Broadcom held its 2025 fiscal third quarter earnings conference call, during which the company presented its financial performance, including non-GAAP financial results, and provided financial forecasts for the fourth quarter along with commentary on the business environment. Investors can access detailed information on the company's website. The conference also mentioned risk factors that could impact actual results and highlighted the comparison between non-GAAP and GAAP indicators in the financial report.

In the third quarter of the 2025 fiscal year, the company's total revenue reached a record $16 billion, a 22% increase year-on-year, mainly driven by strong growth in AI semiconductor and networking technologies. AI semiconductor revenue saw accelerated growth, accounting for 65% of total AI revenue, and is expected to significantly increase in the 2026 fiscal year. In addition, the company introduced a series of innovative networking solutions, including OpenAI and Open Ethernet technologies, to address the need for large-scale computing, particularly in the field of generative AI. Non-AI semiconductor business is expected to recover in the fourth quarter.

The infrastructure software department showed strong performance in Q3, with revenue reaching 6.8 billion US dollars, exceeding expectations, and adding new contract value of over 1 billion US dollars. Through the acquisition of VMware, the integrated cloud platform VMware Cloud Foundation vEd was successfully launched, supporting on-premise or cloud deployment for enterprises, compatible with virtual machines and container applications, providing a strong competitive solution for public cloud. It is expected that Q4 infrastructure software revenue will reach 10 billion US dollars, a year-on-year growth of 10%, driving full year revenue to 15 billion US dollars, with adjusted EBITDA of 5%. Overall performance set a record, with revenue of 16 billion US dollars, a year-on-year growth of 22%, gross margin of 78.4%, operating profit of 10.5 billion US dollars, an increase of 32% compared to the same period last year.

The financial report shows that the company's semiconductor solutions revenue reached $9.2 billion in the third quarter, with accelerated year-on-year growth, mainly due to the growth in the AI field, and the proportion of total revenue increased. Infrastructure software revenue was $6.8 billion, up 17% year-on-year, accounting for 40% of revenue. Free cash flow in the third quarter reached $7 billion, accounting for 44% of revenue. The company expects revenue of $17.4 billion in the fourth quarter, with semiconductor revenue expected to reach $10.4 billion, up 30% year-on-year, and infrastructure software revenue expected to reach $7 billion, up 15% year-on-year.

Discussed the reasons for the accelerated growth of AI business, especially XPU products, pointing out that the addition of a major client and increased demand from existing clients are key factors that have collectively driven significant growth in the business.

The conversation discussed signs of recovery in non-AI businesses after experiencing a low point, including positive performances in areas such as broadband services, storage, and enterprise network upgrades. While overall growth is not yet significant, seasonal improvements have been observed in the wireless and server storage sectors, especially with broadband services showing a continual growth trend. Looking ahead, the continued recovery of non-AI businesses is expected to extend into the next fiscal year, but the specific extent of recovery still needs to closely monitor market dynamics.

Discussed the slow recovery situation in the non-AI semiconductor industry, pointing out that it may show a U-shaped recovery trend, while the broadband business is experiencing a strong rebound after a sharp decline. Order trends show year-on-year growth in the non-AI semiconductor industry, but at a slower pace than AI-related products, with significant signs of recovery expected in the later stages.

Discussed the growth forecast of the AI guidance policy for the fiscal year 2026, pointing out that the growth rate of Ed will accelerate, far exceeding the growth rate of networks and customized businesses, and the proportion of networks and customized businesses is expected to change, with the proportion of network businesses gradually decreasing.

The discussion focused on the company's backlog of $110 billion worth of orders, emphasizing the significant growth in AI-related business, with a large proportion in the semiconductor sector. AI-related orders far outweigh non-AI orders. Software business showed stable growth, with non-AI business also achieving double-digit growth, but still lagging behind AI.

Discussed the performance of AI revenue exceeding expectations and the progress of collaboration with non-script clients. Mentioned that currently four clients have entered the production stage, but remain cautious about specific plans for the future, without providing exact production targets or timelines, and emphasized close collaboration with potential clients to gradually advance their respective project strategies.

Discussions were held on market segmentation targeting cutting-edge model developers, as well as the selection criteria for potential customer groups, emphasizing the focus on a few players who can invest heavily in computational power, while also expressing a cautious attitude towards acquiring new customers beyond the ones already identified.

The conversation revolves around the application of Jericho technology in large-scale GPU clusters, discussing its importance in AI training and inference, as well as how to achieve efficient collaboration between clusters through cross-data center network connections. The maturity and stability of the technology are emphasized, indicating that it has been validated in multiple extremely large-scale users.

The progress of migrating the top 10,000 customers from vSphere to vSphere Cloud Foundation (VCF) was discussed, with over 90% of customers reported to have purchased VCF licenses, but comprehensive deployment still requires effort. The value creation of VCF in private cloud deployments and future potential for advanced service sales were emphasized, while the impact of hardware virtualization on reducing enterprise costs and the adoption of VCF by small and medium-sized enterprises is still being observed.

Discussed the adjustment of revenue structure, especially the impact of decrease in software revenue and increase in wireless and xPU contributions on gross margin, analyzed the trends in semiconductor and software gross margin changes, as well as forecasts for future revenue and gross margin.

The conversation revolves around the classification of the fourth largest customer, making it difficult to clearly distinguish whether they are a super large-scale or traditional llum manufacturer. Mention is made of a billion-dollar order expected to be deployed by the end of the third quarter of the fiscal year, with the specific timing potentially during the Q3 period.

The expansion potential of Ethernet in AI network business was discussed, emphasizing its advantages in decoupling from GPUs, especially in ultra-large scale systems. It was pointed out that the company optimizes network switches and other components to support Ethernet as a choice for open network protocols, while maintaining tight integration with XPUs to meet customer demand for high-performance networks. It is expected that in the next year, Ethernet will have a larger market in cross-GPU cluster expansion.

Discussed the advantages of Ethernet technology in data centers and AI infrastructure, as well as its competitive position relative to other protocols. Emphasized the maturity and scalability of Ethernet, especially the potential for improvement in reducing latency. Also mentioned the competitive landscape in the ASIC market, pointing out that continuous investment and innovation are key to maintaining a competitive advantage.

The conversation discussed the trend that as the scale of AI data centers expands, the proportion of XPU chips used in existing customers will surpass GPU, becoming the primary computing resource. This is attributed to the continuous optimization of XPU chips in efficiency and performance, as well as the improvement in stability and reliability of the software stack, which has increased customers' confidence and gradually increased the use of XPU in multiple generations of updates. In addition, it also mentioned the industry conferences that the company will participate in and the financial report release plan.