Marvell pops 6% on AI chip deal that lets Google buy up to $12.2 billion in shares

Marvell shares jumped ~6% after a filing said Google could buy up to $12.2B of Marvell shares—58,970,907 shares at $206.58 each—under an expanded partnership tied to purchasing targets through FY2033. The agreement is linked to products that attach to Google’s TPU ecosystem, including AI inference accelerators plus storage and network interface controllers. The news underscores intensifying AI custom-silicon competition as Google seeks alternatives to Nvidia.

Analysis

This is less a near-term revenue re-rating than a strategic validation of custom-silicon economics: hyperscalers are still willing to pay for design partners that help them pull AI compute spend away from merchant GPUs. For MRVL, the key market mechanism is not just incremental socket wins but higher perceived durability of its AI attach rate; however, the revenue recognition is stretched over a decade, so today’s pop likely outruns the near-term P&L impact.

The more interesting second-order read-through is to NVDA and AVGO. If Google keeps deepening a TPU-centered stack, it compresses the addressable market for the highest-cost inference workloads first, which is where merchant GPU pricing power is most vulnerable; AVGO is less likely to be a clean loser because its broader custom-chip franchise still benefits from hyperscaler capex, but relative share expectations can swing on who controls more of the architecture. For GOOGL, this is a margin-supportive move structurally, because every unit of workload shifted in-house improves AI economics and reduces dependence on external pricing.

Contrarian view: the market may be treating this as a pure win for MRVL, when the bigger signal is customer concentration risk. A long-only rerate is less compelling if investors conclude Google is simply using a second source to keep bargaining power over suppliers, which would cap multiples for MRVL and AVGO alike. The thesis is falsified if Google commentary shows slowing AI capex, if MRVL fails to convert this into visible 12-18 month revenue acceleration, or if NVDA pricing/mix remains resilient through the next two earnings cycles.

Key Decisions for Investors

• Long MRVL / short NVDA as a 1-3 month relative-value expression: plays custom-silicon share shift and lower merchant-GPU pricing power; stop if NVDA commentary shows no pricing erosion or MRVL guidance does not inflect next quarter.
• Fade the knee-jerk MRVL rally with a short-dated put spread or call spread hedge into the next earnings window; the deal is strategically positive but economically back-end loaded, so near-term upside may be limited.
• Watch/avoid a clean short AVGO: the stock is vulnerable to incremental share-loss headlines, but its broader custom-chip book means the better trade is only on confirmation that Google is reallocating architecture spend away from Broadcom.
• Overweight GOOGL versus NVDA on a 6-18 month horizon if you want the structural margin-improvement story; thesis breaks if AI capex growth slows or if external accelerators reassert performance-per-dollar leadership.