In the narrative of AI chips, Nvidia has captured almost all the attention. Yet another semiconductor giant with a market cap of $2 trillion – Broadcom – is often overlooked by investors. However, that situation may soon change.
Broadcom’s management has signaled clearly that a major transformation is underway, driven by custom AI chips.
Nvidia’s success is built on general purpose GPUs that handle a wide range of AI workloads. But Broadcom has chosen a different path – working directly with AI hyperscalers to create custom designed chips tailored to their specific workloads. Google’s TPU (Tensor Processing Unit) is the most successful example.
The advantage of this customization is extreme optimization for specific workloads, often delivering the best price/performance ratio for training and inference. The trade-off is reduced flexibility – custom chips cannot handle every task as well as general purpose GPUs. But the reality is that most AI companies now have a clear understanding of their workload patterns, making it a logical choice to engage Broadcom to design custom chips to optimize performance. At the same time, they can still maintain some access to Nvidia chips for non-standardized tasks – the two are not a zero-sum game.
Broadcom CEO Hock Tan revealed in March: “Our custom accelerator business grew 140% year-on-year in Q1. This momentum continues in Q2. The ramp of custom AI accelerators across all our five customers is progressing very well.”
The data confirms this explosive growth. In the latest quarter, Broadcom’s AI semiconductor division (which includes custom AI chips and other products) generated 8.4 billion in revenue, up 1068.4 billion in revenue, up 106 billion or more in annual revenue.
This means that within just over two years, a single business line is set to leap from a quarterly run rate of several billion dollars to a hundred billion dollar annual level. The speed of growth and the certainty of orders are enough to reshape the company’s entire growth trajectory.
Today, Broadcom’s market cap is approximately $1.9 trillion (with a recent modest pullback in share price). Valuation metrics such as its P/E ratio do not fully reflect the explosive potential of its custom AI chip business. The market is still viewing Broadcom through the lens of a “traditional semiconductor giant,” ignoring a key fact: custom AI chips are moving from the “periphery” to the “mainstream.”
Thanks to the successful demonstration of Google’s TPU, more and more AI companies are beginning to deploy custom chips. Once this trend gains momentum, Broadcom will transform from a “follower” into another pole that stands shoulder to shoulder with Nvidia. In specific scenarios, custom chips – with their superior power efficiency and lower total cost of ownership – could even overtake general purpose GPUs.
For investors, the current “valuation lag” in Broadcom’s stock offers a rare window of opportunity. Once the 100 billion revenue target for 2027 begins to materialize, the market will have no choice but to reprice the company′s value. At that point, the 100 billion revenue target for 2027 begins to materialize, the market will have no choice but to reprice the company′s value. At that point, the 2 trillion giant’s status may be just the beginning.
Of course, risks remain: high customer concentration (currently relying mainly on five key customers), uncertainty in technology road map evolution, and potential pushback from Nvidia. But given the current landscape, Broadcom has already established an unassailable first mover advantage in the differentiated segment of custom AI chips.
In the AI computing arms race, Nvidia is the supplier of “general purpose weapons,” while Broadcom offers “customized missiles.” The market size for both is expanding in tandem, and the latter is now stepping out of the shadows into the spotlight. Smart investors should see this trend clearly before the big transformation is complete.
