AWS Taps Qualcomm for Custom AI Chips as $496B Backlog Drives Diversification

Amazon Web Services has signed Qualcomm to develop custom AI data center silicon across multiple product generations, marking the first time AWS has publicly named a second external chip partner alongside incumbent supplier Broadcom. The deal positions Qualcomm to supply both compute accelerators and 1.6T optical networking components for AWS inference workloads.

The partnership reflects the scale of demand AWS is attempting to satisfy. The cloud division reported a $496 billion contract backlog in its most recent quarter, with revenue reaching $42.2 billion—up 36.7% year-over-year, the fastest growth in 18 quarters. Amazon is on track to double its power capacity by the end of 2027 compared to 2025 levels, and CEO Andy Jassy stated AWS could eventually become a trillion-dollar annual revenue business.

AWS now operates an AI revenue run rate exceeding $25 billion, with most capacity being contracted under five-year minimum terms. Capital expenditures hit $53.1 billion in the second quarter of fiscal 2026 alone, underscoring the infrastructure investment required to meet demand.

Why it matters

This deal signals AWS is willing to diversify its custom silicon supply chain even as it continues developing proprietary Trainium chips internally. For Qualcomm, the named Amazon partnership validates its pivot into data center markets and provides a second major hyperscaler customer after Meta. For Broadcom, it represents the first visible instance of a top-tier cloud provider dual-sourcing AI accelerators, potentially constraining the incumbent's addressable market as competitors gain production traction.

Qualcomm's Data Center Ambitions Take Shape

Qualcomm has now converted what it previously described as a "leading hyperscaler" engagement into a public Amazon partnership. The company is targeting $5 billion in data center revenue for fiscal 2027 and $15 billion by fiscal 2029, with total non-handset revenue projected to reach $40 billion in fiscal 2029. CFO Akash Palkhiwala confirmed the company has purchase orders in hand and began wafer production, with revenue starting in the December quarter.

The stock climbed 7.4% over the past week following the announcement. However, management cautioned that initial custom chip revenue will carry significantly lower gross margins than baseline products, reducing overall QCT margins by 1.5 to 2 percentage points during the ramp period. Handset revenue declined 20% year-over-year last quarter, making the data center thesis critical to offsetting core business headwinds.

Broadcom's Incumbent Position

Broadcom remains the dominant AI silicon supplier to hyperscalers, reporting $16.70 billion in AI semiconductor revenue last quarter—up 221% year-over-year—and guiding to $21.7 billion in the current quarter. The company expects AI revenue to reach $58 billion in fiscal 2026, $115 billion in fiscal 2027, and $230 billion in fiscal 2028.

CEO Hock Tan estimates Broadcom's content at $20 billion to $30 billion per gigawatt of AI infrastructure deployed. The customer roster includes Google TPUs, Anthropic's Ironwood deployment, OpenAI's Jalapeno accelerator, and Meta's MTIA program. However, management acknowledged that the "vast majority" of AI demand originates from a concentrated group of customers, making dual-sourcing arrangements like the AWS-Qualcomm deal a material risk to future growth assumptions.

The December quarter will provide the first revenue data on Qualcomm's hyperscaler shipments, establishing whether the company can execute against its fiscal 2029 targets or whether Broadcom's established pipelines continue to capture the bulk of cloud infrastructure spending.

Details of the AWS-Qualcomm partnership were first reported by 24/7 Wall St. and CNBC.