Wall Street Analyst Sees Between 19% to 37% Upside in These 5 AI Chip Stocks

Piper Sandler initiated overweight coverage on Nvidia, Broadcom, Arm, AMD and Marvell, projecting 19%-37% upside and forecasting the compute market will expand fivefold to $2.2 trillion by 2030, including roughly $2 trillion for AI chips. The firm cites AI-chip supply constraints, with H100 GPU rental prices up 30% from $2.00 in December to $2.60 per hour and AI-chip order books reportedly full into 2028. Nvidia carries the highest implied upside at 37% to a $300 target, while Broadcom's AI revenue is projected to rise 400% over two years to $230 billion in fiscal 2028.

Analysis

The useful signal is not another bullish sell-side initiation; it is the implied persistence of accelerator scarcity. Rising rental rates for prior-generation GPUs suggest cloud capacity has not caught up with demand, supporting near-term pricing power for NVDA and, more importantly, a broader spend cycle in networking, memory, power, and optical interconnect. AVGO is the cleaner diversified beneficiary because custom silicon adoption by hyperscalers expands its addressable market without requiring a wholesale displacement of Nvidia.

Over the next 1-3 months, consensus target-price upgrades are unlikely to be a durable catalyst after a broadly publicized AI-chip rally. The investable inflection is quarterly evidence that hyperscaler capex is shifting from concentrated GPU purchases toward scale-out infrastructure: AVGO networking, MRVL optical DSPs, and server CPUs from AMD. This would also pressure INTC, whose server-CPU share losses become more consequential if agentic workloads require materially higher CPU-to-accelerator ratios.

The contrarian risk is that booked supply through 2028 reflects non-cancellable capacity reservations rather than final end-demand. A normalization in neocloud GPU rental rates, AI-lab funding, or hyperscaler capex guidance would compress the premium multiples first in ARM and MRVL, where execution expectations leave less room for a digestion period. ARM's move toward selling more complete silicon solutions also creates a licensing-channel conflict with customers, making royalty growth less linear than the market assumes.

For 6-18 months, the key competitive question is inference economics rather than training share. ASIC penetration benefits AVGO and potentially MRVL, while AMD needs disclosed customer deployments and software adoption to prove its inference opportunity is incremental rather than merely a lower-price alternative. NVDA remains structurally advantaged, but its relative returns could lag if AI capex broadens from accelerators into connectivity and custom silicon.

Key Decisions for Investors

• Initiate a 3-6 month long AVGO / short ARM pair, sized beta-neutral. AVGO offers exposure to ASIC and networking monetization with lower valuation fragility; ARM is vulnerable if royalty upside is delayed by customer pushback or server-chip execution. Reassess if ARM reports materially accelerating royalty revenue or AVGO's AI revenue/order commentary weakens.
• Maintain NVDA core exposure but rotate a portion of incremental AI allocation into AVGO ahead of the next hyperscaler earnings cycle. The catalyst is evidence of networking and custom-silicon spend broadening; invalidate the rotation if cloud capex guidance is cut or GPU rental pricing declines sequentially.
• Use MRVL only as a watch-list long rather than a fresh directional position until Alphabet-related revenue timing, optical margin trajectory, and Amazon IP exposure are independently quantified. A confirmed design-win ramp with gross-margin stability would justify entry; loss of a major custom-chip role would make the current multiple vulnerable.
• Express AMD upside through a defined-risk 6-9 month call spread only after customer-specific inference revenue or server-CPU share data confirms the thesis. Falsification: flat-to-down data-center guidance, delayed accelerator ramp, or evidence that CPU demand does not scale with agentic workloads.
• Remain underweight INTC versus AMD for the next 2-3 earnings cycles. The risk/reward improves only if Intel demonstrates sustained server-share stabilization and foundry/customer traction sufficient to offset ongoing margin pressure.