Qualcomm CEO Cristiano Amon on AI smartphones, automotive agents, and acquiring Modular to open-source a CUDA alternative

Qualcomm CEO Cristiano Amon on AI smartphones, automotive agents, and acquiring Modular

Cristiano Amon is at Snapdragon Summit in Maui — more than a decade into holding the event in Hawaii — and his central message is that the AI smartphone inflection point Qualcomm has been building toward for years is finally arriving.

On-device AI

Amon's framing on edge versus cloud AI is deliberately simple. Every phone has 200-plus apps, and most of them don't work in airplane mode. AI will work the same way: some compute runs on the device, some runs in the cloud, and it will be transparent to the user. Qualcomm has been running multi-billion parameter models on device for years already.

Automotive

Automotive is Qualcomm's clearest current example of agentic AI at the edge. The company now serves every major car manufacturer, and Amon sees the in-car experience as a natural fit for agents precisely because there is no legacy OS or app ecosystem to navigate. He describes a use case that is already live: a driver asks an agent about the Yelp rating of a restaurant visible through the windshield and whether it has lunch availability — pulling together camera input, navigation data, and real-time web access in a single query.

“We made an acquisition of a company — the Modular team. The founder is Chris Lattner. He was the inventor of the Apple Swift programming language, the inventor of LLVM. He built a stack which is like CUDA, but it's designed to work on any hardware — NVIDIA, AMD, Qualcomm, or any hardware whatsoever. We bought that company and we're making that open source.”

Compute demand

Amon is unambiguous about the supply picture: the semiconductor supply chain is running at 100% capacity, and demand for compute exceeds availability across the board. His view on the CPU-versus-GPU question is that the data center is evolving into heterogeneous compute, with distinct engines for different tasks — CPUs handling orchestration and agent logic, and separate engines for inference pre-fill, decode, and decode-attention. CPU demand, he argues, will keep rising on both the data center side and the device side as agents proliferate.

New device categories

Amon says the majority of emerging consumer device categories — earbuds with cameras, jewelry, pins, glasses, rings, wristbands — are already running on Qualcomm silicon. He is most bullish on glasses, which he describes as "prime real estate" given their proximity to eyes, ears, and mouth. The "see what I see, read what I read, hear what I hear" agentic pattern, he argues, makes glasses a natural form factor for the next wave of personal AI.

Modular acquisition

The sharpest disclosure is Qualcomm's acquisition of Modular, founded by Chris Lattner, who created both Apple's Swift programming language and LLVM. Modular built an AI software stack that is hardware-agnostic — designed to run across CPUs, GPUs, and accelerators from any vendor, including Nvidia, AMD, and Qualcomm itself. Amon says Qualcomm is open-sourcing the stack. His stated rationale is explicitly industry-wide: a single proprietary stack concentrates AI innovation in one company, and an open alternative scales from data center to edge across competing hardware. No acquisition price was disclosed.

Amon's willingness to celebrate a stack that runs on Nvidia and AMD hardware is the tell. The Modular bet is less about locking developers into Qualcomm silicon and more about owning the layer that makes CUDA optional — which matters most to Qualcomm in the markets, automotive, edge, and mobile, where Nvidia's grip is weakest.