Broadcom: ASIC demand to increase in 2025 (FY24Q3 earnings call)

Broadcom (AVGO.O) released its 2024 third-quarter financial report after the U.S. stock market on September 6th (as of August 2024):

The following is a summary of the Broadcom 2024 third-quarter financial report conference call. For an interpretation of the financial report, please refer to " Broadcom "Surge"? AI Can't Support the Collapse of Traditional Semiconductors 》

I. $Broadcom(AVGO.US) Financial Report Highlights:

II. Detailed content of Broadcom's financial report conference call

2.1. Key points from the management's statements:

1. Business Highlights:

① Software Business:

• Infrastructure software revenue in the third quarter was $5.8 billion, a year-on-year increase of 200%, with VMware contributing $3.8 billion.
• VMware's business model transformation is progressing smoothly, with ABV reaching $2.5 billion, a 32% increase quarter-on-quarter.
• Significant cost control, with VMware expenses decreasing from $1.6 billion in Q2 to $1.3 billion in Q3, expected to achieve or exceed the adjusted EBITDA target of $8.5 billion in the 2025 fiscal year.

② Semiconductor Business:

• Network revenue in the third quarter was $4 billion, a 43% year-on-year increase, mainly driven by the demand for AI networks and custom AI accelerators in super-scale data centers.
• Revenue from custom AI accelerators increased 3.5 times year-on-year, with significant growth in Ethernet switching and optical interconnect device revenue.
• Non-AI network products increased by 17% quarter-on-quarter in Q3, a 41% year-on-year decrease, with revenue expected to decline by 30% year-on-year in the fourth quarter, but overall network revenue is expected to increase by over 40% year-on-year.

③ Other Semiconductor Businesses:

• Server storage connection revenue increased by 5% quarter-on-quarter, expected to continue growing in Q4.
• Wireless business is expected to grow by 20% quarter-on-quarter in Q4, flat year-on-year.
• Broadband business performance is weak due to telecom company spending pauses, expected to recover in 2025.
• Industrial and other revenue is close to the bottom, with a expected rebound in Q4.

④ Future Outlook

1. Financial Highlights

① Free Cash Flow: Free cash flow is $4.8 billion, accounting for 37% of revenue. Excluding restructuring and integration costs, free cash flow is $5.3 billion, a 14% year-on-year increase.

② Debt: At the end of the third quarter, the company held $10 billion in cash and total debt was $72.3 billion. The company replaced $5 billion of floating rate debt and plans to repay $1.9 billion of fixed rate bonds in the fourth quarter.

③ Capital Allocation:

• The company paid $2.5 billion in cash dividends to shareholders in the third quarter, and the cash dividend per share in the fourth quarter will increase to $0.53.
• It is expected that the Q4 non-GAAP diluted share count will be approximately 4.91 billion shares.

2.2, Analyst Q&A

Q: What is the revenue ratio of Compute and Networking? How do you view this ratio for the expected $3.5 billion in the fourth quarter? Looking ahead to fiscal year 2025, although you did not provide guidance for the overall AI business, do you have confidence in Broadcom's ability to grow in sync with or surpass the AI chip industry?

A: As mentioned in the last earnings conference call, about two-thirds of revenue in the previous quarter came from Compute, and one-third came from Networking. We expect to maintain a similar trend in the fourth quarter. While we have not provided guidance for fiscal year 2025, we anticipate strong growth in AI revenue.

Q: From suppliers and the AI ecosystem, NVIDIA mentioned that their revenue this quarter shifted from cloud service providers to enterprises. Given your primary focus on cloud customers, do you think this shift will slow down your revenue prospects? Do you see this trend, or do you have a different view?

A: We do not focus on the enterprise AI market. Our AI products are mainly focused on AI accelerators, XPUs, and networking, especially for large-scale data centers, cloud computing, and large platforms. We have little exposure to enterprise AI, so we have not seen this trend.

Q: VMware's business performed well in the third fiscal quarter, but traditional Broadcom software business seems to have declined. What caused the fluctuations in traditional Broadcom business? Secondly, have we reached the $4 billion benchmark related to VMware in the fourth quarter? What are the main driving factors and challenges for future growth rates?

