Qualcomm and Amazon Expand AI Data-Center Partnership Into Custom Silicon and 1.6T Networking

Qualcomm Technologies and Amazon are expanding their relationship into a multi-generation customized-silicon effort for AWS artificial-intelligence infrastructure, with the companies also working on optical connectivity that extends to 1.6T and future generations. Qualcomm said the silicon work is focused on AI inference, moving the partnership beyond the companies’ earlier use of Qualcomm accelerators inside Amazon EC2.

The September 8 announcement gives Qualcomm a potentially significant position inside one of the world’s largest cloud platforms at the same time the chipmaker is pushing deeper into data centers. Qualcomm said the companies will collaborate on customized silicon at scale, high-performance optical links using Qualcomm SerDes and optical DSP technologies, and the use of AWS AI infrastructure for Qualcomm’s own chip-design workflows.

A regulatory filing adds financial detail that was not spelled out in the product announcement. Qualcomm disclosed in a Form 8-K filed September 8 that it issued an Amazon affiliate a warrant to acquire as many as 25 million Qualcomm common shares at an exercise price of $161.26 per share. The warrant is tied to a strategic collaboration involving purchases of Qualcomm Technologies server-chip products, technology, systems and manufacturing services by Amazon.

The SEC filing links the warrant to purchasing milestones

The warrant was issued September 3 to Amazon.com NV Investment Holdings LLC, according to Qualcomm’s filing. It permits cashless exercise and expires September 3, 2036. Vesting occurs in tranches tied to commercial arrangements, binding purchase orders and actual purchases by Amazon during the warrant term.

Qualcomm said the vesting schedule can extend through a maximum of $60 billion in payments for covered products and services. That figure should not be read as a $60 billion purchase commitment. The filing describes it as the upper end of the payment milestones tied to warrant vesting, and it does not say Amazon has committed to spend that amount.

The initial relationship is already meaningful enough to vest part of the equity incentive. Qualcomm said 3.75 million warrant shares vested when the warrant was issued based on initial purchase commitments. The remaining shares are subject to later milestones. Until exercised, the warrant does not give the Amazon affiliate voting rights or other rights of a Qualcomm shareholder, and Qualcomm said it expects to register the resale of warrant shares after the filing date.

The equity structure gives both sides a long runway. Amazon can gain additional economic exposure to Qualcomm if the commercial relationship grows and later milestones are met, while Qualcomm has a direct incentive to win more AWS-related purchases over the warrant’s ten-year life. The filing does not identify the specific server products that will satisfy each milestone or disclose a deployment timetable for the customized silicon announced September 8.

Qualcomm is broadening its data-center role beyond earlier AWS accelerators

Amazon and Qualcomm already had a cloud-hardware relationship. In 2023, AWS made EC2 DL2q instances generally available using Qualcomm Cloud AI 100 accelerators for AI inference. Those instances brought Qualcomm’s AI accelerator technology into Amazon’s cloud and were also positioned for developing and validating workloads that could later run on Qualcomm-powered devices.

The new collaboration goes further because Qualcomm is now describing multi-generation customized silicon for large-scale AWS AI infrastructure rather than a single EC2 instance family using an existing accelerator. The company has not named the first custom chip, said whether it will carry a Dragonfly brand, or provided a launch date. That distinction matters because Qualcomm unveiled a much broader data-center roadmap in June that includes Dragonfly CPUs and AI accelerators as well as separate custom-silicon offerings.

At its June investor event, Qualcomm introduced the Dragonfly C1000 data-center CPU, its High Bandwidth Compute technology, the Dragonfly AI300 inference accelerator and a connectivity portfolio. The AI300 joined the AI200 and AI250 in what Qualcomm described as an annual, multi-generation accelerator roadmap. Qualcomm also presented custom silicon as its own category, designed around customer-specific data-center requirements.

That broader portfolio helps explain why the Amazon relationship can span several layers without implying that AWS has selected every Qualcomm data-center product. The September announcement specifically confirms customized silicon for AI inference and advanced connectivity. It does not say Amazon will deploy the C1000, AI200, AI250 or AI300, so those products provide strategic context rather than confirmed AWS design wins.

1.6T optical connectivity and chip-design workloads widen the scope

The networking piece extends the relationship beyond compute. Qualcomm said the companies will work on high-performance optical connectivity for Amazon data-center networks using Qualcomm’s SerDes and optical digital-signal-processing technology. Its current data-center portfolio describes connectivity designed for 800G and 1.6T performance across optical modules, active optical cables and active electrical cables.

Moving data between accelerators, servers and racks has become a major constraint as AI systems scale, which makes networking performance part of the economics of an AI data center rather than a secondary component. Qualcomm’s announcement says the Amazon work will extend up to 1.6T and later generations, but it does not specify an exact networking product, the number of links Amazon plans to deploy or when the first 1.6T systems will enter service.

The collaboration also runs in the opposite direction. Qualcomm plans to deepen its own use of AWS AI infrastructure, including Amazon Bedrock, for electronic design automation workloads. The company said it is targeting shorter chip-design cycles. That gives Amazon a role not only as a buyer of Qualcomm technology but also as a cloud provider supporting part of Qualcomm’s development process.

For AWS, the agreement sits alongside a long-running in-house silicon strategy that includes Graviton processors and the Trainium and Inferentia AI chips. The Qualcomm announcement does not indicate that those AWS-designed products are being displaced. Instead, it adds a customized-silicon and connectivity relationship with an outside semiconductor supplier to infrastructure that already mixes Amazon-designed hardware with products from multiple chipmakers.

For Qualcomm, the arrangement is another step in its effort to build a larger business in data-center computing after decades in mobile and edge devices. The warrant provides a measurable commercial framework, but the product announcement leaves several details open: the first customized chip has not been named, the 1.6T deployment schedule has not been disclosed, and the filing does not establish that Amazon will reach the maximum $60 billion payment threshold. Future product disclosures and purchase milestones will determine how much of the potential relationship becomes deployed infrastructure.