Qualcomm Sold Server Chips Once Before and Quit Within a Year. History Says What Has to Be Different This Time.

Qualcomm signed an agreement under which Amazon could purchase up to $60 billion of server chips, systems, and services through 2036, with 25 million Qualcomm warrant shares vesting based on binding purchase orders and actual spending. The deal gives Qualcomm an anchor customer for its renewed data-center push after its Centriq server-chip effort was dismantled in 2018; 3.75 million warrant shares vested immediately on initial commitments. Management targets more than $15 billion of data-center revenue by fiscal 2029 and expects non-handset revenue growth to exceed 60% in fiscal 2027, but execution risk remains as the $60 billion figure is a vesting cap rather than guaranteed revenue. Qualcomm's fiscal Q3 2026 handset revenue fell 20% year over year to $5.1 billion, while total revenue declined 4% to $9.9 billion.

Analysis

The key underwriting issue is not addressable market but whether QCOM can convert a hyperscaler-specific design into repeatable, high-margin platform revenue. AWS has strong incentives to dual-source inference capacity and retain architectural control through its Graviton/Trainium/Inferentia stack; that makes QCOM a potentially valuable capacity and cost lever, but also limits pricing power. The warrant package is effectively customer-acquisition consideration: up to roughly 2%-3% potential dilution and a below-market strike mean investors should assess incremental gross profit net of dilution, not gross qualified spend.

Near term, the announcement supports a multiple re-rating because it reduces the probability that QCOM's data-center effort is another unfunded experiment. Over the next 1-3 months, the critical evidence is disclosed purchase-order timing, manufacturing commitments, and whether management quantifies fiscal-2027 revenue and gross-margin contribution; absent those, the market will likely treat the arrangement as a framework rather than backlog. The most important 6-18 month risk is that custom inference silicon becomes a lower-margin ASIC/services business, with AWS capturing most of the savings while NVDA retains premium workloads and QCOM's handset weakness leaves less room for execution misses.

Consensus may be underweight the negative read-through for INTC: another credible Arm-based hyperscaler supply path further weakens the probability of an x86-only server refresh. Conversely, the move is not yet a material threat to NVDA, whose moat is software, networking and training-scale deployment rather than commodity inference alone. A QCOM rerating is justified only if data-center revenue becomes visible before handset declines require it; a fiscal-2027 guide below the implied non-handset acceleration, or no order/revenue disclosure by the next two earnings reports, would falsify the thesis.

Key Decisions for Investors

• Maintain a QCOM watch-list long rather than chase the initial reaction; initiate only after the next one to two earnings reports show named data-center revenue or firm fiscal-2027 contribution guidance. Target a 10%-15% upside from multiple expansion if execution is validated; exit on a material handset-guide cut without offsetting data-center bookings.
• Express the competitive implication as a 6-12 month pair: long QCOM / short INTC in equal beta-adjusted dollars. The thesis is incremental Arm/custom-silicon adoption pressuring INTC server-unit and gross-margin expectations; cover if INTC demonstrates sustained data-center share stabilization or QCOM fails to disclose commercial ramp milestones.
• Do not short NVDA on this development. Reassess only if AWS reports broad displacement of GPU inference across external customer workloads, not merely internal customized deployments; until then, NVDA's software and full-stack economics remain largely insulated.
• Monitor QCOM's reported gross margin, warrant-share dilution, and capex/supply commitments alongside revenue. A revenue ramp that requires disproportionate incentive expense or dilutes consolidated gross margin would make the apparent data-center win value-destructive despite headline spend figures.