Broadcom, AMD race past S&P 500 on AI chip demand

Broadcom and AMD represent two distinct approaches to capitalizing on AI infrastructure growth. Broadcom dominates the ASIC market, providing custom chips directly to hyperscalers, while AMD competes in the broader GPU and CPU space. The divergence in their product strategies is becoming strategically significant as the AI industry evolves. Broadcom's guidance for 84% year-over-year revenue growth in fiscal Q3 suggests momentum in ASIC demand is accelerating faster than expected, with AI semiconductor revenue alone anticipated to more than triple year over year. AMD's 41% guidance for next quarter, while lower in percentage terms, reflects a different growth trajectory driven by its established position in both GPUs and CPUs across the broader data center market.

The structural shift toward agentic AI is reshaping the economic calculus of AI infrastructure in ways that may favor CPU-dependent architectures. Historically, AI training required one CPU per eight GPUs; the move toward agentic systems is reversing this ratio, with industry guidance suggesting four CPUs per GPU or even one-to-one ratios in certain deployments. This change has material implications for AMD, whose CPU business stands to benefit from increased infrastructure spending on processors as hyperscalers rebuild their systems to support these new ratios. Neither company's guidance has been fully updated to reflect how deeply this ratio shift will penetrate AI data centers, which leaves room for upside surprises as deployment patterns stabilize.