A: VMware's business performed well. We achieved good results by converting customers from perpetual licenses to full-stack VCF subscription licenses, with a high proportion of new VCF subscription users. We expect this trend to continue in the fourth quarter and even extend into 2025. Therefore, in addition to the guidance for the fourth quarter of 2024, the overall trend in bookings and growth is accelerating.

Q: In terms of semiconductors, non-AI networking business declined by over 50%, and other businesses are also far below peak levels. Besides cyclical factors, are there other reasons why these businesses cannot return to previous levels? Is there hope for recovery when the market rebounds? **

A: We have experienced a typical downturn in the non-AI sector, mainly due to inventory adjustments at various stages of the supply chain. Although the impact could not be completely avoided, we have seen clear signs that the lowest point has been passed. In the third quarter, non-AI semiconductor orders grew by 20%, indicating that recovery is underway. The pace of recovery varies in different markets, but we have seen rebounds in enterprise data centers and IT spending. While the broadband market has not yet bottomed out, orders are on the rise, indicating that the bottom is near. Overall, the non-AI semiconductor industry has emerged from the downturn, is recovering, and is expected to return to previous levels.

With the expansion of AI in enterprise and digital native fields, there will be an increase in hardware upgrade demand, which may trigger a new upturn cycle, possibly surpassing previous cycles. Therefore, we believe that 2024 is the lowest point of the recovery.

Q: On the software side, non-VMware business has returned to a quarterly level of around $2 billion. Is this due to the decline in Brocade? Is this $2 billion quarterly level already the bottom, and will it be the growth benchmark for future non-VMware software business?

A: Regarding the software segment, I believe we have reached a stable level, with Brocade showing significant fluctuations, but overall stable. The stability of non-VMware software revenue is evident. Next, our main focus will be on VMware's performance in the next year to year and a half.

Q: Regarding semiconductor and AI revenue. Is the approximately $3.1 billion in the third quarter in line with expectations? Are there any parts that did not meet expectations? Where does the expected increase to $3.5 billion in sequential growth come from? Additionally, you mentioned that AI revenue will increase significantly next year, is this due to the addition of hyperscale data centers and consumer internet customers?

A: Our third-quarter data basically met expectations, and the fourth-quarter expectations support our raising of the full-year AI revenue guidance to over $12 billion. If there are no other changes, this indicates that the trend will continue to be strong next year, mainly coming from hyperscale data centers, cloud computing, and digital native companies, combined with AI accelerators and network business, based on existing backlog orders. As for whether there will be more customers, we will wait and see.

Q: Can you discuss the issue of repatriating IP to the United States, which resulted in a $4.5 billion tax liability. Historically, Broadcom adjusted its place of registration before a transaction, and some investors are asking if this time is related to asset sales for debt repayment. Can you clarify?

A: This is just a timing arrangement for our relocation of intellectual property, unrelated to asset sales. We relocated the intellectual property, resulting in a $4 billion expense, but with deferred tax liabilities, which are basically non-cash and have minimal cash impact.

Q: I would like to ask about the comparison between your AI revenue and GPU growth rates. This year, AI revenue growth is roughly the same as GPU computing. You mentioned that AI revenue will continue to grow next year, but major customers will introduce new custom ASICs next year, and some believe they may partially switch back to purchasing GPUs. Do you think the growth of AI revenue next year will still be roughly in sync with GPU computing growth? A: This is a difficult question involving two parts. Regarding GPU growth, one should inquire about companies focusing on GPUs such as NVIDIA and AMD. I do not engage in the enterprise market, but mainly focus on hyperscale data centers, which are not directly related. In the long term, a few large hyperscale data centers relying on running large language models and AI demands will eventually develop their own custom chips and accelerators, a transition that may take several years to complete. Broadcom focuses on providing accelerators for these customers rather than enterprise AI workloads.

While some companies are entering the AI market for hyperscale data centers, this is a transitional phase and the two are not directly related. As the transition occurs, our business model of providing accelerators and networks for hyperscale data centers will benefit.

Q: I would like to understand VMware's profit situation. Historically, the operating profit margin of the software business has exceeded 70%. I want to know if, after all adjustments, VMware's profit situation will be similar to other software businesses?

