- Lead. Qualcomm has issued Amazon warrants to acquire 25 million shares at $161.26 apiece — roughly $4 billion — as part of a long-term partnership under which Amazon could purchase up to $60 billion of Qualcomm AI data-centre chips and optical networking technology.
- Fact. Qualcomm shares rose 4% on the announcement; the warrant vests in tranches tied to purchasing milestones, structurally aligning Qualcomm’s commercial incentives with Amazon’s willingness to commit at scale.
- Stake. For Amazon Web Services, the deal offers a meaningful diversification away from Nvidia; for Qualcomm, it validates a years-long attempt to enter the server-class AI chip market as its smartphone business faces the anticipated loss of its Apple modem account and persistently soft handset demand.
The companies announced the agreement on September 8, describing it as covering “multiple generations of customised silicon” designed for AWS infrastructure. The deal’s focus on inference workloads — running a trained model to generate outputs — distinguishes it from the large-scale training partnerships that have dominated AI chip deal-making. Inference is increasingly the dominant compute workload at hyperscaler scale, and it places a premium on power efficiency and cost-per-query metrics where Qualcomm contends its architecture has structural advantages over Nvidia’s GPU-centric design. CNBC reported the warrant structure links Amazon’s equity rights directly to purchasing milestones.
Qualcomm’s Server Pivot
Qualcomm’s move into data-centre AI silicon reflects a deliberate response to structural pressure on its core business lines. The company generates most of its earnings from smartphone chipsets and from royalties on its wireless technology patent portfolio; both face medium-term pressure. Handset volumes have been subdued, and the anticipated loss of Apple’s iPhone modem business — expected within the next product generation — removes a significant and reliable revenue stream that has historically supported Qualcomm’s margins.
The company has spent the past two years courting cloud providers with custom AI inference chips, positioning itself as a cost-effective alternative to the Nvidia processors that dominate AI infrastructure spending. The Amazon deal is the largest such arrangement Qualcomm has disclosed and provides the kind of committed demand signal that guides fabrication capacity planning at its manufacturing partners. The warrant structure also gives Amazon a financial stake in Qualcomm’s success in the server market, creating incentives for a deeper technology relationship over time.
AWS and the Nvidia Dependency
Amazon Web Services has been the most aggressive hyperscaler in developing proprietary silicon through its Trainium training chips and Inferentia inference processors, but AWS has also demonstrated consistent willingness to expand its supplier base as AI workloads grow faster than any single vendor can serve. The $60 billion ceiling on potential chip purchases from Qualcomm is large even by AWS capital-expenditure standards, reflecting growth expectations for inference demand through the end of the decade rather than an immediate commitment.
Marvell’s agreement granting Google a $12.2 billion stake option in a custom AI chip alliance established a template for warrant-linked silicon partnerships between hyperscalers and non-Nvidia chipmakers; Qualcomm’s arrangement follows a similar structure but at a larger disclosed commercial ceiling. Analysts expect the deal to cover Qualcomm’s Cloud AI 100 series processors and successor inference architectures, though Qualcomm has not publicly confirmed which specific product lines are included.