Qualcomm announced the acquisition of Modular and simultaneous guidance calling for $15 billion in annual data-center revenue by 2029. The company reported zero data-center sales as recently as 2021. The $15 billion target represents roughly 40% of Qualcomm's fiscal 2024 total revenue of $38.9 billion, effectively requiring the construction of a second business equal in scale to its current automotive or IoT divisions within five years.
Modular built a compiler and runtime framework designed to abstract AI workloads from underlying silicon, allowing models to run across heterogeneous chip architectures without vendor lock-in. Qualcomm did not disclose purchase price or retention terms. The acquisition follows 18 months of public commentary by CEO Cristiano Amon positioning the company's Snapdragon chips as inference accelerators capable of displacing Nvidia in edge and hybrid-cloud deployments. Qualcomm has shipped AI-capable SoCs into 2.2 billion mobile devices since 2020 but holds negligible share in the $67 billion data-center accelerator market that Nvidia commands at 92% share.
The $15 billion by 2029 guidance assumes compound annual growth exceeding 110% from an effective 2024 baseline near zero. For context, Nvidia's data-center revenue grew from $10.6 billion in fiscal 2023 to $47.5 billion in fiscal 2024, a 348% increase driven by GPU scarcity and H100 allocation queues extending into mid-2025. Qualcomm's path requires capturing 5-7% of a projected $220 billion total addressable market for AI silicon by decade-end, according to analyst consensus compiled in January 2025. That share must come from greenfield workloads or displacement, neither of which Qualcomm has demonstrated at data-center scale.
Modular's abstraction layer solves a specific problem: enterprises running inference at edge or in hybrid environments need to avoid rewriting codebases when switching from Nvidia CUDA to alternative substrates. Qualcomm's Snapdragon X Elite chips, shipping in volume since Q4 2024, already run Llama and Stable Diffusion models locally on Windows laptops at inference speeds competitive with cloud API calls for sub-13-billion parameter models. The strategic question is whether on-device and private-cloud inference expands quickly enough to justify $15 billion in chip sales, or whether centralized training and serving by hyperscalers keeps the majority of AI revenue inside Nvidia's installed base through 2029.
Allocators should monitor three data points over the next 18 months: Qualcomm's quarterly data-center revenue emergence from the "other" line item into standalone disclosure, expected by Q3 fiscal 2025; design-win announcements with named hyperscalers or sovereign cloud providers, particularly in the Middle East and Asia where non-Nvidia procurement is policy-driven; and Modular's customer retention rate post-acquisition, since its value proposition depends on remaining architecture-neutral even under Qualcomm ownership. Nvidia's next-generation Blackwell GPU family begins volume shipments in Q2 2025, compressing the window in which alternative architectures can claim performance parity before the benchmark resets.
Qualcomm has not yet shipped a data-center chip at commercial volume. The $15 billion figure sits in guidance documents, not purchase orders.