Qualcomm and Amazon are expanding their semiconductor partnership into large-scale AI data centers, with a multi-generation collaboration focused on custom silicon for AI inference and optical connectivity reaching speeds of up to 1.6 terabits per second. The agreement gives Qualcomm a larger role in the rapidly expanding AI infrastructure market while strengthening Amazon Web Services’ strategy of combining internally developed and customized processors with high-bandwidth networking to control the cost and energy demands of AI workloads.
The next phase of AI infrastructure competition is increasingly moving beyond the question of who can build the fastest accelerator.
Cloud providers now need to optimize the entire data-center stack—from processors and memory to networking, optical interconnects, software and power consumption. Qualcomm’s expanded collaboration with Amazon reflects that shift, bringing the chipmaker into a deeper relationship with AWS around both AI inference silicon and connectivity.
Under the agreement, Qualcomm Technologies and Amazon will work across multiple generations of customized silicon for large-scale AI data centers, with an initial emphasis on inference. The companies will also develop optical connectivity solutions capable of supporting up to 1.6T bandwidth and future generations.
The distinction between training and inference is increasingly important.
Training frontier AI models requires enormous amounts of compute for comparatively concentrated workloads. Inference is different: once models are deployed, every user request, recommendation, search result, agent interaction or enterprise workflow can generate additional compute demand. As AI applications become more widely used, inference can therefore become a persistent infrastructure cost.
Gartner expects global spending on AI-optimized infrastructure-as-a-service to reach $42.3 billion in 2026, representing 96.4% growth from 2025. More notably, Gartner forecasts global spending on AI inference infrastructure at $23.3 billion this year, compared with $19 billion for training.
That economics is helping drive interest in specialized silicon.
AWS already operates its own accelerator portfolio, including Trainium for AI training and Inferentia for inference. Amazon has said its custom silicon business has become one of the world’s three largest data-center chip businesses, while reporting strong demand for its Trainium products.
Qualcomm’s role adds another dimension. The company has spent decades optimizing processors for power efficiency and heterogeneous system integration in smartphones, PCs, automotive systems and edge devices. Bringing that expertise into data centers could help AWS pursue customized architectures where performance per watt and cost per inference matter alongside absolute compute performance.
The collaboration also extends into networking.
As AI clusters grow, accelerators need to exchange enormous amounts of data. Faster processors alone cannot solve bottlenecks if network links and interconnects cannot move data quickly enough between compute, memory and storage resources.
Qualcomm and Amazon therefore plan to work on optical connectivity using Qualcomm’s SerDes and optical digital signal processing technologies. The first target is connectivity of up to 1.6 terabits per second, with future generations planned beyond that.
This is becoming an increasingly important layer of AI infrastructure. High-speed optical links can help connect large numbers of accelerators while addressing bandwidth, latency and power constraints that become more difficult as clusters scale.
The partnership also has an unusual feedback loop: Qualcomm plans to increase its use of AWS AI infrastructure for electronic design automation (EDA), including Amazon Bedrock, with the goal of reducing chip-design cycles.
In other words, AWS infrastructure will help Qualcomm design future semiconductor products that can potentially become part of AWS infrastructure itself.
That vertical integration is becoming a defining characteristic of hyperscaler AI strategies.
Amazon is competing with Microsoft, Google and Meta not only through cloud services but also through increasingly customized infrastructure. Microsoft has developed Maia accelerators, Google has its TPU family, and Meta has been developing its own AI silicon while continuing to deploy large volumes of commercial accelerators.
At the accelerator level, NVIDIA remains the dominant provider for AI computing, particularly for training and high-performance workloads. Qualcomm’s agreement with Amazon is therefore unlikely to displace NVIDIA overnight. Instead, it illustrates how hyperscalers are expanding the number of processor architectures available for specific workloads and attempting to reduce dependence on a single supplier.
Qualcomm’s expansion into data-center AI also represents a strategic shift for the chipmaker.
The company has historically been strongly associated with mobile processors and wireless technologies. It is now building a broader AI infrastructure business around data-center compute, networking and custom silicon. Reuters reported that the Amazon agreement could support Qualcomm’s broader goal of reaching $15 billion in data-center revenue by fiscal 2029.
For Amazon, the benefit is greater silicon and systems optionality. For Qualcomm, it is access to one of the world’s largest cloud infrastructure operators and a pathway into a market where AI-related semiconductor demand continues to grow.
The timing is significant. Gartner forecasts worldwide AI spending will reach $2.59 trillion in 2026, with AI infrastructure—including optimized servers, networking and AI processing semiconductors—accounting for more than 45% of overall spending.
The competitive question is now shifting from whether AI data centers need more compute to how efficiently that compute can be delivered.
Qualcomm and Amazon are betting that customized inference silicon and faster optical connectivity can become part of that answer.
Market Landscape
AI infrastructure is entering a phase of system-level optimization. Hyperscalers are increasingly developing custom accelerators, networking technologies and software stacks to improve performance per dollar and performance per watt.
AWS already uses Trainium and Inferentia alongside NVIDIA hardware, while Google, Microsoft and Meta are developing their own accelerator strategies. Amazon’s custom silicon business has also attracted major AI customers, including Anthropic and others using Trainium for model development and deployment.
Qualcomm adds a different strength: decades of experience optimizing compute under tight power and thermal constraints. Its collaboration with Amazon suggests those capabilities are being adapted for data-center-scale inference and high-speed connectivity.
The networking component may prove equally important. As AI clusters become larger and more distributed, optical interconnects and advanced SerDes technology increasingly become part of the AI accelerator equation rather than separate networking concerns.
Top Insights
- Qualcomm and Amazon will develop customized AI data-center silicon across multiple generations, initially targeting increasingly important inference workloads.
- The companies are also pursuing optical connectivity reaching 1.6T, addressing bandwidth requirements created by increasingly large AI clusters.
- Gartner expects inference spending to exceed training spending globally in 2026 as AI applications move into continuous production workloads.
- Qualcomm is using AWS AI infrastructure and Amazon Bedrock for EDA workloads, potentially shortening future semiconductor development cycles.
- The partnership highlights hyperscalers’ growing preference for customized silicon as they seek better AI performance, efficiency and cost control.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
