Meta Platforms and Google recently announced encouraging updates for Broadcom shareholders.
Big tech is prepared for a colossal investment spree in 2025, with the top four tech giants collectively planning to allocate a staggering $320 billion in capital expenditures (capex). The majority of this budget will be allocated to constructing and modernizing AI data centers, primarily for training and utilizing generative AI.
Nvidia is expected to reap significant benefits from this enhanced spending. Notably, major tech companies have highlighted the importance of their associations with Nvidia, the leading GPU designer. However, remarks from Meta Platforms and Alphabet's management hint at a promising future for another chipmaker as well.
Broadcom could emerge as a significant winner from the subsequent surge in big tech spending. Here's why:
The Ascent of Custom AI Chips
Broadcom is instrumental in the AI data center operations of big tech players like Meta and Google through its custom AI chips. Broadcom engineers and manufactures chips for network switches, ensuring efficient data flow between data centers' servers. By tightening the performance gap between Nvidia chips (GPUs) and its network switches, Broadcom plays a crucial role in optimizing data center efficiency.
Besides network switches, Broadcom also produces specialized chips that Meta and Google utilize in their data centers. Its collaboration with Meta resulted in the creation of the Meta Training and Inference Accelerator (MTIA), and Google's 2015 partnership led to the development of Tensor Processing Units (TPUs).
Meta and Google openly commended their custom-built AI chips during their fourth-quarter earnings calls. Meta's CFO, Susan Li, revealed plans to further integrate MTIA into its AI data centers, with the aim of eventually handling core AI training workloads for the company. Google CEO Sundar Pichai expressed optimism over the robust adoption of its sixth-generation TPUs, suggesting a promising 2025 for these chips.
Custom AI accelerator chips have proven to be more energy-efficient and suitable for various AI workloads than general-purpose GPUs. The increasing scale of big tech data centers necessitates leveraging these microscopic efficiencies, which in turn leads to substantial savings in the long term. This impetus drives the importance of expanding the applications for custom chips and carrying out full control over the technology stack as capex budgets grow.
Broadcom: Poised for Growth
Broadcom isn’t just limited to Meta and Google. Ranked among its three largest customers is TikTok's parent firm ByteDance. Additionally, Broadcom announced two new clients crafting custom next-generation chips during its fourth-quarter earnings call, sparking speculation over the identities of these companies, with Apple and OpenAI among the frontrunners.
Apple demonstrated a preference for Broadcom's chips by using Google's previous-generation TPUs in its AI system in 2023. If Apple decides to take control of its AI training and development, it will quickly become Broadcom's most significant client.
Broadcom anticipates that the market for custom AI accelerators will reach between $60 billion and $90 billion within three years. In the fiscal year 2024, Broadcom dominated this market, accounting for roughly 70% of AI accelerators and generating nearly $12 billion in revenue. Given Broadcom's diversified operations, a doubling of the AI-driven semiconductor market growth could dramatically uplift its overall revenue.
However, while Broadcom is poised for impressive growth, its stock’s premium valuation may deter prospective investors. Broadcom's share price currently trades at 37 times forward earnings, and analysts expect around 30% earnings growth this year. For a more reasonable valuation or a potential growth beyond expectations, investing in Broadcom might prove attractive when its stock price falls to around 30 times its 2025 earnings, or if the adoption of its custom chips surpasses expectations in Meta and Google's data centers.
Broadcom is an enticing stock to monitor for artificial intelligence investors. Yet, its current valuation is significantly steep, making it hard to justify an investment.
- The investment spree in 2025 by big tech giants, worth $320 billion, will see a significant portion allocated to constructing and modernizing AI data centers, with a focus on training and utilizing generative AI.
- Broadcom, a leading chipmaker, is instrumental in AI data center operations of major tech companies like Meta and Google, providing custom AI chips and ensuring efficient data flow with its network switches.
- Meta's CFO, Susan Li, revealed plans to integrate Meta Training and Inference Accelerator (MTIA) into its AI data centers, aiming to handle core AI training workloads, while Google's CEO, Sundar Pichai, expressed optimism over the robust adoption of its sixth-generation TPUs in 2025.
- Broadcom anticipates that the market for custom AI accelerators will reach between $60 billion and $90 billion within three years, potentially making it the dominant player in this market and significantly boosting its revenue with a doubling of AI-driven semiconductor market growth.
