Marvell Q2: Buying This Dip Looks Premature

The market is likely punishing the duration shift, not the absolute economics. When a hyperscaler program is already embedded three years out, the near-term reacceleration narrative loses convexity, so MRVL can de-rate even while the fundamental track remains intact. That usually creates a brutal but temporary mismatch: the stock price reacts to slower slope, while the earnings model only changes modestly unless management is forced to walk down FY27/FY28 again.

Second-order, this is constructive for GOOGL's unit economics if the underlying silicon stack is increasingly internalized. The benefit is not top-line; it's margin durability and less dependence on merchant suppliers as AI workload intensity rises. That also matters for peers like AVGO and other custom-ASIC incumbents: if one hyperscaler can keep more inference economics in-house, it raises the bar for every external chip vendor chasing the same wallet share.

The key risk is that the selloff becomes self-fulfilling if investors start discounting a longer plateau in MRVL's AI revenue mix. Over 1-3 months, the stock likely trades on whether management can show incremental non-Google design wins or improved visibility into backlog conversion; over 6-18 months, the thesis flips only if Google programs expand beyond current assumptions or if another hyperscaler ramps enough to offset the timing shift. The contrarian point: this may be more of a multiple reset than a fundamental break, so the downside should be bounded unless guidance is revised again.