Facing a counterattack after “Fright Crash”? UBS backs Broadcom (AVGO.US): AI revenue for the 2026 fiscal year may exceed 60 billion dollars. The pullback is an overreaction

The Zhitong Finance App learned that last week, Broadcom (AVGO.US) and Oracle (ORCL.US) experienced a shock in stock prices for one day after releasing financial reports. Broadcom plummeted one after another, losing 17% of blood within two days. Last Friday, UBS and the Broadcom (AVGO.US) management team held an offline investor meeting. Based on the content of the conference communication, the bank believes that the market reaction on Friday was an overreaction. Based on management's statement on the AI semiconductor business revenue for the 2026 fiscal year, the bank raised its performance expectations. It is expected that AI semiconductor revenue for the 2026 fiscal year will exceed 60 billion US dollars, an increase of nearly 3 times over the previous year, and the target price will be raised from 472 US dollars to 475 US dollars.

In view of the sharp correction in the company's stock price on the same day and the concentration of investors' questions after the announcement of this quarter's results, the bank believes that the stock has become a generally optimistic bully in the market.

Key highlights of the conference were as follows:

1. In the earnings conference call, the company revealed that the AI business order backlog reached 73 billion US dollars, covering a period of 18 months; however, management clearly stated during the conference that the data is extremely conservative, and the actual delivery period will be close to 12 months.

2. In the fourth fiscal quarter (ending October), the company's total order backlog increased 50% month-on-month, with the AI semiconductor business order backlog doubling; if Anthropic's new $11 billion order is excluded, the remaining order backlog will still increase by 20-25 billion US dollars, mainly from Google and Meta, and OpenAI orders are not yet included.

3. Broadcom believes that due to the progress of preparations for power components, the $21 billion rack order delivered to Anthropic may be extended to the 2027 fiscal year; the company emphasized that Anthropic is the only rack customer in the order backlog, and OpenAI (once included in the order backlog) will become a traditional ASIC customer.

4. The company did not specify the gross profit margin of the $21 billion rack order, but it did not refute market expectations of 45%-50%; at the same time, it reiterated that due to this, gross margin for fiscal year 2026 will decline year on year, but the operating profit margin (OPM) will be at least the same as the previous year (the bank predicts that operating profit margin for fiscal year 2026 will drop 30 basis points from fiscal year 2025).

Furthermore, the company reiterated that the gross margin of the XPU business is about 55% and the gross margin of the AI network business is about 80% — the comprehensive gross margin of Broadcom's own components in the AI rack is about 60%, but due to the inclusion of resale components, the overall gross margin of this $21 billion revenue will fall to the 45%-50% range. The bank believes that Anthropic's rack-scale order is a one-time cooperation, and in the long run, the relationship between the two sides will shift to a standard ASIC/XPU supply model.

5. A year ago, Broadcom estimated the target market size (TAM) of the XPU+ network business in 2027 (mainly based on training load); now the company says the data has expired and the actual scale will expand significantly — on the one hand, the XPU business benefits from the accelerated growth of AI inference demand, and on the other hand, due to the complexity and speed of innovation of the AI technology stack, AI Big Model Lab is seizing the enterprise market originally anticipated by Broadcom (originally planned to use GPUs). More and more companies are choosing to use OpenAI, Anthropic and other manufacturers Services rather than fine-tuning big models yourself.

Ultimately, Broadcom management is confident that the AI business revenue for the 2026 fiscal year will exceed current market expectations and will be higher than any predictions it sees from buyers or sellers. Overall, the bank raised its revenue forecast for the 2027 fiscal year to US$135 billion (2% higher than market consensus), raised its earnings per share (EPS) forecast to US$14.15 (1.7% higher than the market consensus), and raised its target price from US$472 to US$475.

In addition to this, UBS also mentioned in the research report that Anthropic's $21 billion order may be shipped at the end of fiscal year 2026 and continue until the first half of fiscal year 2027. Anthropic wants the order to be delivered in the 2026 fiscal year, but Broadcom believes it will probably continue until the 2027 fiscal year. The bank currently models Anthropic's shipment volume as $15 billion in fiscal year 2026, with the remaining $6 billion delivered in fiscal year 2027.

Broadcom reiterated that the growth rate of AI business revenue in FY2026 will surpass that of FY2025.

The overall gross margin of Broadcom's AI business is close to 60% - the gross margin of the XPU business is over 55%, and the gross margin of the AI network business is over 80%. Broadcom expects the comprehensive gross margin of AI revenue to be above 60%, but due to the inclusion of resale components, the gross shipping margin for rack products will be even lower. In the end, gross margin for fiscal year 2026 will depend on the structure of new AI orders, but Broadcom expects gross margin for fiscal year 2026 to decline year on year, while operating profit margin is likely to remain the same year on year.

Broadcom believes that the growth rate of the XPU business will not exceed the AI network business; at the same time, it emphasizes that the AI network business is currently strong. This is driven by large-scale expansion driven by AI expansion demand and early Ethernet capacity expansion requirements. According to Broadcom, the AI switch business revenue for the 2025 fiscal year was 3 billion US dollars, but order data shows that it may reach 12 billion US dollars in the 2026 fiscal year.

In addition to switches, Broadcom expects DSP business revenue to grow from $1 billion in fiscal 2025 to $5 billion in fiscal 2026. The bank is currently adjusting the year-on-year growth rate of the AI network business for the 2026 fiscal year from 55% previously estimated to 103%. Notably, the gross margin of Broadcom's AI network business reached 80%.

Broadcom's enterprise order backlog increased from US$110 billion to US$162 billion (an increase of more than 50%), with the backlog of AI-related orders doubling month-on-month.

Broadcom responded to concerns about “Google's direct connection with foundries,” saying that this is unlikely to happen within the next 5 years; however, Broadcom's long-term strategy is to expand diversified customers to diversify the risk of customer concentration.

Broadcom pointed out that there is currently a trend of “AI laboratories seizing more markets” — these markets were originally the enterprise-level AI markets expected by Broadcom (which relies on GPUs), but now they are gradually shifting to AI laboratories. The increase in demand for AI laboratories is due to the complicated process of putting AI technology stacks into production; therefore, more and more enterprise customers choose to use the services of vendors such as OpenAI and Anthropic rather than fine-tune large language models themselves.

Broadcom is open to monetization forms for custom chips, and will not rule out introducing a model similar to royalties if customers need it. Under this model, in the average chip sales price (ASP) confirmed by the company, the portion related to Broadcom IP will decrease, but the overall gross margin/operating profit margin will be higher. At present, Broadcom has achieved control over the complete XPU solution in the HBM business, but will adjust it flexibly according to customer needs.