Amazon and Qualcomm have entered a long-term partnership to develop custom AI chips for data centres, in a deal that could see Amazon purchase up to $60 billion worth of Qualcomm chips and related products.
The deal gives Qualcomm a major foothold in the data-centre market as it looks to reduce its dependence on smartphones and challenge Nvidia in AI computing. For Amazon, it adds another chip supplier as cloud companies increasingly look for alternatives to Nvidia's dominant AI processors.
The companies will focus on AI inference chips, which process requests and generate responses from already-trained AI models. They will also work on optical connectivity technology capable of supporting speeds of up to 1.6 terabits per second, aimed at handling the growing data-transfer requirements of AI data centres.
The partnership comes as Qualcomm makes a major push into data-centre infrastructure. The chipmaker has set a target of generating $15 billion in data-centre chip revenue by 2029.
Qualcomm is looking to diversify as its traditional smartphone business faces pressure from weaker handset demand, rising component costs and Apple's move away from its modem technology.
“As AI demand accelerates, data center infrastructure will require advances in both computing and connectivity to deliver greater performance with more efficiency,” Qualcomm CEO Cristiano Amon said.
Amazon Looks Beyond Nvidia
Amazon is already one of Nvidia's largest customers and has developed its own AI chips through AWS. Its custom silicon business, including Trainium and Inferentia, has become an increasingly important part of its cloud strategy. AWS had an annualised revenue run rate of more than $25 billion at the end of June.
The Qualcomm partnership gives Amazon another option as AI infrastructure costs rise and demand for compute continues to surge. Microsoft and Meta are also working with Qualcomm on AI chips, giving the semiconductor company access to some of the world's biggest cloud and AI customers.
The focus on inference is particularly significant. As AI models move from training into widespread use, the amount of computing required to generate responses is growing rapidly. This is creating a new opportunity for chipmakers that can deliver inference at lower costs and with better energy efficiency.
Qualcomm will also increase its use of Amazon Web Services for chip-design workloads, with the aim of shortening development cycles.
The deal does not immediately challenge Nvidia's dominance. But it highlights a growing shift in the AI chip market: the next battle may increasingly be over making AI inference cheaper and more efficient, rather than simply building larger models and faster processors.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.