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 as AI-compute demand continues to exceed chip supply. The firm forecasts the compute market will expand fivefold to $2.2 trillion by 2030, including roughly $2 trillion for AI chips, citing sold-out supply through next year and order books extending into 2028. Nvidia carries the highest target at $300, or 37% upside, while Broadcom's AI revenue is projected to rise 400% over two years to $230 billion in fiscal 2028.

Analysis

The incremental analyst initiation is unlikely to alter fundamentals, but it reinforces a crowded expectation that AI infrastructure shortages will persist. The key differentiation is monetization durability: NVDA and AVGO retain pricing power because their platforms sit at the highest-value bottlenecks, while AMD, MRVL, and ARM require design wins to convert into material earnings. Reported multiyear capacity reservations should be discounted until backed by customer capex guidance and supplier backlog disclosures; hyperscaler commitments can be renegotiated if model efficiency improves or utilization disappoints.

AVGO is the cleaner relative beneficiary of an inference-led spend cycle: custom silicon and networking content scale with hyperscaler deployments even where customers seek to reduce dependence on NVDA. A second-order risk is that successful ASIC adoption caps the addressable portion of GPU inference, making NVDA's revenue mix and gross-margin trajectory more sensitive to the pace of training demand than consensus assumes over the next 6-18 months. Conversely, higher GPU rental rates are not unambiguously bullish: if they reflect scarce supply rather than sustainably high utilization, they invite customer-owned capacity and ASIC substitution.

ARM has the most asymmetric downside among the large-cap AI beneficiaries because moving further into physical designs risks channel conflict with the licensees that generate its royalty stream; valuation leaves little tolerance for delayed server CPU adoption. MRVL's upside is similarly concentrated in a small number of custom-silicon and optical programs, making it vulnerable to program timing, qualification delays, and customer insourcing. INTC is the underappreciated hedge: sustained CPU demand for agentic workloads could stabilize server share losses, but this requires evidence of product competitiveness rather than merely a larger market.

Near term, the sector can trade higher on capex revisions and supply commentary; over 1-3 months, earnings revisions and hyperscaler 2027 capex guides matter more than price targets. The trend is falsified by declining accelerator rental utilization, a sequential cut in AMZN/GOOG capex plans, or evidence that inference token costs are falling faster than deployment volumes rise.

Key Decisions for Investors

• Initiate a 3-6 month pair: long AVGO / short ARM, sized beta-neutral. AVGO has more direct content-per-deployment capture and less valuation dependence on a distant server-CPU outcome; target 15% relative upside, stop at 8% relative underperformance or on ARM royalty guidance accelerating above expectations.
• Maintain NVDA exposure but fund part of it with a small short MRVL basket over the next 1-3 months if MRVL rallies without incremental customer backlog disclosure. The thesis is concentration and program-timing risk versus NVDA's broader installed base; cover if MRVL provides binding volume commitments or raises forward gross-margin guidance.
• Do not chase AMD on generalized agentic-AI enthusiasm. Upgrade to a long only after a disclosed MI-series backlog conversion, server-CPU share data, or gross-margin guidance demonstrates that inference wins are accretive; absent that evidence, use AMD as the higher-beta expression only on broad AI-capex pullbacks.
• Set alerts around AMZN and GOOG capex guidance, GPU-rental utilization/pricing, and AVGO AI semiconductor bookings. A synchronized capex increase supports longs across AVGO/NVDA; a utilization decline or hyperscaler capex reset warrants reducing semiconductor beta before the next earnings cycle.