Qualcomm and AWS Ink $60 Billion Deal for Custom AI Silicon

Qualcomm and Amazon Web Services (AWS) have entered a multi-generational partnership to develop customized silicon for AI infrastructure. Announced on September 8, 2026, the deal signals a massive shift in how cloud giants source the hardware required to power the next decade of artificial intelligence.

Under the terms of the agreement, Amazon has committed to potential hardware purchases valued at up to $60 billion through 2036. The collaboration focuses specifically on the development of customized silicon designed for large-scale AI inference and high-speed optical connectivity. The partnership aims to push optical connectivity speeds up to 1.6 Tbps, ensuring that data movement between chips does not become a bottleneck as models grow in complexity.

The Shift Toward Custom ASICs

This partnership arrives as the world's largest cloud providers move to reduce their heavy reliance on general-purpose GPUs. To lower the operational costs of running massive AI models, providers are increasingly investing in Application-Specific Integrated Circuits (ASICs) tailored for specific workloads. While Amazon already develops its own proprietary chips, including Trainium and Inferentia, this new deal integrates Qualcomm's specialized expertise in power-efficient silicon directly into the AWS ecosystem.

Industry Implications

For Qualcomm, the deal represents a strategic pivot to diversify its revenue streams. Long dependent on the mobile handset market, the company is now aggressively expanding its footprint into the data center. By securing a long-term commitment from one of the world's largest cloud providers, Qualcomm establishes a stable, decade-long pipeline of high-value hardware sales.

For Amazon, the partnership addresses the primary cost driver of deploying Large Language Models (LLMs) at scale: inference. By optimizing the silicon specifically for the inference phase—where the model generates a response—AWS can potentially lower the cost per query and increase the efficiency of its AI services for millions of end users.

Looking Ahead

As the partnership rolls out, the industry will be watching to see how quickly these custom chips can be deployed across AWS data centers and whether they can meaningfully erode the market dominance of general-purpose AI accelerators. While the financial commitment is vast, the ultimate success of the venture will depend on the performance of the 1.6 Tbps optical interconnects and the power efficiency of the resulting silicon in real-world production environments.