Qualcomm announced the acquisition of Modular, a developer tools company, and set guidance for $15 billion in data center revenue by 2029. The chipmaker had no meaningful data center sales as recently as 2021. The Modular deal price was not disclosed. Qualcomm's inference-focused silicon strategy now extends from edge devices into cloud workloads, targeting the post-training AI market where Nvidia currently holds 90% share.
The $15 billion target represents roughly 15% of Qualcomm's total revenue if the company maintains its current growth trajectory. Modular's technology sits between AI models and hardware—software that compiles frameworks like PyTorch and TensorFlow into chip-specific instructions. Qualcomm has shipped custom AI processors to Meta for inference workloads and secured design wins with unnamed hyperscalers. The company disclosed in February that its AI inference accelerators already run production workloads at three cloud providers. Modular co-founders Chris Lattner and Tim Davis built compiler infrastructure at Apple and Google before founding the company in 2022. Modular raised $100 million at an undisclosed valuation in August 2023.
The revenue target matters because it assumes Qualcomm can win inference share while training budgets compress. Hyperscalers spent $200 billion on capex in 2024, with 65% allocated to Nvidia GPUs optimized for training. Inference—running trained models at scale—requires different economics. Qualcomm's chips consume 40% less power per inference operation than comparable Nvidia products, according to internal benchmarks the company published in January. If energy costs $0.10 per kilowatt-hour and a hyperscaler runs 10 million inferences daily, the annual savings approach $1.5 million per rack. At cloud scale, that creates procurement optionality. The Modular acquisition removes a software integration barrier. Without native tooling, AI teams must hand-optimize models for non-Nvidia silicon—work that takes months and dedicated engineering.
Operators should watch two deadlines. First, Qualcomm's April earnings call, where management typically discloses design win progress and hyperscale customer counts. The company has historically announced cloud partnerships 60-90 days before volume shipments begin. Second, Modular's integration timeline. If Qualcomm ships a unified hardware-software stack by late 2025, it compresses evaluation cycles for enterprise AI buyers who currently default to Nvidia's CUDA ecosystem. The $15 billion guidance assumes Qualcomm captures 8-12% of the inference accelerator market by 2029, based on analyst estimates of total addressable spend. That requires at least four hyperscale customers running production inference workloads at meaningful scale. Qualcomm currently names two publicly: Meta and an unnamed Tier 1 cloud provider.
The company's investor deck, updated March 2025, shows data center revenue growing from $400 million in fiscal 2024 to the $15 billion target in five years. Modular gives Qualcomm the compiler layer it lacked. The question is timing—whether inference budgets grow fast enough to support a new supplier before training spending recovers.