Amazon secured a $60 billion deal with Qualcomm to expand its AI inference hardware portfolio.

Amazon and Qualcomm will develop custom silicon and high-speed optical connectivity for AI inference, potentially providing AWS users and data center operators with new hardware options as power and infrastructure costs continue to rise. Revealed by Qualcomm on Sept. 8, the partnership opens another hardware pathway for AI inference as energy constraints push data center architects toward more specialized infrastructure. Both companies are targeting processors and connectivity specifically engineered to address these limitations.

The collaboration spans multiple generations of customized AI inference silicon and optical connectivity designs, with speeds reaching up to 1.6 terabits per second for large-scale data centers. Stephen Sopko, practice lead for semiconductor and deep tech at HyperFrame Research, emphasized that focusing on inference is critical because its requirements diverge significantly from AI training. “Power is the constraint that decides which architectures get built,” Sopko said in an email. While training workloads prioritize peak floating-point performance and high-bandwidth memory, inference places greater emphasis on tokens per watt, tokens per rack, and available power headroom. For cloud buyers evaluating AWS inference services, the agreement expands available hardware options. Similarly, data center operators considering in-house AI infrastructure now have another processor and connectivity supplier to consider, though neither company disclosed specific product roadmaps or production timelines.

Commercial commitments under the agreement extend through September 2036. According to a Qualcomm SEC filing, Qualcomm issued Amazon.com NV Investment Holdings LLC a warrant to acquire up to 25 million shares at $161.26 per share. This warrant is tied to commercial arrangements, binding purchase orders, and actual purchases of Qualcomm server chip products, technology, systems, and manufacturing services, with total payments potentially reaching $60 billion over the warrant’s term. Upon issuance, 3.75 million shares vested based on initial purchase commitments. Matt Kimball, vice president and principal analyst for Data Center Technologies at Moor Insights & Strategy, noted that the multi-generational framework distinguishes this partnership. “It's definitely significant … because of the multiple-generational aspect of the partnership,” Kimball said in an email. “Hyperscalers don't typically commit to multiple generations for commodity parts.” He added that the structure indicates Amazon is planning around a long-term hardware architecture rather than evaluating isolated components.

While Amazon continues to develop its own data center silicon through its Trainium AI accelerators and Graviton server CPUs, Qualcomm is supplying customized silicon for a distinct segment of the AI infrastructure stack. The partnership also encompasses multi-generational development of high-bandwidth network connectivity. “Connectivity is not an accessory to the accelerator story; it is half the system,” Sopko said. Qualcomm will leverage its SerDes and optical digital signal processor technologies, bolstered by high-speed connectivity capabilities gained through its acquisition of Alphawave Semi. Kimball cautioned that AWS already utilizes multiple connectivity suppliers, including Marvell and Broadcom, so the networking collaboration should not be interpreted as a vendor replacement. Instead, he highlighted the strategic significance of Qualcomm combining custom compute and connectivity capabilities within a single ecosystem.

Beyond hardware, Qualcomm is expanding into software and AI infrastructure, planning to deepen its reliance on AWS services—including Amazon Bedrock—for electronic design automation workloads to accelerate chip design cycles. The company’s current data center portfolio features the Dragonfly C1000 server CPU, AI200, AI250, and AI300 inference accelerators, high-bandwidth compute technology, and 800G and 1.6T connectivity. In June, Qualcomm announced a separate multi-generational data center CPU agreement with Meta, which plans to deploy the C1000 in its next-generation server fleet with production slated for the second half of 2028. Following its July acquisition of Modular, which adds software and AI infrastructure capabilities, Qualcomm’s broader data center strategy encompasses AI accelerators, server CPUs, high-bandwidth compute technology, and custom silicon. The company has targeted more than $15 billion in data center revenue by fiscal 2029.