Qualcomm Takes On Nvidia With Its First AI Data Center Chip

Qualcomm — the company whose Snapdragon chips power the majority of Android smartphones — just crossed into territory it has never occupied before: the data center. The company unveiled the Dragonfly C1000, its first processor built specifically for the server racks that run AI agents. Meta has already agreed to deploy it starting in 2028, and investors responded by pushing Qualcomm's stock up 15% after hours.

The scale of Qualcomm's ambition here is significant. The company is revising its long-term revenue targets dramatically: non-smartphone business is now projected to hit $40 billion by 2029 — nearly double the previous forecast — with data centers alone expected to contribute $15 billion of that total. This isn't a tentative experiment in a new market. It's a direct move to build a substantial second business from the ground up.

The chip is built for how AI actually runs today

The Dragonfly C1000 is designed specifically around AI agent workloads — the kind of continuous, interactive inference that's become the dominant pattern as AI assistants, coding agents, and autonomous systems have gone from experiments to production infrastructure. That's a meaningful distinction. General-purpose AI training chips (Nvidia's core business) are optimized for a different task than the chips that need to run a model efficiently at scale, responding to millions of requests per day.

Alongside the hardware announcement, Qualcomm is acquiring Modular for approximately $4 billion. Modular builds software that lets AI applications run across different chip architectures without needing to be rewritten for each one. Combined with the Dragonfly, this gives Qualcomm both sides of what a data center customer needs: the chip itself, and a software layer that makes switching costs lower. That matters for the competitive picture against Nvidia, which benefits from deep lock-in through its CUDA software ecosystem.

Why this matters for the cost of AI

Nvidia currently holds dominant market share in AI server chips — a position that comes with significant pricing power. The chips that run the AI models behind ChatGPT, Claude, Gemini, and every other major service run almost entirely on Nvidia hardware. That near-monopoly is one reason AI API costs, while falling, remain expensive relative to underlying compute.

Qualcomm entering the data center market — with a chip a company the size of Meta is willing to stake a deployment on — is a real competitive signal. AMD has been chipping away at Nvidia's position; now Qualcomm brings another credible alternative. More competition in AI silicon historically translates to lower prices further down the stack, which eventually reaches the developers and builders who pay per token or per API call. The timeline isn't immediate — Meta's deployment doesn't start until 2028 — but the direction is set.