and customer expansion.