Qualcomm secures its first Western hyperscaler client in AWS, forging a multi-year collaboration that spans AI inference hardware and high-speed optical networking, with a potential $60 billion deal to accelerate cloud AI infrastructure.
Amazon Web Services has become Qualcomm’s first Western hyperscaler customer for custom AI inference silicon, in a multi-generational collaboration that also covers high-speed optical networking for data centres. Qualcomm said the arrangement is designed to support AWS’s AI infrastructure at scale, while the two companies will jointly develop connectivity hardware that can reach 1.6 terabits per second for intra-cluster traffic. Qualcomm also said it will use AWS services, including Amazon Bedrock, to speed up its own chip design work.
The commercial terms are unusually explicit. Qualcomm disclosed in a securities filing that Amazon’s affiliate can buy as many as 25 million Qualcomm shares, but the full warrant only vests if Amazon spends as much as $60 billion on Qualcomm chips, networking gear and manufacturing services by September 2036. Qualcomm’s finance chief, Akash Palkhiwala, said at the Goldman Sachs Communacopia & Technology Conference that revenue from the deal will start in the company’s fiscal first quarter of 2027, with chips already in production.
Qualcomm’s pitch is that inference requires different hardware from training. The company uses LPDDR memory rather than the high-bandwidth memory found in Nvidia and AMD accelerators, arguing that inference workloads are often limited more by capacity than by peak bandwidth. Qualcomm says its AI200 card can support up to 768GB of LPDDR, and it is positioning its next-generation AI250 around near-memory computing to reduce data movement and power use. Palkhiwala said the firm’s “High Bandwidth Compute” approach is intended to suit decode-heavy inference tasks, with broader use expected later.
The optical networking element of the deal is tied to Qualcomm’s Alphawave Semi acquisition, completed late last year. Alphawave’s SerDes and interconnect portfolio includes PCIe Gen 6, CXL 3.0, Ethernet and chiplet connectivity technologies, which Qualcomm says can be adapted for the data paths that connect racks inside hyperscale facilities. That matters because AI deployment is increasingly constrained not only by the chip in the server, but also by the fabric linking thousands of servers together.
For Qualcomm, the AWS agreement is also a strategic hedge. The company remains heavily exposed to mobile handsets, and Apple’s move towards in-house modem silicon is set to reduce that dependence over time. Qualcomm has been trying to build a larger data centre business, and the AWS deal gives it a tangible Western hyperscaler customer rather than a prospect or pilot. Palkhiwala also said a second global hyperscaler is looking at a similar technology scope, suggesting the AWS relationship could become a repeatable model.
The deal does not mean AWS is abandoning its own silicon efforts. Its Trainium and Inferentia programmes remain separate internal accelerator lines, while Qualcomm’s chips are aimed at broader cloud infrastructure and external inference workloads. In market terms, the agreement strengthens the idea that hyperscalers will mix and match specialised silicon by workload, rather than rely on a single architecture for every AI task. Qualcomm shares rose sharply after the announcement, while investors in Amazon appeared to take a more cautious view.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
