Qualcomm and Amazon Agree to Develop Custom AI Chips for AWS

Qualcomm and Amazon agreed to develop custom AI inference silicon and high-speed optical connectivity for Amazon Web Services data centers under a multi-generation collaboration, Qualcomm said Tuesday.

The deal gives Qualcomm a hyperscale customer for its data-center technology. A performance-based warrant could also let an Amazon affiliate buy up to 25 million Qualcomm shares. A Qualcomm filing with the U.S. Securities and Exchange Commission sets the exercise price at $161.26 per share, or about $4.03 billion if the warrant is exercised in full. The warrant is an option tied to commercial milestones, not a completed equity purchase.

The companies plan to create customized silicon at scale for AI inference — the process of running trained models to generate outputs. Their connectivity work will extend to optical links of up to 1.6 terabits per second and later generations, using Qualcomm serializer-deserializer and optical digital signal processor technologies to move data between systems.

In Qualcomm’s announcement, AWS Vice President Prasad Kalyanaraman said the collaboration is intended to deliver faster, more efficient and more cost-effective infrastructure for customers. The companies did not disclose the financial terms of the engineering and supply relationship.

Qualcomm issued the warrant on September 3. It expires September 3, 2036. The filing says 3.75 million shares vested at issuance based on initial purchase commitments. The remaining tranches are tied to commercial arrangements, purchase orders and Amazon’s actual purchases of Qualcomm server-chip products, technology, systems and manufacturing services.

Those milestones cover up to $60 billion in payments, but that figure is a vesting ceiling, not a guaranteed Amazon purchase commitment or the stated value of the partnership.

The partnership adds Qualcomm technology to an AWS infrastructure portfolio that already includes Amazon-designed chips. Amazon’s custom-silicon overview lists Trainium for AI workloads and Graviton for general-purpose computing. AWS’s latest AI capital-spending outlook provides context on Amazon’s broader infrastructure buildout.

Separately, Qualcomm introduced its Dragonfly data-center roadmap in June. Neither the announcement nor the SEC filing specifies a first-delivery date, deployment schedule, manufacturing partner, process node, accelerator architecture, committed unit volume or exclusivity terms.

Qualcomm also said it plans to expand its use of AWS AI infrastructure, including Amazon Bedrock, for electronic design automation workloads aimed at shortening chip-design cycles.