Marvell shares gain on report of deal talks with Google to develop two AI chips

Marvell Technology shares jumped 7% in premarket trading after reports that Google is discussing two new AI chips with the company, including a memory processing unit and a new TPU. The potential deal suggests Google may be diversifying chip supply away from Broadcom as AI demand accelerates. Marvell could add more than $9 billion in market value if the move holds, with analyst sentiment already at a buy rating and a $125 median target.

Analysis

The important signal here is not just incremental chip demand for Marvell, but a potential change in customer concentration risk within custom AI silicon. If Google is widening its vendor set, that weakens Broadcom’s quasi-franchise status and creates a bidding channel for second-source design wins across hyperscalers that want bargaining leverage, faster iterations, and supply-chain redundancy. In the near term, the market is likely to extrapolate a multi-quarter revenue uplift for MRVL, but the larger second-order effect is margin mix: custom silicon programs often start with favorable optics and later compress as competitive pricing and NRE amortization normalize.

For Google, the strategic value is optionality. A memory-processing adjunct plus a new TPU would help reduce inference bottlenecks and unit cost per token, which matters most if AI product usage keeps scaling faster than public cloud capex efficiency improves. That is mildly negative for NVDA at the margin because every credible in-house accelerator roadmap chips away at TAM growth, but the bigger overhang is on AVGO if hyperscalers increasingly dual-source design services to avoid single-vendor dependency. Meta’s prior commitment reinforces the broader message that hyperscaler silicon budgets are becoming more fragmented, not more concentrated.

The move in MRVL can persist for weeks if sell-side models start capitalizing design-win probability rather than revenue recognition timing, but the stock is vulnerable if the story remains “talks” rather than tape-out, since the revenue is 12-24 months out and execution risk is high. Contrarian view: the market may be underestimating how little of this flow converts to near-term EPS; AI custom chip programs are strategically important but usually lumpy, delayed, and heavily discounted by customers once they gain leverage. That means the cleanest expression is not chasing MRVL outright, but positioning around relative winners and losers in the custom silicon supply chain.

Key Decisions for Investors

• Long MRVL / short AVGO as a 1-3 month relative-value trade: if Google is truly diversifying, MRVL gets the incremental narrative premium while AVGO faces de-rating risk from vendor-share concerns; stop if AVGO reasserts as sole design partner.
• Buy MRVL on a pullback after the opening squeeze rather than at the gap: the right entry is post-news consolidation, targeting a 6-12% upside window with a tighter stop below the post-event VWAP.
• Small tactical short NVDA via weekly call spreads or a 1-2 month put spread: use this only as a hedge against broader AI-customization headlines, since the fundamental impact is second-order and likely capped unless more hyperscalers announce similar moves.
• Add GOOGL on weakness, not strength: custom-chip optionality improves long-run cloud economics and inference margins; the risk/reward is better over 6-12 months than on a single-day headline move.
• Avoid chasing META on this headline alone: its custom-silicon strategy is already broadly recognized, so the incremental alpha from this article is low versus MRVL/AVGO pair exposure.