Amazon and Qualcomm have agreed to develop multiple generations of customised silicon for AI data centres, with the collaboration covering AI inference and high-speed optical connectivity.

Qualcomm announced the agreement recently, saying the companies will develop silicon for Amazon’s AI infrastructure. They are also working on optical connectivity extending to speeds of up to 1.6T, with future generations of the technology covered by the collaboration.

The arrangement adds Qualcomm to an AWS infrastructure portfolio that already includes Amazon-designed Trainium and Inferentia accelerators, alongside EC2 instances based on Nvidia GPUs. AWS continues to deploy both first-party and third-party processors across its AI infrastructure.

AWS broadens its AI infrastructure mix

Amazon has developed specialised processors for several parts of its cloud infrastructure. Graviton supports a broad range of cloud workloads, while Trainium and Inferentia are AWS-designed AI accelerators. AWS positions Trainium for training and inference, while Inferentia is designed specifically for inference.

Anthropic said in April that it was using more than one million Trainium2 chips to train and serve Claude through Project Rainier. Its expanded agreement with Amazon secures up to 5 GW of additional AWS capacity and spans Graviton and Trainium2 through Trainium4, with nearly 1 GW of Trainium2 and Trainium3 capacity expected online by the end of 2026.

The Qualcomm collaboration adds another source of specialised silicon to that mix. Neither company has disclosed whether the customised Qualcomm processors will be offered directly as AWS instances or used internally within Amazon’s data centres.

The Qualcomm agreement comes as AWS continues to expand EC2 capacity based on Nvidia’s Blackwell architecture while also deploying its own Trainium and Inferentia processors. AWS made P6-B300 instances available in its Jakarta region on September 3 after earlier deployments in regions including Oregon, Northern Virginia, Seoul, Hyderabad, and São Paulo. Each instance is based on eight Nvidia Blackwell Ultra B300 GPUs.

Amazon has not said that the Qualcomm collaboration is intended to reduce or replace its Nvidia deployments.

Qualcomm’s work with Amazon focuses partly on inference, an area where AWS already offers its Inferentia processors. Qualcomm’s broader data centre accelerator roadmap also centres on inference, with its Dragonfly portfolio spanning the AI200, AI250, and AI300 systems.

The Dragonfly portfolio also includes the C1000 CPU and separate connectivity products. Qualcomm said in June that its data centre plans cover CPUs, AI accelerators, customised silicon, and networking technology.

Qualcomm has not disclosed how the customised processors being developed with Amazon relate to its Dragonfly products. Amazon has also not said whether the Qualcomm silicon will complement Inferentia or serve a separate set of workloads.

The collaboration also extends to networking. Qualcomm said the companies are working on optical connectivity extending to 1.6T, using its SerDes and optical digital signal processor technologies for high-bandwidth links inside Amazon data centres.

Qualcomm’s wider data centre connectivity portfolio includes technology for 800G and 1.6T links across optical modules and other interconnects. The companies have not provided deployment dates for the customised processors or the 1.6T connectivity technology.

Qualcomm expanded its connectivity capabilities through its acquisition of Alphawave Semi, completed in December 2025. The deal, initially announced at an implied enterprise value of about $2.4 billion, brought Alphawave’s high-speed wired connectivity, custom silicon, and chiplet technologies into Qualcomm’s data centre business. Alphawave co-founder and former CEO Tony Pialis subsequently took responsibility for Qualcomm’s data centre operations.

A separate Qualcomm filing with the US Securities and Exchange Commission details the commercial terms of the arrangement. Qualcomm issued an Amazon affiliate a warrant to acquire up to 25 million Qualcomm shares at an exercise price of $161.26 per share.

The shares vest in stages based on commercial agreements, binding purchase orders, and purchases of Qualcomm server chips, technology, systems, and manufacturing services. Those purchases are tied to a maximum of $60 billion in payments over the term of the warrant.

Qualcomm said 3.75 million shares vested when the warrant was issued based on initial purchase commitments, while the warrant expires in September 2036. The filing does not state that Amazon has committed to spend $60 billion; the figure represents the maximum level of payments linked to the warrant’s vesting conditions.

The SEC filing covers server chip products, technology, systems, and manufacturing services purchased by Amazon during the warrant period.

Qualcomm builds out its data centre business

Qualcomm also plans to expand its use of AWS infrastructure for semiconductor development, including Amazon Bedrock for electronic design automation workloads. The companies did not disclose the value of Qualcomm’s AWS usage under the agreement.

The Amazon agreement follows a wider expansion by Qualcomm into server processors, AI accelerators, connectivity, and custom silicon. At its June 2026 investor day, the company set a target of more than $15 billion in data centre revenue by fiscal 2029.

Its Dragonfly roadmap combines the C1000 CPU with the AI200, AI250, and AI300 inference accelerators and high-speed connectivity products. Qualcomm has said the AI accelerator range will follow an annual product cadence.

The company is also developing custom silicon separately from its standard Dragonfly products. Qualcomm’s September announcement did not identify the architecture, manufacturing process, memory configuration, power envelope, or launch schedule for the processors being developed with Amazon.

Other hyperscale technology companies are pursuing their own specialised processors alongside third-party hardware. Microsoft runs its Maia accelerators alongside Nvidia and AMD hardware, while Google Cloud offers its own TPU processors alongside other accelerator options.

Meta has also continued developing its MTIA processors for internal workloads. Its MTIA 300, detailed in August, is designed for training ranking and recommendation models and integrates networking components alongside the accelerator.

(Photo by Brecht Corbeel)

See also: AWS and Microsoft launch multicloud link between AWS and Azure

Want to learn more about Cloud Computing from industry leaders? Check out Cyber Security & Cloud Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.

Cloud Computing News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.