Qualcomm Bets on Data Centers: An AI Breakout Amid Declining Smartphone Sales

Qualcomm is undergoing a critical strategic pivot. According to its latest earnings call, smartphone business sales fell about 20% year-over-year compared with 2025, and the company—long centered on mobile chips—is now positioning data centers and AI infrastructure as its next growth engine.

Smartphone Business Under Pressure, Transformation Imminent

The weakness in the smartphone market is not a sudden development, but a 20% year-over-year decline is enough to make Qualcomm reconsider its revenue structure. For a long time, Qualcomm's revenue has been highly dependent on smartphone chip shipments and patent licensing, a model that looks fragile during a market downturn. On the earnings call, management did not shy away from this pressure, but explicitly positioned the data center business as the "second growth curve."

The logic behind this shift is not hard to understand: compute demand driven by AI training and inference continues to expand, and the growth space in the data center chip market is far larger than that of the maturing smartphone market. For Qualcomm, the question is not whether to transform, but whether it can carve out a share in a data center race already occupied by players such as Nvidia.

Second growth curve is a concept originating in management consulting that refers to a new business line a company opens up before its core business slows or peaks, in order to carry future revenue growth. For Qualcomm, smartphone chips and patent licensing constitute the first curve, whose growth is constrained by the ceiling on global smartphone shipments; data centers and AI infrastructure are positioned as the second curve. The core assumption of this strategic framework is that the new business must be cultivated before the old one declines, otherwise the company will fall into an awkward transitional period. This is precisely the risk Qualcomm faces—its smartphone business has already seen a 20% year-over-year decline, while its data center business is still in an early stage of announced partnerships and products not yet at volume. Whether the time windows of the two curves can connect is the key to the success of the transformation.

Partnering with Amazon and Meta to Build an AI Data Center Collaboration Network

Qualcomm is not going it alone in entering data centers. The key information mentioned in the earnings report is that the company has established partnerships with hyperscalers such as Amazon and Meta, and the related plans were announced earlier. This means Qualcomm's AI data center chips are expected to enter the infrastructure systems of leading cloud providers and internet platforms directly, rather than building a customer network from scratch.

Tying up with hyperscale customers means more to a chipmaker than orders. It means products need to be validated under real large-scale workloads, and it also means Qualcomm must prove that its solutions have differentiated advantages in energy efficiency, cost, or specific workloads. Given the scale of AI compute investment by Amazon and Meta, such partnerships are themselves a strong signal.

Hyperscalers generally refer to companies such as Amazon AWS, Microsoft Azure, Google Cloud, and Meta that operate millions of servers and build large-scale data centers of their own. They are both the largest buyers of AI chips and important influencers of chip design. Partnering with such customers means chipmakers must meet extremely high standards: a single cluster can scale to tens of thousands or even hundreds of thousands of accelerators, with stringent requirements for interconnect bandwidth, failure rates, energy efficiency, and software stack compatibility. At the same time, hyperscalers generally adopt multi-vendor strategies and actively develop their own chips (such as Amazon's Trainium/Inferentia and Meta's MTIA) to reduce dependence on a single supplier. Qualcomm's ability to enter this system shows that its solutions have gained initial recognition for specific workloads, but it also means it must continuously prove its value against customers' in-house chips and other suppliers.

Where Qualcomm's Differentiation Lies

In the field of data center AI chips, Qualcomm faces a fairly clear competitive landscape: Nvidia holds a dominant position, players such as AMD continue to catch up, and cloud providers' in-house chips are also diverting demand. Qualcomm's entry point is likely related to its long-accumulated low-power design capabilities on mobile—against the backdrop of data center energy consumption becoming a core constraint, performance per watt is a persuasive selling point.

However, going from smartphone chips to data center chips is not simply a matter of scaling up. Data centers have vastly different requirements from mobile for interconnect bandwidth, software ecosystems, and memory architecture, and Qualcomm has quite a few gaps to close. The information disclosed on the earnings call remains mostly at the strategic level, with specific product roadmaps and performance metrics not yet detailed, which is also the main uncertainty for outsiders assessing its competitiveness.

Performance per Watt in the data center context refers to the compute provided per unit of power consumption, and is one of the core constraint metrics for AI infrastructure today. As GPU clusters scale up, power supply and cooling costs account for a significantly rising share of total data center cost of ownership. The electricity demand of some large AI data centers has reached hundreds of megawatts, approaching the limits of power grids and land capacity. Therefore, every tier of improvement in energy efficiency translates directly into lower operating costs and higher deployment density. Qualcomm's long-accumulated low-power design experience in smartphone SoCs—including heterogeneous computing scheduling, DVFS dynamic voltage and frequency scaling, and adaptation to advanced process nodes—can in theory be transferred to data center accelerators. But note that data center workloads are mainly sustained high throughput, which differs markedly from the intermittent workload patterns of mobile devices. Whether the energy efficiency advantage can be realized in real clusters still needs to be supported by measured data.

Software ecosystem is a threshold repeatedly mentioned in data center AI chip competition. One of Nvidia's core moats is its CUDA programming model and supporting libraries (cuDNN, TensorRT, etc.), which after more than a decade of accumulation have formed a huge developer community and a mature toolchain, with mainstream AI frameworks (PyTorch, TensorFlow) best supporting it. Even if new entrants have competitive hardware metrics, they need to provide easy-to-use compilers, operator libraries, framework adaptation, and migration tools; otherwise, customer migration costs are too high. AMD invested for years through ROCm before gradually narrowing the gap, and cloud providers' in-house chips are mostly first validated on internal workloads. For Qualcomm, completing the software stack is a long-term task running in parallel with hardware development, and this is also the hidden cost that outsiders most need to watch when assessing its data center strategy.

Whether the Transformation Pays Off Remains to Be Seen

Qualcomm's bet is essentially about finding a new revenue pillar during the window of decline in its smartphone business. Partnerships with Amazon and Meta provide a ticket to entry, but competition in the data center market is far more intense than in smartphone chips, and customers are more demanding about performance, ecosystem, and long-term supply capability.

In the short term, pressure from the smartphone decline will still weigh on overall performance; in the medium to long term, whether the data center business can generate revenue at scale depends on whether Qualcomm can turn partnerships into mass-producible, competitive products. The signal from the earnings call is clear: Qualcomm no longer intends to be just a smartphone chip company.

Whether this judgment is correct will be answered by upcoming product launches and customer orders.

One-line Summary

As smartphone sales fall 20%, Qualcomm teams up with Amazon and Meta to bet on AI data centers in pursuit of a second growth curve.

Paragraph Summary

Qualcomm's latest earnings call disclosed that smartphone business sales fell about 20% year-over-year compared with 2025, and the company explicitly positioned its AI data center business as its "second growth curve." Qualcomm has established partnerships with hyperscalers such as Amazon and Meta, with related plans announced earlier, giving its AI data center chips a ticket into the infrastructure systems of leading cloud providers. But in a data center race dominated by Nvidia, chased by AMD, and diluted by cloud providers' in-house chips, Qualcomm needs to prove its differentiated advantages such as performance per watt and close gaps in interconnect bandwidth and software ecosystem. Earnings information currently remains at the strategic level, with product roadmaps and performance metrics not yet disclosed, and whether the transformation pays off still awaits validation through mass-produced products and customer orders.