Piper Sandler says AI is causing 'insatiable demand' for compute capacity. These two stocks will benefit most

Piper Sandler initiated overweight coverage on AMD, Nvidia, Broadcom and Arm, citing agentic AI-driven compute demand and forecasting average EPS CAGR of 45% through 2030 for the group. The firm set a $600 AMD target, implying 15% upside, and a $300 Nvidia target, implying 34% upside; GPU prices have risen 25%-40% year-to-date. Piper views compute valuations at 14-22x P/E as attractive and estimates roughly 30% average upside, while assigning neutral ratings to Intel and Qualcomm.

Analysis

The investable distinction is shifting from aggregate AI-capex exposure to who monetizes inference at scale. NVDA retains the highest pricing power where proprietary software and networking lock in full-rack deployments; its upside depends less on another model-training cycle than on conversion of announced capacity into revenue-bearing enterprise deployments over the next 2-4 quarters. AMD offers higher operating leverage if hyperscalers diversify supply, but its valuation case requires evidence that deployments extend beyond a small concentration of strategic customers; without that, incremental demand may primarily improve utilization rather than establish durable share.

The underappreciated constraint is physical deployment: grid interconnects, power conversion, cooling, and networking can delay GPU revenue recognition even when chip demand is firm. That favors ETN and VRT as lower-beta second-order beneficiaries of multi-gigawatt data-center commitments, while AVGO is positioned to capture the networking/custom-silicon alternative if customers seek lower total cost of ownership versus merchant accelerators. Over 6-18 months, growing inference mix could pressure accelerator ASPs but expand total compute demand; the critical question is whether token economics improve fast enough to prevent customer ROI scrutiny.

Consensus likely overweights the reported accelerator-price signal and underweights timing risk: capacity reservations are not equivalent to installed, powered clusters. A 1-3 month pullback is plausible if hyperscaler capex commentary remains strong but deployment schedules slip; that would create a better entry in the leaders rather than invalidate the structural theme. The thesis is falsified by sequential hyperscaler AI-capex reductions, material extension in data-center commissioning timelines, or evidence that inference workloads migrate materially to lower-cost ASIC/CPU architectures.

Key Decisions for Investors

• Maintain/establish long NVDA on 5-8% weakness over the next 1-3 months; target a 15-20% re-rating if enterprise-scale deployment wins translate into forward revenue visibility. Reduce if next earnings shows data-center backlog conversion slowing or gross-margin guidance falls materially below consensus.
• Use a pair trade: long AMD / short INTC for 6-12 months, sized modestly given AMD execution risk. The trade captures server-CPU share transfer and AI optionality; exit if AMD fails to sustain server revenue growth or INTC demonstrates credible foundry/customer traction that narrows its execution discount.
• Prefer AVGO plus ETN or VRT as a diversified AI-infrastructure basket over adding only accelerator exposure; initiate in tranches through the next two earnings cycles. This hedges against accelerator share volatility while retaining exposure to network, power, and cooling intensity; reassess if hyperscaler capex guidance turns negative or data-center project delays widen.
• Do not initiate a directional QCOM long solely on this theme until management discloses material high-performance compute design wins, volume commitments, and associated margin profile; treat such disclosure as a catalyst alert rather than a current recommendation.