Although there have been signs of a rotation out of the tech sector in the stock market this year, the artificial intelligence (AI) super cycle is far from over. However, the focus of this cycle has shifted, which will benefit a new batch of AI stocks.
The initial boom in AI infrastructure was driven by the demand for raw computing power to train large language models (LLMs). This race once generated enormous profits for Nvidia. But the next phase of AI will center on networking, inference, and agentic AI—and a select group of stocks is poised to benefit from these trends.
Broadcom (AVGO) sits at the intersection of two major new trends within the AI super cycle. The first trend is networking. Broadcom is a leading supplier of Ethernet switches and other data center networking components. As the number of AI chips in a single AI cluster surpasses one million, the focus is no longer solely on the peak performance of individual chips but on the performance of the entire cluster. Broadcom’s Ethernet switching solutions, optical interconnects, and other components play a critical role in reducing congestion and ensuring the system operates at optimal levels.
At the same time, Broadcom is also a leader in application-specific integrated circuit (ASIC) technology, helping customers design custom chips for specific tasks. AI ASICs are particularly well-suited for the inference market because their tailored design maximizes data throughput and significantly reduces energy consumption.
Broadcom’s partner in the AI ASIC space is Alphabet (GOOGL) (GOOG). With Broadcom’s help, Alphabet has developed the world’s most advanced AI ASIC—the Tensor Processing Unit (TPU). Having started development a decade ago, Alphabet has been able to optimize its entire hardware and software stack around the TPU, gaining a significant structural cost advantage over competitors that rely primarily on Nvidia GPUs.
Alphabet applies this advantage to its own AI workloads. According to estimates from SemiAnalysis, Alphabet’s total cost of ownership is 44% lower than a comparably configured Nvidia server. The high efficiency of these chips makes them ideal for large-scale inference tasks. Today, Alphabet has begun offering TPUs to its customers, with Anthropic placing a $21 billion order through Broadcom. Some of these chips will be rented out via Google Cloud, while others will be sold directly.
Another company set to benefit from the growing demand for data center networking is Arista Networks (ANET). While Broadcom provides networking components, Arista Networks packages those components with its Extensible Operating System (EOS) software platform, offering customers an easy-to-manage networking solution.
Arista Networks’ largest customers include Microsoft (MSFT) and Meta Platforms (META). Like Broadcom, it will benefit from the ever-increasing scale of AI chip clusters. Its newly launched Blue Box program provides enhanced diagnostic capabilities for hardware platforms, adding another growth driver for the company.
Summary: Overall, the AI super cycle is not over; it has simply shifted from pure computing-power training toward new directions such as networking, inference, and customized chips. Broadcom, with its dual strengths in Ethernet switching and ASIC design; Alphabet, leveraging the cost and efficiency advantages of its self-developed TPUs; and Arista Networks, through its software-integrated networking platform, are emerging as representative growth stocks in the next phase of the AI wave. Investors should focus on this structural shift rather than merely continuing the chase after traditional AI hardware giants.
