Andy Jassy Just Made a Move That Should Excite Micron and Broadcom Investors

Amazon plans to raise 2026 capital spending to $220 billion from $200 billion to expand AI infrastructure, with CEO Andy Jassy saying AWS will still lack sufficient capacity to meet demand through 2026 and likely 2027. Jassy also characterized demand visibility for 2028 as "striking," reinforcing a multi-year AI data-center investment cycle. The spending outlook is positive for Micron's AI memory demand and Broadcom's networking and custom-chip businesses, while countering investor concerns that current AI infrastructure investment may be outpacing end demand.

Analysis

The investable read-through is stronger for AI-networking and custom-silicon content than for generic cloud exposure. Incremental hyperscaler capacity requires a disproportionate increase in bandwidth, switching and interconnect as clusters scale, supporting AVGO’s higher-value content per deployed AI rack; revenue recognition should lag construction commitments by roughly 2-4 quarters. AMZN benefits only if utilization and AI-service pricing hold up: sustained capex without an accompanying acceleration in AWS growth or operating-income conversion would turn the narrative from demand-led expansion to free-cash-flow dilution.

MU is the least direct beneficiary and carries the greatest near-term expectation risk. Tight memory markets can produce sharp earnings torque, but capacity announcements and customer inventory normalization typically reverse pricing faster than data-center capex cycles; Amazon’s spending does not establish its HBM/DRAM sourcing mix or guarantee MU share. The more important second-order implication is a relative headwind for NVDA if hyperscalers continue migrating portions of inference to internally designed ASICs, although aggregate AI infrastructure demand can still support NVDA unit growth.

Consensus is likely too quick to annualize a multiyear capacity shortage. The critical 1-3 month catalyst is supplier commentary on lead times, networking attach rates, and booked custom-silicon backlog; the 6-18 month risk is that power availability, permitting, or enterprise AI monetization becomes the binding constraint rather than compute supply. Falsification for the broad thesis would be sequential deceleration in AWS growth alongside rising capex intensity, or supplier guidance indicating that deployment schedules—not component availability—are slipping.

Key Decisions for Investors

• Initiate a 3-6 month long AVGO / short NVDA relative-value position, sized beta-neutral. The thesis is a mix shift toward custom accelerators and higher network complexity; exit if AVGO reports custom-silicon backlog or AI-networking growth below guidance, or if NVDA’s hyperscaler demand commentary reaccelerates materially.
• Maintain AMZN as a core long only through an AWS monetization checkpoint: add after evidence of sequential AWS revenue acceleration and stable/improving AWS operating margin. Avoid treating infrastructure spend alone as a catalyst; reduce if free-cash-flow conversion deteriorates for two quarters without corresponding cloud-growth upside.
• Do not chase MU on this signal. Set an alert for confirmed HBM qualification/share gains and continued contract-memory price increases; absent those data, use any capex-driven rally to favor AVGO exposure over MU because MU remains more exposed to a conventional memory-cycle correction.
• For downside protection on the AI-capex complex over the next 6-12 months, consider a small SMH put spread financed against profitable AVGO/MU gains. The principal tail risk is a deployment bottleneck from power, data-center construction, or weaker AI workload monetization rather than an immediate collapse in stated capex plans.