Qualcomm announced Tuesday that Amazon will hold rights to purchase up to $60 billion in AI data-center chips and related products under a multi-year supply agreement, paired with an option to acquire approximately $4 billion in Qualcomm equity. The partnership marks Qualcomm's largest non-mobile design win and Amazon's first public commitment to a non-Nvidia silicon architecture at hyperscale.
The structure is unusual. Amazon does not commit to $60 billion in purchases outright but secures the *right* to pull that volume over an unspecified term, likely five to seven years based on comparable hyperscaler agreements. The $4 billion equity option, exercisable in tranches tied to chip purchase milestones, functions as a de facto warrant package — Amazon pays market rates but gains board influence and pricing leverage as deployment scales. Qualcomm's stock rose 3.8% in after-hours trading; Amazon moved 1.1% on the news.
The move matters because it opens a second silicon lane for AWS at a moment when Nvidia's H200 and Blackwell lead times stretch past 18 months and pricing remains non-negotiable. Qualcomm's AI accelerators, built on its Oryon CPU architecture and Adreno GPU heritage, target inference workloads rather than training clusters — the segment where Amazon has the deepest moat with Bedrock and SageMaker deployments. If Qualcomm chips can deliver 65-75% of Nvidia's inference performance at 40-50% of the cost per FLOP, Amazon's gross margins on AI services improve materially without re-architecting its training stack. That calculation matters more than peak performance.
For Qualcomm, the deal provides revenue visibility that offsets handset-market compression. The company reported $35.8 billion in total revenue for fiscal 2024, with mobile chips contributing 68%. A $60 billion chip agreement, even if only 50% materializes over six years, represents $10 billion annually — roughly 28% of current revenue. More important is the signal: Microsoft, Meta, and Google now have proof that Amazon sees Qualcomm as production-grade. Expect procurement conversations at those three within 90 days.
The equity stake introduces execution risk. If Amazon exercises the full $4 billion option, it will own approximately 2.8% of Qualcomm at current valuations, enough for a board observer seat and veto rights on certain supply-chain decisions. That influence cuts both ways. Amazon gains pricing control and roadmap input; Qualcomm becomes structurally dependent on a customer that has historically built its own silicon when leverage tips too far. Amazon's Graviton chips replaced Intel and AMD in over 60% of EC2 compute instances within four years. Qualcomm must deliver without teaching Amazon how to replace it.
Watch three follow-on events. First, Qualcomm's October earnings call will clarify whether the $60 billion figure includes networking and storage components or chips alone. Second, Amazon's re:Invent conference in December will likely showcase the first Qualcomm-powered inference instances, with pricing that undercuts Nvidia by 30-40%. Third, any equity-option exercise will appear in Qualcomm's 10-Q filings within 60 days of the trigger event, signaling actual deployment velocity versus announced intent.
The partnership does not displace Nvidia in AWS training clusters. It creates a cost-efficient inference tier that improves Amazon's unit economics on the $26 billion in annualized AI revenue it now runs through AWS. Qualcomm's revenue diversification arrives exactly when its mobile business no longer compounds.