Qualcomm and Amazon will collaborate across multiple generations on chips and network connectivity for AI data centers. The companies are focusing, among other things, on custom chips for AI inference and optical connections with speeds of up to 1.6 Tbit/s.
The partnership could ultimately reach enormous proportions. Documents filed by Qualcomm with the U.S. regulatory authority indicate that cumulative payments from Amazon over a ten-year period could total up to $60 billion, according to Neowin. That amount is not a guaranteed order but is tied to commercial agreements and milestones.
Part of the agreement is a warrant that allows Amazon to purchase up to 25 million shares of Qualcomm at $161.26 per share. The right to acquire those shares becomes available in stages as certain commercial milestones are reached.
These conditions are linked, among other things, to agreements, orders, and payments for Qualcomm’s server products, technology, systems, and manufacturing services. According to Neowin, 3.75 million shares have already been released based on Amazon’s initial purchase commitments. The warrant is valid through September 2036.
Custom chips for inference
Qualcomm and Amazon are discussing a partnership expected to span multiple generations of products. As part of this, the companies are developing custom silicon for large-scale AI data centers. A key area of application will be inference: the actual execution of previously trained AI models.
Details regarding the architecture or launch date of the first chips have not yet been disclosed. Qualcomm is also working on Dragonfly, a data center platform that includes the C1000 CPU and AI300 accelerator. With this, the company is seeking to gain a stronger foothold in the AI infrastructure market, where Nvidia currently dominates.
In addition to computing power, network connectivity will play a key role. Qualcomm will deploy its SerDes and optical DSP technology for connections within Amazon’s data center networks. The companies are working on optical solutions up to 1.6 Tbit/s and are also looking toward next-generation technologies.
AWS used for chip development
The collaboration also works in the opposite direction. Qualcomm aims to make greater use of AWS’s AI infrastructure for its own chip development. To that end, Amazon Bedrock, among other tools, will be used for Electronic Design Automation (EDA) workloads.
Qualcomm aims to shorten the development cycles of new chips through this approach, although the company has not provided specific figures on the expected time savings.
The agreement brings Qualcomm more firmly into the ecosystem surrounding AWS’s own AI infrastructure. Amazon has long been developing its own chips for its data centers. The collaboration with Qualcomm covers both custom chips for AI and the high-speed connections needed to link increasingly large AI systems together.