Broadcom is sitting on an order book that most chip companies can only dream of, and the market is treating it like an ordinary semiconductor business. That is the argument at the center of a Seeking Alpha analysis published this week, which contends that Broadcom’s roughly $30 billion in backlog, driven by AI networking chips and custom accelerators, is not reflected in the stock.
The backlog is the measure of the shift inside Broadcom. The company’s AI revenue comes from two sources: the networking chips that tie AI servers together, and the custom silicon it designs for the largest cloud providers, including the tensor processing units that power some of the biggest AI training clusters.
Both businesses are booked years in advance. Broadcom’s custom chip contracts are design engagements that run for years, and its networking products are sold to the same hyperscalers that are committing record sums to data centers. The backlog gives the company unusual visibility into its own revenue.
The market’s skepticism has a logic. Broadcom has been classified as a semiconductor company for its entire existence, and semiconductor stocks trade on the memory of brutal cycles: boom, glut, crash. Investors have been burned before by chip companies whose order books evaporated when demand turned.
The counterargument is that Broadcom’s AI business does not behave like a cyclical chip business. Its customers are a handful of the largest companies in the world, spending on AI infrastructure they have described as a multiyear commitment. The backlog is not dealer inventory; it is contracted engineering work.
The company has been building the foundation for years. Broadcom’s acquisition of VMware gave it a software business that smooths out the semiconductor cycle, and its networking division has become the backbone of the AI data center. The combination is unlike anything else in the chip industry.
The earnings picture supports the thesis. Broadcom’s AI-related revenue has grown severalfold over the past two years, and the company has guided to continued growth as custom chip programs move from design into production. The backlog is the reason the guidance carries weight.
The valuation question is where the argument gets interesting. Broadcom trades at a discount to the growth rates its AI business is delivering, and the analysis argues the market is still pricing the company on the old semiconductor playbook. If AI capital spending continues to be revised upward, Broadcom is among the best positioned to benefit.
The risks are the same ones facing every AI supplier. If hyperscalers slow their buildouts, the backlog would shrink and the cycle would return with a vengeance. The concentration of Broadcom’s AI revenue among a few customers is a structural risk that no amount of backlog can eliminate.
Broadcom’s shape is unlike any other chip company. Built through a long series of acquisitions, it sells a portfolio that spans networking chips, wireless components, storage controllers and, since the 2023 acquisition of VMware, enterprise software. The software business, which generates recurring subscription revenue, is the cushion that most semiconductor companies lack.
The custom silicon business is the growth engine. Broadcom designs specialized chips for a handful of large customers, most prominently Google, with which it co-develops the tensor processing units that power some of the largest AI training clusters. These are multiyear engineering engagements, and once a design enters production, the revenue streams for years.
The networking side is equally tied to AI. The largest AI systems connect tens of thousands of chips, and the switches and interconnect silicon that tie them together are dominated by Broadcom. Every new AI cluster, whether built by Microsoft, Alphabet or Amazon, is a potential networking order.
The cycle risk is the counterargument, and it is not trivial. Broadcom’s customers are few, and their decisions to build or delay data centers can swing the company’s results. The backlog provides visibility, but visibility is not immunity, and the history of the semiconductor industry is littered with companies whose backlogs evaporated when the buyers paused.
The market will get fresh evidence soon. Broadcom reports its next quarterly results in the coming weeks, and investors will compare the AI revenue growth against the stock’s valuation. If the growth is there, the analysis says, the stock has room to re-rate; if it slows, the discount to the sector will look justified rather than cheap.
For investors, the analysis frames a choice. One reading says Broadcom is a semiconductor company in disguise, due for the correction that has followed every chip boom. The other says it is an AI infrastructure company wearing a chip company’s valuation, and that the market will eventually notice. The backlog, sitting between those two readings, is the evidence that decides which one wins.
