Micron (MU), AMD, and Marvell (MRVL) are positioned to benefit from sustained AI capital expenditure cycles through 2028, but their exposures to memory, GPU processors, and custom silicon create fundamentally different demand and margin profiles. The convergence of these three companies under a single AI capex macro narrative masks critical divergences in competitive positioning and technology cycles.
The semiconductor value chain's heterogeneity means that AI demand acceleration does not translate uniformly across chipmakers. Memory (DRAM/NAND) faces commodity pricing pressure despite volume tailwinds, while GPU-centric players capture pricing power from architectural specialization. Custom silicon operators sit between these poles, exposed to both volume leverage and customer concentration risk. Factor model differentiation suggests the market has not fully priced in these structural differences.
Risk profiles vary materially: memory suppliers face cyclical downside if capex normalizes, GPU vendors risk competitive saturation as entrants proliferate, and custom silicon producers depend on sustained customer relationships vulnerable to in-house silicon strategies. The 2028 timeline introduces technology transition uncertainty—next-generation process nodes, chiplet architectures, and efficiency improvements could materially reshape competitive outcomes versus baseline projections.
Sector implication: The semiconductor complex remains correlated to broad tech but increasingly fragmented by sub-segment. Investor thesis clarity requires disaggregated analysis rather than monolithic AI-tailwind narratives. 2028 price targets inherently carry elevated model risk given the 4+ year horizon and unstable competitive dynamics in specialty silicon.