Focus on Companies Building Global AI Infrastructure
Investors looking for sustainable growth opportunities should consider focusing on companies that are pivotal in developing the global artificial intelligence (AI) infrastructure. A recent earnings call from a different corporation can sometimes reveal key insights about another company’s potential. This was the case for Broadcom (AVGO), a prominent semiconductor company, which gained unexpected insights into its growth prospects during Oracle’s (ORCL) first-quarter fiscal 2026 earnings call, which concluded on August 31.
Broadcom Benefits from Expanded Data Center Capacity
During Oracle’s earnings call, management emphasized a significant uptick in spending related to AI infrastructure, which could have substantial consequences for its suppliers, including Broadcom. Oracle’s CEO, Safra Catz, along with chairman Larry Ellison, discussed ambitious goals to construct a global AI infrastructure, backed by $455 billion in remaining performance obligations (RPO) — a measure of the contracts they have yet to fulfill — marking an impressive 359% increase year-over-year.
The Wall Street Journal also reported that OpenAI has entered into a five-year, $300 billion agreement with Oracle for computing capacity, necessitating a swift expansion of Oracle’s data center operations. Such extensive projects will rely on suppliers like Broadcom for critical components, including custom accelerators, high-speed Ethernet switches, and optical interconnects, which facilitate high-speed communications within and across AI data centers using fiber optic technology. With demand for data center capacity exceeding current availability, Oracle plans to invest nearly $35 billion in capital expenditures for fiscal 2026, primarily for data center equipment, and is accelerating deployments to optimize its ability to monetize this capacity. This suggests that Broadcom could experience a more robust and consistent demand in the years to come.
Potential for Increased Inferencing Workloads
Oracle’s leadership expressed their intention to become a dominant player in the AI inferencing market, which involves deploying models in real-time. The company is utilizing its status as the largest custodian of private enterprise data by vectorizing data — converting it into numerical formats suitable for AI models. This allows clients to safely connect their data, stored across Oracle’s databases and cloud systems, to leading AI reasoning models, such as those from OpenAI, Alphabet, xAI, and Meta, all hosted on Oracle Cloud Infrastructure (OCI). This integration brings AI inference closer to where enterprise data resides.
However, it is crucial to understand that AI inferencing differs from AI model training. While training involves sporadic, heavy workloads, inferencing is continuous and necessitates minimal latency and high bandwidth for data transfer among various data centers. The anticipated growth in inferencing workloads is likely to boost demand for Broadcom’s ultra-fast Ethernet switches, optical interconnects, and specialized AI chips.
Broadcom’s Financial Performance Indicates Growth
Broadcom’s financial results for the third quarter of fiscal 2025 (ending August 4) reflect a positive trajectory. The company reported a revenue increase of 22% year-over-year, totaling $16 billion, with adjusted earnings before interest, taxes, depreciation, and amortization rising by 30% to $10.7 billion. Additionally, Broadcom announced a record backlog of approximately $110 billion. AI continues to serve as the primary driver of this growth, with AI-related revenue soaring 63% year-over-year to $5.2 billion, 65% of which was attributed to its custom accelerator (XPU) segment. This surge was fueled by heightened demand from three major hyperscale clients rapidly expanding their AI infrastructure. Broadcom has also secured $10 billion in XPU-based AI rack orders from a new client. Management anticipates that AI revenue growth in fiscal 2026 will outpace that of fiscal 2025, driven by Oracle’s aggressive data center expansion and increased inferencing workloads, which could further elevate demand for Broadcom’s custom accelerators.
Networking Challenges Ahead for Oracle and Broadcom
The demand for Ethernet-based networking has surged as large language models advance and AI clusters grow larger. Once these clusters exceed 100,000 GPUs or XPUs, high-bandwidth networking becomes essential for sharing memory across nodes. Broadcom’s Ethernet-based Tomahawk switches and Jericho Ethernet fabric routers have proven highly efficient in connecting GPUs/XPUs within a rack, between racks in a data center, and across data centers. Oracle’s strategy to increase the number of data center regions and enhance inference on enterprise data aligns with this trend, necessitating more nodes, links, traffic, and networking components to interconnect them all.
Broadcom has strategically designed its Ethernet networking solutions independently from its accelerators, offering clients the flexibility to select components that best fit their needs. Hyperscalers prefer Ethernet due to its open-source nature, proven reliability, and familiarity among architects and engineers constructing AI data centers. This preference, coupled with support from various suppliers, mitigates the risk of vendor lock-in for clients. As AI clusters continue to expand, the demand for Ethernet-based networking is expected to rise significantly in the coming years.
Broadcom’s Elevated Valuation
Currently, Broadcom is trading at 36 times its forward earnings, which is considered relatively high. However, with a backlog of $110 billion, increasing demand for its AI-optimized accelerators and Ethernet networking, and Oracle’s substantial investment in AI infrastructure poised to drive sustained spending, the company appears well-positioned for multi-year growth. Long-term investors may want to consider acquiring small stakes in Broadcom stock during any market pullbacks.
