AI Expectations Hit a Fever Pitch

The near-term loser is the memory complex, not because demand is weak, but because positioning is too one-way and balance-sheet/capex optics are starting to matter. MU and the Korean memory cohort are vulnerable to a 1-2 quarter de-rating if buyers see any pause in hyperscaler ordering or if new supply from capital raises keeps wafer pricing from tightening as fast as the market expects. In contrast, the real beneficiaries of the AI spend race are the platforms with the cheapest funding access: AMZN and GOOGL can keep extending compute/infrastructure capex without stressing leverage, which supports NVIDIA’s revenue bridge in the next two quarters even if the equity tape stays volatile.

In defense, the headline risk is backwards: cheap drones do not automatically destroy LMT/NOC/RTX economics; they shift mix toward sensors, autonomy, underwater, and integration layers where primes retain pricing power and contract access. LMT’s bolt-on into undersea warfare is strategically useful because it increases share of wallet in a priority budget line, while LHX is better positioned than the legacy airframe names if digital payloads and ISR continue to displace pure metal-bending. The second-order loser is the standalone middle tier: AVAV/KTOS can win share, but the market may still overestimate how much margin survives at consumables-like economics.

Contrarian view: the selloff in AI memory may be overdone if investors are conflating ‘high expectations’ with ‘cycle peak.’ The faster falsifier is not price action but evidence that hyperscaler capex growth slows or that memory ASPs stop improving over the next 1-2 quarters. If that does not happen, today’s reset could become a buying opportunity in MU/SSNLF rather than the start of a downturn. In defense, the consensus that drones are pure disruption misses that primes are often the acquirers and integrators, not the displaced incumbents.