IN Brief:
- Qualcomm and Amazon will collaborate across multiple generations of customised silicon for AWS AI inference infrastructure.
- The companies are also developing optical connectivity extending to 1.6T using Qualcomm SerDes and optical DSP technology.
- Qualcomm plans to increase its use of AWS AI infrastructure, including Amazon Bedrock, for EDA workloads.
Qualcomm Technologies and Amazon are collaborating across multiple generations of customised silicon for AWS AI infrastructure, alongside optical-connectivity technology extending to 1.6T. The programme takes Qualcomm further into customer-specific data-centre silicon while linking the compute work to SerDes and optical DSP technology intended to handle the bandwidth demands surrounding large AI systems.
The companies say the silicon collaboration will focus on AI inference, but they have not disclosed the architecture, process node, packaging approach, memory system, tape-out date, or deployment volume for the first device. The announcement therefore establishes a multi-generation development programme rather than introducing a finished processor or accelerator.
That distinction is important as cloud operators increasingly use custom silicon alongside merchant processors. A device designed for one infrastructure operator can be shaped around a narrower workload mix, software environment, power budget, memory architecture, and deployment model than a component intended to serve a broad commercial market.
Qualcomm is entering the collaboration with a wider data-centre portfolio that already includes CPUs, AI accelerators, connectivity devices, and custom-silicon engineering. Its Dragonfly programme also covers advanced packaging and design-through-manufacturing services, giving the company an engineering base that extends beyond licensing processor IP or supplying an off-the-shelf accelerator.
Amazon brings its own substantial silicon-development capability through AWS, so the collaboration should not be read as a simple outsourcing arrangement. Custom semiconductor programmes can divide responsibility across system architecture, reusable IP, physical implementation, packaging, firmware, software, verification, and manufacturing, depending on which functions the customer wants to control directly.
Neither company has yet described that division of work. Until the first device specifications emerge, claims about performance, power efficiency, or cost relative to existing AWS silicon would be premature. What has been confirmed is that Qualcomm and Amazon intend to work together on customised inference silicon over several product generations.
The optical element is more specific. Qualcomm says the companies are developing connectivity solutions extending to 1.6T and future generations using its SerDes and optical DSP technologies. Those components sit in the data path between compute, memory, networking equipment, and other racks, where bandwidth and power consumption increasingly constrain the scale of AI infrastructure.
At those data rates, the interconnect becomes an electronics problem in its own right. SerDes circuitry has to recover and transmit high-speed signals with acceptable bit-error performance, while optical DSPs compensate for channel impairments and modulation effects without consuming enough power to undermine rack-level efficiency.
AI infrastructure is already pulling optical conversion closer to compute and switch silicon, changing decisions around photonics, packaging, thermal management, manufacturing, and test. Qualcomm’s AWS collaboration sits within the same movement, although the companies have not said whether their first 1.6T implementation will use pluggable optics, active optical cables, co-packaged optics, or another physical arrangement.
Qualcomm’s broader connectivity roadmap includes electrical and optical links for 800G and 1.6T infrastructure, covering SerDes, PAM4, coherent-lite DSP technology, and different reaches within and between data-centre facilities. The Amazon programme gives those capabilities a named hyperscale customer context rather than leaving them solely as parts of a supplier roadmap.
The collaboration also extends into Qualcomm’s own engineering workflow. Qualcomm plans to deepen its use of AWS AI infrastructure, including Amazon Bedrock, for electronic design automation workloads, with the stated aim of shortening chip-development cycles.
AI tools are increasingly being evaluated for verification support, design-space exploration, code generation, documentation, and engineering search, but they do not remove deterministic semiconductor sign-off. Functional verification, timing closure, physical verification, power integrity, and foundry rules still have to be satisfied before tape-out, irrespective of whether generative tools helped produce or analyse part of the design.
The immediate engineering significance of the agreement therefore lies in its breadth rather than in an undisclosed benchmark. Qualcomm is supplying custom-silicon capability, high-speed connectivity IP, and parts of its design methodology into a multi-generation AWS programme. The first useful comparison with competing AI infrastructure will come when the resulting silicon, interfaces, and deployment specifications are disclosed.