Nvidia Just Piled $2 Billion Into This Chip Stock, and It Can Still Climb Higher From Here

Nvidia has taken a stake in Marvell Technologies, signaling that XPUs and custom AI chips could take some share from GPUs in future data-center deployments. Marvell’s interconnect revenue is expected to rise 50% this year, while analysts project earnings growth of 23% this year, 41% in fiscal 2028, and 36% the following year. With a forward P/E of 34, the article argues Marvell remains attractive despite the competitive implications of Nvidia's investment.

Analysis

The market is likely underestimating the signaling value of Nvidia’s capital allocation more than the dollar amount itself. If the dominant GPU supplier is underwriting an XPU specialist, that is effectively a hedge against its own product mix over a 2-5 year horizon and a tacit acknowledgment that custom silicon is becoming a bigger share of AI capex. The second-order winner is not just Marvell, but any infrastructure vendor that sits across both architectures: interoperability lowers switching costs and makes “GPU vs. XPU” a portfolio decision for hyperscalers rather than a binary vendor lock-in choice.

Marvell’s real operating leverage comes from being a toll collector on data movement, not just from accelerator wins. As AI clusters scale, networking, optical, and connectivity spend should rise faster than raw compute spend because every incremental training/inference node increases traffic complexity; that can keep Marvell’s growth resilient even if XPU ramp timing slips. This also creates a subtle competitive threat to smaller point solutions in optical interconnect and custom ASIC design, which may struggle to match Marvell’s breadth plus Nvidia-aligned compatibility.

The key risk is consensus extrapolation: investors may be paying today for a clean multi-year acceleration that depends on hyperscaler capex staying elevated and Microsoft’s Maia ramp arriving on schedule. If big cloud customers pause spending or push out custom silicon deployment by even 2-3 quarters, the multiple compresses before earnings catch up, especially with a mid-30s forward P/E. The trade is therefore less about direction and more about timing: Marvell can work, but only if the market tolerates a long runway without a growth air pocket.

Key Decisions for Investors

• Long MRVL vs. short a basket of slower-growth semiconductor infrastructure names for a 6-12 month relative-value trade; thesis is that interconnect growth plus XPU optionality should outgrow peers even if the broader AI tape cools.
• Buy MRVL on pullbacks only, preferably after post-earnings digestion or any hyperscaler capex scare; target a 12-18 month hold with upside tied to Maia and optical interconnect ramp, but cap downside with a hard stop if custom chip delays appear.
• Use a pair trade: long MRVL / short NVDA small size for a 1-2 quarter horizon as a hedge on AI architecture diversification; this benefits if custom silicon adoption rises faster than GPU replacement risk is priced, but should be sized modestly given Nvidia’s stronger execution.
• Consider selling cash-secured puts on MRVL 10-15% below spot into volatility spikes; the implied forward growth already supports valuation, and put premium is attractive if you want to own the stock on any capex-driven selloff.
• Watch MSFT next custom-silicon commentary as the highest-signal catalyst; if Microsoft confirms accelerated Maia deployment, add to MRVL immediately, since the market will likely re-rate the stock before reported revenue inflects.