MXL’s near-term win is not “AI exposure” in the abstract; it is socket-level share capture in power-constrained hyperscale builds. If Keystone’s watt-per-bit advantage is real, it should improve design-win durability and force competitors to compete on total platform economics rather than specs alone, which is harder for narrower optical vendors to defend. The second-order beneficiary is hyperscalers themselves: lower power per port can raise rack density and defer incremental power-capex, which tends to favor faster deployment cycles over the next 1-3 quarters.

The stock, however, is now in a prove-it window. Over the next 1-3 months, the market will care less about TAM narratives and more about whether revenue growth converts into stable margins and repeat orders; if gross margin stays near guide while optical revenue is re-accelerating, that supports further multiple expansion, but any mix or pricing miss will invite a sharp de-rating. Longer term, the real catalyst is Rushmore qualification; delays into 2028 would push out the bull case and expose how much of the current move is already discounting optionality that has not yet become revenue.

Contrarianly, consensus may be underestimating how fragile a concentrated design-win story can be after a 270% YTD move and a premium sales multiple. AVGO and MRVL do not need to “lose” broadly to cap MXL; they only need to bundle sufficiently good solutions across switches, DSPs, and retimers to keep MXL from converting Keystone momentum into a durable 1.6T share position. The setup is attractive, but upside from here likely depends more on repeated guide raises than on the AI optics theme itself.