At first glance, there is no operational deterioration in Broadcom's third-quarter results that would explain the decline. The company's revenue increased 86% year over year to $29.6 billion, and adjusted earnings per share increased 96% to $3.32. Free cash flow hit a record $13.7 billion, while AI semiconductor revenues grew 221% to $16.7 billion. AI now accounts for 56% of Broadcom's total revenue.

But there's a key shift underlying the results that the market may question: Broadcom's gross margin is shrinking as AI grows. Gross margin fell from 77.1% in Q2 to 75% in Q3, and the company expects it to decline to approximately 73% in Q4.

The management clearly shows the reason for this as the increasing weight of XPUs in revenue and the use of more memory in new generation chips. Despite this, operating margin reached 67.9% in Q3, thanks to economies of scale; Q4 expectation is approximately 66%. So the company generates much more AI revenue, but the margin profile of that revenue is lower than the software business.

The most striking part is the future expectations. While Broadcom expects AI revenue of approximately $58 billion in 2026, it expects to increase this to $115 billion in 2027 and $230 billion in 2028. While there is a long-term TPU agreement with Google, Anthropic will reach 5 GW capacity in 2027 and an additional 10 GW capacity in 2028; OpenAI is planned to exceed 1.3 GW in 2027 and 5 GW in 2028.

But Broadcom management's message is important: demand alone is no longer the driver. Hock Tan says customers are asking for more chips than the company anticipated, but it's unclear whether these chips can be installed in data centers in time. Land, electricity, data center buildings, advanced manufacturing capacity, substrates, and HBM memory can each become bottlenecks at different times. The company states that the demand for 2027 is over $115 billion, but it limits this figure according to current supply and installable data center capacity.

Therefore, it is more accurate to interpret the decline in the stock as an increase in the level of expectations rather than a weak balance sheet. Broadcom's AI story gets stronger; But from now on, the market will price not only revenue growth, but also how much margins will be maintained and how much of the hundreds of billions of dollars of demand can be physically converted into data centers.

$AVGO | There was an interesting point in Broadcom's earnings report. CEO Hock Tan stated that customers are demanding more chips than the company had anticipated, but it remains uncertain whether these chips can be installed in data centers on schedule. He noted that land, electricity, data center facilities, advanced manufacturing capacity, substrates, and HBM memory could each become a bottleneck at different stages. So, what does this mean? Essentially, it presents a challenge for management. To me, it implies: "I can manufacture and sell the best chips, but I can't quite predict how my customers will fare when it comes to actually bringing that capacity online."

As you can see, high demand alone isn't enough. The equipment needs to be installed, and the companies driving that demand need to be able to generate revenue from the gear they purchase. There are significant obstacles to this in the U.S. The existing infrastructure is insufficient to support the data centers currently under construction or the capacity that companies aim to build.

Power grids are aging, water shortages exist, and companies are contending with various local regulations. These are major factors slowing down investment. That said, Broadcom projects its revenue will double by 2028. In my view, the primary issue is no longer the chip shortage itself, but rather the infrastructure companies responsible for deploying that capacity and the memory supply chain.

The Trump administration has signed an emergency executive order to accelerate the strengthening of energy infrastructure. This creates new opportunities in these sectors as well. I won't be adding to my Broadcom position today. The stock is trading below the 200-day SMA on the daily chart, suggesting that downward pressure will persist for a while. I’ll be monitoring the weekly charts, and you might want to do the same. I’ve set alerts for two specific buy points; I’ll step in as a buyer when those levels are reached. Broadcom holds a significant slice of the AI ​​market, and I want to get my own share of that slice. I will do my part before the elections.