AI Infrastructure Shift: Broadcom and AMD Positioned for Next Supercycle

The artificial intelligence landscape is transitioning from a focus on individual chip performance to the efficiency of massive chip clusters, creating new opportunities for infrastructure providers. While Nvidia dominated the first wave of AI, the emerging 'next supercycle' is expected to favor companies specializing in high-speed networking and custom accelerators. Broadcom is leveraging its leadership in data center networking, specifically its Tomahawk Ethernet solutions, to support clusters that are projected to reach one million chips. Beyond networking, the company is a leader in application-specific integrated circuit (ASIC) technology, providing the intellectual property necessary for customers to manufacture custom chips at scale. Broadcom has already supported Alphabet's tensor processing units (TPUs) and has deals with Anthropic, OpenAI, and Meta. The company projects custom chip sales will reach $100 billion in fiscal 2027. Simultaneously, Advanced Micro Devices (AMD) is positioning itself at the intersection of inference and agentic AI. The upcoming MI450 GPU utilizes a chiplet design to increase memory and improve inference performance. AMD has secured significant commitments from OpenAI and Meta, with each pledging 6 gigawatts' worth of new GPUs. These partnerships include the issuance of warrants, granting the AI giants a stake in the company. Furthermore, the rise of agentic AI is expected to increase the CPU-to-GPU ratio in server racks, as CPUs are required to handle the sequential logic and real-world interactions of AI agents. This trend provides a substantial tailwind for AMD's data center CPU business as the industry moves toward more complex, autonomous AI workflows.