Qualcomm enters datacenter AI chip market to compete with NVIDIA

Qualcomm announced a full line of AI datacenter chips — including the Dragonfly C1000 CPU — and signed multi-generation deals with Microsoft and Meta. This is the company's most ambitious bet outside the smartphone market, and the most direct competitor entry to NVIDIA from a tier-one player in years.

Qualcomm's chips compete with NVIDIA's H100/B200 GPUs in the inference segment — where already-trained models respond to production requests. It is not the training market (where NVIDIA still dominates by a large margin), but it is where most operational AI costs sit, and where price elasticity is highest.

Why Qualcomm may succeed where others failed

Intel, AMD and dozens of startups tried to carve out market share from NVIDIA. Most failed by lacking a customer base or by arriving too late to software (NVIDIA's CUDA has an 18-year head start). Qualcomm has a different advantage: decades of mobile miniaturization created unique capabilities in energy efficiency per AI operation — which is critical in datacenters where electricity is the dominant variable cost.

The Microsoft and Meta deals are the most important part of the announcement. They do not need to be the fastest chip — they need to be fast enough at a lower cost per operation. With two of the world's largest AI infrastructure buyers signing multi-generation contracts, Qualcomm has what chip startups lacked: guaranteed demand to amortize R&D.

What changes in the market

NVIDIA will not lose dominance in training. But in at-scale inference, competition changes the calculation for all infrastructure buyers. With more hardware options, GPU pricing tends to fall — or at least not grow at the current pace.

For anyone building AI systems in production, hardware diversification is good news. Dependence on a single accelerator vendor (NVIDIA) creates supply risk and price risk. Qualcomm and AMD as real alternatives change that equation.

For 10Dobro

Our 26 production applications run primarily on cloud (inference via API). We are not direct GPU buyers. But what matters: when Azure, AWS and GCP have more hardware options, the cost of API tokens — which directly affects our margin — tends to fall in the medium term.