Table of Contents
Qualcomm CFO and COO Akash Palkhiwala described the Amazon transaction as “a landmark deal” for the company’s data center business
In sum – what to know:
Amazon validates strategy – Qualcomm says the multiyear agreement makes its $5 billion fiscal 2027 data center revenue target “very high confidence.”
AI compute disaggregates – Qualcomm sees growing demand for specialized solutions across training, prefill, and decode, with its HBC technology targeting memory bandwidth constraints in inference.
Four pillars expand – Custom silicon, AI accelerators, CPUs, and connectivity form Qualcomm’s broader data center strategy, supporting its $15 billion fiscal 2029 target.
Qualcomm sees its multiyear agreement with Amazon as a major step in building a data center business spanning custom silicon, AI accelerators, CPUs, and connectivity, with the company saying the deal strengthens its ability to meet previously announced revenue targets.
Speaking at the Goldman Sachs Communacopia + Technology Conference, Qualcomm CFO and COO Akash Palkhiwala described the Amazon transaction as “a landmark deal” for the company’s data center business. The agreement covers multiple generations of customized silicon as well as optical connectivity solutions starting with 1.6T and follow-on generations.
The transaction also includes a warrant agreement tied to purchases of up to $60 billion from Amazon for data center products over the next 10 years. Palkhiwala said about 15% of the warrants vest upfront and are associated with upfront commitments from Amazon.
The agreement is expected to contribute to Qualcomm’s previously announced data center targets. Palkhiwala said the company’s fiscal 2027 revenue target of approximately $5 billion is now “very high confidence,” while the Amazon agreement is also one of the core components supporting the company’s fiscal 2029 target of $15 billion.
Qualcomm expects to begin generating revenue from Amazon in the December quarter and is already in production with the customer. The company is also proceeding with another data center customer it previously discussed at its Investor Day.
Four-part data center strategy
Palkhiwala outlined four parts of Qualcomm’s data center strategy, beginning with custom silicon. Beyond Amazon, Qualcomm has disclosed an engagement with another global hyperscaler. The company is bringing together its compute and connectivity technologies, advanced-node chip expertise, high-performance and low-power capabilities, and high-yield manufacturing expertise for custom silicon.
The second area is AI accelerators, where Qualcomm sees the market moving away from a single solution for different AI workloads.
“What we’re seeing in the market is a disaggregation of compute, where there is certain kind of solutions, GPUs would be very good for certain things, XPUs would be very good for other things.”
As workloads are divided between training and inference, and inference is further divided into prefill and decode, Palkhiwala said hyperscalers increasingly want customized solutions for different components.
Qualcomm is initially targeting a memory-bandwidth constraint in decode workloads with its High-Bandwidth Compute technology, which stacks compute and memory together to deliver high bandwidth.
The third area is CPUs. Qualcomm is bringing its CPU technology from phones, PCs, automotive and industrial devices into data centers, targeting high performance at low power and low cost. Meta is its first disclosed customer for data center CPU silicon.
The fourth area is connectivity. Qualcomm’s acquisition of Alphawave brought SerDes technology that is integrated into these solutions, while Alphawave’s optical connectivity products are also part of the Amazon agreement.
Power and memory bandwidth shape AI infrastructure
Qualcomm expects demand for AI tokens to grow significantly as enterprises use AI across activities including software development, chip design, and customer support.
At the same time, Palkhiwala identified power, wafer availability, and memory availability as constraints on data center expansion. He argued that these constraints play to Qualcomm’s strengths, particularly its focus on performance per watt and its scale in wafer usage.
The company also sees AI computing becoming distributed between data centers and the edge rather than concentrated in one location. On the edge, Qualcomm is developing a compute-plus-memory stack as a coprocessor for devices including phones, cars, PCs, industrial devices, and robotics.
Qualcomm also expects its data center opportunity to extend beyond its fiscal 2029 target. Palkhiwala said the company’s longer-term strategy is not dependent on CPU products alone, but combines CPUs, AI accelerators, custom chips, and connectivity.
For Qualcomm, the Amazon agreement therefore represents both a customer commitment and an early validation of a broader data center strategy. Palkhiwala said the company expects its engagement to remain multi-generation, multi-product, and multi-customer rather than relying on one or two products or customers.
