Beyond Nvidia: AMD and Broadcom Positioned for AI Inference Cycle

While Nvidia remains the dominant force in AI infrastructure, a strategic shift is occurring as the industry moves from the training phase of large language models (LLMs) toward the inference phase. Inference, where AI models apply trained data to generate outputs, is projected to become a significantly larger portion of the market and is less technically demanding than initial training. Advanced Micro Devices (AMD) is carving a niche in this space, bolstered by new partnerships with Meta Platforms and OpenAI. Beyond GPUs, AMD is positioned to benefit from the rise of 'agentic AI,' which increases demand for high-performance CPUs to handle sequential logic. This sector is already seeing supply constraints, as evidenced by Amazon's inability to meet all customer demand for its custom Graviton CPUs. Broadcom (AVGO) is emerging as a primary beneficiary through its leadership in application-specific integrated circuit (ASIC) technology. By designing custom chips hardwired for specific tasks, Broadcom offers superior energy efficiency compared to programmable GPUs, which is critical for managing the ongoing costs of inference. Broadcom's role in the custom silicon space is further highlighted by its $21 billion TPU supply agreement with Anthropic and an extended five-year partnership with Alphabet. As data center operators seek to lower operational costs and optimize performance, the shift toward ASICs and specialized CPUs may diversify the AI hardware landscape.