RBC raises Marvell stock price target on AWS chip strength By Investing.com

RBC Capital raised its Marvell price target to $170 from $115 and kept an Outperform rating, citing stronger near-term visibility from PAM-4 optical connectivity and AWS Trainium chip production. RBC now estimates about $1.6B in custom silicon revenue from AWS this year, up roughly 17%, and expects the AWS-Anthropic partnership to support double-digit growth into fiscal 2028 and beyond. The stock has already risen more than 200% over the past year and is trading near its 52-week high, which may temper further near-term upside.

Analysis

The market is correctly reading this as more than a single-vendor increment for one chip designer; it is a signaling event that re-rates the entire AI infra stack. A larger, longer-duration Anthropic buildout increases confidence that hyperscalers will keep funding multiple custom-compute architectures simultaneously, which is supportive not just for the direct silicon supplier but also for optical interconnect, packaging, and power-delivery vendors that sit in the bottleneck path. The second-order implication is that capacity scarcity shifts from ‘AI demand uncertain’ to ‘supply chain allocation uncertain,’ which tends to preserve pricing power deeper into the cycle.

The near-term constraint is not demand, it is execution capacity: advanced-node wafer availability, CoWoS-like packaging, high-speed optics, and deployment cadence. That means upside can still accrue even if headline AI spending is unchanged, but the beneficiaries will rotate toward vendors with the cleanest share of wallet and the least manufacturing friction. Conversely, any slip in AI training ROI, Anthropic utilization, or customer concentration narrative would hit multiples fast because the stock reaction has already discounted several years of growth.

Consensus is underestimating how much this helps Marvell’s optics and custom silicon mix relative to generic AI exposure. The more customized the cloud stack becomes, the more valuable the ‘picks and shovels’ layers become — especially where switching costs are high and redesign cycles are long. The risk is that this turns into a crowding trade: if investors extrapolate every AWS/Anthropic dollar into perpetual high-multiple growth, the stock can become vulnerable to even a modest guide-down or supply hiccup over the next 1-2 quarters.

For the broader basket, Nvidia is still structurally positive, but this is a reminder that the marginal dollar of AI capex is now diffusing into adjacent infrastructure winners. That creates a relative-value opportunity versus names that are more exposed to one customer, one node, or one product cycle. The cleanest expression is to own the constrained enablers, not the most obvious AI beta.

Key Decisions for Investors

• Long MRVL vs short a high-multiple semicap proxy on a 3-6 month horizon: the thesis is not just revenue growth, but improved visibility into custom-silicon pull-through and optical attach rates. Use a market-neutral pair to isolate AI infra share gain; target 15-20% relative outperformance if AWS capex stays firm.
• Add to NVDA on dips, but hedge with a small MRVL long: this captures continued AI capex strength while expressing that incremental spend is rotating into ecosystem beneficiaries. Best entry is on any 3-5% pullback in NVDA; risk is that supply-chain constraints cap near-term upside across the group.
• Sell near-dated MRVL upside via covered calls or call spreads if already long: the stock has re-rated aggressively, so implied growth is rich. Prefer 1-2 quarter tenors to monetize elevated expectations while retaining exposure to the secular story.
• Tactically overweight optical/interconnect exposure over generic AI hardware for the next 6-12 months: the second-order beneficiary set should benefit from sustained bandwidth demand and deployment complexity. Favor names with diversified customer exposure and visible backlog to reduce single-customer event risk.
• Watch for any sign of packaging or wafer allocation slippage; if that emerges, reduce MRVL first rather than the broader AI basket. A 5-10% drawdown on an execution miss is plausible given how much optimism is already embedded.