Broadcom’s Next Catalyst Has Wall Street Asking an Awkward Question: Are We Late?

Long after the closing bell rang on a Tuesday afternoon in lower Manhattan, the screens on the trading floor continued to glow. The ticker for Broadcom, AVGO, continued to show up in investor chats, analyst notes, and the quiet side discussions that typically take place when something significant might be changing. It wasn’t precisely an optimistic or pessimistic question. It was easier, but maybe a little awkward: are investors already running late?

Since Broadcom has been in business for a considerable amount of time, it doesn’t always exude the aura of a technological marvel. Its enterprise software, networking chip, and custom semiconductor design business has historically sounded more industrial than glamorous. But recently something unusual has been happening. The company’s artificial intelligence chip business is growing rapidly, and it’s becoming more difficult to ignore the numbers.

According to the most recent data, AI semiconductor revenue more than doubled from the previous year, reaching approximately $8.4 billion in a single quarter. Wall Street is usually awakened by such growth. The more intriguing part, however, came during a recent earnings call when CEO Hock Tan calmly stated a goal that sounded almost casual: over $100 billion in revenue from AI chips by 2027.

This is the kind of statement that causes analysts to hesitate. Nvidia has dominated discussions about AI infrastructure for months. And it makes sense. Modern AI models, from OpenAI’s systems to the massive computing clusters being constructed by firms like Meta and Amazon, are powered by Nvidia’s graphics processors. Many investors stopped questioning whether other chipmakers could profit from the company’s total dominance.

However, it appears that Broadcom is creating its own path. The company creates custom AI accelerators—chips made for particular clients—instead of selling standardized GPUs. These chips, which are frequently referred to as XPUs, are designed to precisely match the workloads of hyperscale data centers. Imagine a toolmaker creating equipment specifically for a single contractor’s job instead of selling a general-purpose tool to everyone. It’s a more subdued approach, but it might be more resilient.

That idea seems to appeal to some of the biggest tech companies. For a long time, Google has relied on specialized processors called TPUs, many of which Broadcom designs or produces. As AI models become more sophisticated, analysts now predict that millions of these chips may be shipped in the upcoming years. Additionally, there are rumors—sometimes verified, sometimes conjectural—that businesses like OpenAI are experimenting with partnerships for custom hardware in an effort to manage the exorbitant expenses associated with training AI systems.

The magnitude of that demand is easier to comprehend when one stands outside a massive data center facility in Northern Virginia, where rows of cooling fans hum like far-off turbines. These windowless, warehouse-sized structures use enormous amounts of silicon and electricity. Infrastructure worth billions of dollars is represented by each one.

Additionally, Broadcom seems to be involved in the supply chain more and more. Still, the stock market’s reaction has been oddly cautious. Over the last few years, Broadcom shares have increased significantly, rising more than 600% in just five years. However, due in part to investor concerns about margins, the stock has periodically fluctuated or stalled in recent months.

Although profitable, custom chips frequently have marginally lower profit margins than Broadcom’s conventional semiconductor products. That’s the trade-off. Revenue grows and volume increases, but the financial math gets a little trickier.

How investors will ultimately weigh that balance is still unknown. The long-term estimates for spending on AI infrastructure have been quietly increased by analysts at a number of significant banks. According to some estimates, the cost of AI-related data-center systems worldwide may reach $1.4 trillion by the end of the decade. Although that amount seems almost unreal, new capital expenditures are announced by another hyperscaler every few months.

Watching this unfold, there’s a feeling that the semiconductor landscape is shifting in subtle ways. Nvidia continues to make headlines. However, Broadcom appears to be developing a business behind the headlines—less ostentatious, more methodical.

That brings us back to the awkward query. Investors are frequently reluctant to push a stock higher when it has been rising steadily for years. After a significant rally, nobody enjoys purchasing. Ironically, though, a lot of long-term winners appear pricey at almost every stage of their development. Apple did. Amazon most definitely did. Only a few years ago, even Nvidia did.

It’s possible that Broadcom is going through the same psychological stage. The company’s leadership seems exceptionally self-assured for the time being. Hock Tan hardly ever uses ambiguous language. He described specific production goals, supply-chain commitments, and customer relationships in recent calls with analysts. In a field that frequently favors cautious language, that degree of clarity is uncommon.

It appears that investors value that. However, they are also making calculations. Because whether Broadcom can expand isn’t the main concern on Wall Street. It can, according to the numbers. Whether the market is still underestimating the potential size of this infrastructure buildout or if the next phase of the AI boom has already been priced into the stock is the real uncertainty.

It’s difficult to ignore how frequently the same discussion is currently taking place in investment circles. Silently, occasionally in between earnings calls or over coffee.

Broadcom may already be one of the biggest beneficiaries of the AI era. The unanswered question is whether investors were aware of it in a timely manner.