Deep|QCOM: Better Late Than Never – Massive CPU TAM Growth Sufficient to Accommodate New Entrants

Changes in CPU-to-GPU Ratios and TAM Forecast

In the traditional LLM inference era, a typical configuration consisted of 1 CPU driving 4 to 8 GPUs. However, in architectures driven by Agentic AI, the massive scheduling requirements behind single-GPU inference are shifting the CPU-to-GPU ratio toward 1:1. In certain specific scenarios, the importance of the CPU is even higher. This surge in demand has already triggered a chain reaction in the supply chain. By mid-2026, delivery lead times for server CPUs have extended significantly from the original 1–2 weeks to 8–12 weeks. Both Intel and AMD have implemented price increases for their data center CPU products. Cloud giants such as Amazon and Google are actively expanding their CPU-intensive server clusters to meet the growing demands of AI agent workflows. For Qualcomm, the rise of Agentic AI driving CPU demand presents the optimal timing to enter the server CPU market.

Source: AMD, Funda AI

The Technical Awakening of Nuvia Genes and the Oryon Architecture

Qualcomm’s technical breakthrough in the server CPU field stems from its 2021 acquisition of the startup Nuvia. Nuvia’s founding team was led by Gerard Williams, the former Chief Processor Architect at Apple, who headed the development of Apple’s A-series and M-series chip architectures. This acquisition brought top-tier processor design talent to Qualcomm and provided a custom Arm architecture designed from the ground up, later named the “Oryon” CPU core. The emergence of the Oryon architecture marks Qualcomm’s complete departure from the model of licensing standard cores, moving instead to utilize Arm’s Architecture License Agreement (ALA) to design its own microarchitectures.