Qualcomm down 35% as smartphone weakness masks AI infrastructure upside

Qualcomm's near-term troubles in handset chips are real but cyclical, while the company's positioning in AI infrastructure reflects a structural shift in semiconductor demand. The smartphone weakness stems from two sources: temporary memory supply constraints squeezing handset OEMs' chip purchases, and a permanent structural loss as Apple transitions to modems built in-house. Both pressures are already visible in the valuation—the stock trades at only 16x forward earnings despite projections of $10.20 per share in fiscal 2027, down slightly from $10.50 in fiscal 2026. Once the Apple transition completes by fiscal 2027, the handset comparisons stabilize on a smaller but steadier base. Meanwhile, the company's legacy cash-generation engine (the smartphone and licensing business) produced close to $10 billion in operating free cash flow, providing a cushion while the AI business develops. The opportunity that follows taps into Qualcomm's core strength: decades of optimization for performance per watt, the metric now constraining AI data-center expansion more than pure compute. Power infrastructure is the bottleneck—securing grid connections in key U.S. markets takes 36 to 84 months, and more than 2,000 GW of generation and storage capacity sits in the interconnection queue. Inference, the step of running trained models in production (as opposed to training them), is projected to account for roughly two-thirds of AI compute by 2029, and agentic AI systems that execute tasks autonomously could further drive demand for efficient inference hardware. Qualcomm is already moving: it has announced the AI200, AI250, and AI300 systems optimized for power efficiency and total cost of ownership, secured a 200 MW commitment from HUMAIN for 2026 onward, and obtained a multi-generation agreement from Meta for server CPUs based on the Oryon architecture acquired from Nuvia.