A: I mentioned in the third quarter that VMware had revenue of $3.8 billion and expenses of $1.3 billion, which can give an indication of our operating profit margin and EBITDA profit margin trends. We expect revenue to continue to grow in the fourth quarter, while expenses are expected to stabilize and continue to decrease.

Q: Good afternoon, thank you for answering the questions. I am interested in the software gross margin. After acquiring VMware, our gross margin dropped from the low 90s to the high 80s, and is now recovering. As we approach the $4 billion target and look ahead to the 2025 fiscal year, how should we view the trend of the software gross margin?

A: The software gross margin is not crucial for us unless it is large-scale SaaS operations. Although some products are subscription-based, most are not SaaS, so the gross margin will remain around 90%.

Q: Can you discuss custom AI revenue, especially the contributions from customers other than the largest customer? How significant are the contributions from other customers? With new projects launching, what are your expectations for next year?

A: Currently, we have three important customers who have deployed production-grade accelerators in AI data centers. For any customer, this is not an easy product to deploy, and we do not consider concept validation as mass production customers; there must be actual shipments to consider them as customers.

Q: This year, the acquisition of Seagate's hard drive SOC assets was completed. Could you talk about the specifics of this acquisition, its impact on the company's finances, and whether it will accelerate the development of the storage business in the coming years?

A: This acquisition is more like a collaboration. We see hard drives as a long-term storage medium for hyperscale data centers, with significant room for hard drive technology development, from the current 22-24TB to future 30-50TB. This collaboration aims to drive hard drive technology to reach 50TB within five years, and we believe that hard drive storage will remain important in the next five years and even longer.

Q: Looking ahead to NVIDIA's Blackwell product cycle, there are indications that Broadcom may have more opportunities in the optical field. Do you think this presents a deeper involvement opportunity compared to previous products, or is it a general participation, especially in DSP and other related fields? **

A: We are not directly involved in NVIDIA's Blackwell roadmap or market, but the foundational technologies we develop can be applied, such as optical components or DSP interconnect support. We are willing to participate in this ecosystem, but we are not directly in this market.

Q: What is your view on future mergers and acquisitions? Are acquisitions still part of the plan? If so, will you continue to focus on software, or is the semiconductor sector also of interest to you?

A: Currently, we are fully focused on transforming VMware's business model, which is progressing beyond expectations. I need another one to two years to complete this transformation thoroughly, so there are currently no other acquisition plans.

Q: You mentioned that AI customers accelerated the development and deployment of XPU ASICs, especially in the second half of the year. Considering packaging and HBM memory constraints, the supply of XPUs and network devices is very tight. Have you seen additional demand in the second half of the year, can you meet these demands, or are you constrained by supply? Has the demand for AI exceeded your supply capabilities?

A: We continue to see orders and additional demand, especially as large-scale data centers accelerate the deployment of AI infrastructure. I anticipate more of this demand by 2025, although it is not written in the guidance, the deployment of XPUs and infrastructure acceleration will continue to grow, and we may see more of this in the next 12 months. We are able to meet this additional demand.

Q: You previously mentioned that AI computing may shift from ASICs to the commercial market, but the current trend seems to be the opposite. Do you still believe this is a long-term trend? Additionally, considering ASICs have better performance per watt than GPUs, why do we not see more companies moving towards custom ASICs? Especially large enterprises like AWS, would they customize chips for specific workloads? Overall, what is the trend of ASICs in the AI field?

A: I used to believe that general-purpose chips would prevail, but my view has changed. The AI accelerator market is divided into two categories: the enterprise market is unable to develop custom chips, while large-scale data centers have the scale and economic strength to develop their own custom accelerators, especially for training large language models. These companies will gradually shift towards custom ASICs or XPUs, although the process is slow, the direction is clear. Meanwhile, there is still demand in the enterprise market for commercial silicon.

Q: With the increase in costs and power expenditures, will ASICs become more attractive, so will the curve trend differently?

A: This is an accelerating curve, although it may take longer than expected, it is definitely accelerating. The demand scale of large-scale data centers will rival that of the enterprise market.

Risk Disclosure and Statement of this article: Dolphin Research Disclaimer and General Disclosure