Share with your CTO
Qualcomm is betting that edge AI, running inference directly on Snapdragon chips inside phones, PCs, and vehicles rather than in the cloud, is a viable long-term position against data center-focused rivals. The numbers say otherwise. Analysts project Qualcomm’s revenue growing at a 6% CAGR through fiscal 2028, against 61% for Nvidia and 62% for Broadcom over comparable windows. Despite that gap, Qualcomm trades at 26 times forward earnings while Nvidia and Broadcom trade at 15 and 22 times, respectively.
What this means for your business
The chip that runs your enterprise AI workload determines who captures the value your spending creates. If your AI roadmap leans on cloud inference through hyperscalers like AWS, Google Cloud, or Azure, your dollars are flowing directly into the Nvidia and Broadcom ecosystem. If your architecture is shifting toward on-device inference, whether for latency, privacy, or connectivity reasons, you’re in Qualcomm’s lane. The gap in growth projections is a signal about which ecosystem is attracting the next wave of capital and therefore the next wave of capability.
Qualcomm’s Cloud AI 100 accelerator is a genuine attempt to compete for inference workloads in the data center, and inference, not training, is where most enterprise budgets actually land. But the market Qualcomm is entering is one where Nvidia’s CUDA software ecosystem (the developer tooling and libraries that programmers already know) creates a switching cost that has nothing to do with raw silicon performance. Broadcom has a different angle, building custom chips tailored to specific hyperscaler architectures, which means its customers are locked in by design. Qualcomm is trying to be a general-purpose alternative in a market that is actively rewarding specialization.
The valuation inversion is the sharpest signal here. When the slower-growing company in a sector trades at a higher earnings multiple than the faster-growing one, the market is usually pricing in a risk that the headline growth story hasn’t absorbed yet. For CTOs evaluating AI infrastructure vendors over a three-year horizon, the question isn’t whether Qualcomm’s chips are technically capable. It’s whether a vendor whose core growth engine is consumer device manufacturers has the pull to stay current with the pace of model development that hyperscaler-funded silicon enables. I’d revise this view if Qualcomm lands a named hyperscale inference contract that demonstrates data center pricing power.
Concept deep-dive: Edge inference
Edge inference means running an AI model’s output-generating step locally on a device rather than sending data to a cloud server for processing, similar to how a calculator computes on the spot instead of phoning home for the answer. It exists because latency, privacy regulations, and unreliable connectivity create real limits for cloud-only approaches. The business connection is direct: edge inference shifts AI chip spending from hyperscaler data centers toward device manufacturers, which is a structurally smaller and more price-sensitive market.
Based on reporting from Qualcomm Doesn’t Look Cheap Next to Its AI Chip Peers. Here’s Why., originally published 2026-10-11 13:30:00.
