Qualcomm Renews Apple Patent License as Chip Decoupling Continues

On Sept. 24 Hawaii time, as the Snapdragon Summit entered its final stage, Qualcomm announced it had renewed its global patent licensing agreement with Apple. The new agreement is set to take effect on April 1, 2027, with no term or financial details disclosed. According to reporting by Huxiu’s Liang Kaer, the renewal was seen as reassuring Wall Street for now.

The key point is that the renewal covers patents, not chips. Apple is putting its in-house C1 and C2 series basebands into more iPhones, moving away from Qualcomm-made semiconductors. Royalty payments will continue, even as Apple replaces purchased components with its own.

Patent Renewal Defends Profit Floor

The renewal removes the risk that Qualcomm’s royalty revenue could suddenly disappear after 2027. According to Bernstein analyst Stacy Rasgon and existing market estimates, Apple buys about $5 billion to $7 billion a year in basebands and RF front-ends from Qualcomm, plus about $1.5 billion to $1.7 billion in patent royalties.

The former is shrinking as Apple expands use of its in-house basebands. The latter is smaller, but Qualcomm’s licensing business, QTL, has long maintained pre-tax margins above 70%. Keeping that high-margin business supports earnings.

The pricing terms of the new deal were not disclosed, so analysts’ estimates may not apply directly. Still, continued royalty income is now assured, protecting the profit floor. The challenge remains that it will not sustain the growth trajectory.

In-House Basebands Keep Eroding Orders

Apple still cannot work around Qualcomm’s communications patents, but it is reducing its dependence on Qualcomm-made chips. Wider adoption of the C1 and C2 series means the loss of $5 billion to $7 billion a year in component orders. The patent renewal does nothing to stop that shift.

Sourcing is also changing in the Android camp. MediaTek is deepening joint tuning with device makers through its Dimensity open architecture. Longtime customers such as Xiaomi are also said to be strengthening their in-house SoCs.

In an interview, Qualcomm China Chairman Meng Pu struck a calm tone. He said OPPO and vivo have long used Qualcomm and MediaTek platforms side by side, choosing different chips for different products. He said customer-developed SoCs are not unusual, driven by technology strategy and a desire to showcase in-house capabilities.

For device makers, multi-sourcing and in-house development are technology choices. They are also ways to retain control over product definition, lower procurement costs and gain bargaining power. Qualcomm is in no position to keep relying on a single market.

Qualcomm Declares Shift to Agent-Centric Strategy

At an event ahead of the renewal announcement, Qualcomm President and CEO Cristiano Amon signaled a change in direction.

“We are moving from an ‘app-centric’ experience to a new ‘agent-centric’ era”

The entire summit was built around that call. Qualcomm is seeking to evolve from a supplier of mobile chips into an AI computing platform spanning phones, PCs, cars, wearables and data centers. Qualcomm calls the vision the computing continuum.

On the opening day, Sept. 22 Hawaii time, Amon divided human-machine interaction into two dimensions.

“One dimension is you performing tasks yourself, the other is your agents performing tasks for you”

In his vision, phones will not be replaced by smart glasses. Instead, phones will shift from devices that run apps to control hubs that orchestrate personal agents. Earbuds, glasses, PCs and cars will handle sensing and execution, while phones connect different devices and personal data.

In the third quarter of fiscal 2026, Qualcomm’s handset chip revenue fell to nearly 50% of total sales. In the same period, automotive revenue rose 61% year on year. By fiscal 2029, Qualcomm aims to raise non-handset revenue to $40 billion, building three pillars spanning handsets, automotive, IoT and data centers.

Chip Architecture Backs Continuum Vision

Traditional apps take a single instruction and perform a defined task. Agents must continuously sense their environment, understand context, break down goals, call tools and plan their next moves based on results. Peak compute alone cannot handle that.

In on-device agent workflows, the sensor hub handles continuous sensing, the CPU handles planning and task allocation, the NPU handles model inference and the GPU handles parallel computing. Data moves frequently between compute engines, so compute power, memory bandwidth, latency and power consumption can all become bottlenecks.

In the sixth-generation Snapdragon 8 Elite, the Oryon CPU breaks the 5GHz clock speed for the first time. It introduces a FlexCache shared cache pool, while shared memory for the Hexagon NPU is up 50%. It supports context windows of up to 32K tokens.

The more important change is a redesign of on-chip data movement. Durga Malladi, Senior Vice President of Qualcomm Technologies and General Manager of Technology Planning, Edge Solutions and Data Center, explained that agents generate large amounts of temporary data during inference, planning and tool calls. Storing that data in phone main memory would require shuttling it back and forth to the NPU on each use, increasing latency and power consumption.

Qualcomm’s solution uses large on-chip shared memory near the NPU and second-generation Adreno dedicated fast graphics memory on the GPU side. Frequently accessed data blocks and intermediate results are pinned directly on-chip to minimize access to external system memory. Large models cannot bypass main memory entirely, but data transfers can be reduced.

30B Model Shown Running Fully On-Device

With partners, Qualcomm ran a 30B MoE model entirely on a reference design. It deployed the StepEdge-Omni 30B MoE model on a reference design powered by a next-generation Snapdragon platform. In the demo, the model completed a continuous series of tasks locally, including email understanding, scheduling, calendar syncing, flight and hotel recommendations and email drafting, with no cloud calls needed.

The message on stage was clear: to support the shift of models from cloud to device through platform capability, not a single model. The idea is to extend CPU, NPU, connectivity and software capabilities to as many devices as possible.

Qualcomm’s strength used to be that customers needed Snapdragon. Going forward, the question is whether customers will still want to pay for Snapdragon even with more options. Its value as a cross-device AI platform will have to justify the price.

Cost, Permissions Challenge On-Device AI Rollout

The 30B-scale demo relies on high-capacity memory, advanced process nodes and thermal management. With memory prices surging and 2nm chip costs rising, bringing on-device AI to mainstream price tiers faces real hurdles. It may work in flagships, but reaching volume price segments will take time.

Permission barriers in the agent ecosystem are also a constraint. For on-device agents to call across multiple apps, OS-level interfaces, privacy and payment permissions must be opened up. A chipmaker cannot lead that alone. Qualcomm expects full cross-vendor, cross-device AI interoperability in 2027 to 2028.

Securing patent revenue shows the old business model still works. What Amon is talking about is what Qualcomm can become beyond phones. Apple still pays royalties, but as a customer it no longer wants to depend on Qualcomm.

Data Center Business Poses Biggest Uncertainty

Of the $40 billion target by fiscal 2029, automotive and IoT have a clear path. Data centers, by contrast, must grow from zero to more than $15 billion in revenue. That requires rapid scale in a market led by Nvidia and cloud providers’ custom ASICs.

The 61% growth in automotive supports the continuum vision. The falling share of handset chip revenue shows progress in reducing dependence. Success or failure in data centers will decide whether the overall target is met.

When handset chips alone can no longer support growth, the only path is to extend CPU, NPU, connectivity and software capabilities across diverse devices. This Snapdragon Summit can be seen as declaring that shift.

Editorial Opinion

On the short-term impact: Over the next three to six months, we expect Qualcomm’s earnings outlook to stabilize on reassurance from the patent renewal. Continued royalty income should ease investor concerns. Meanwhile, component orders from Apple will keep falling, and price negotiations for handset chips will likely get tougher. Growth in automotive should continue to offset the drag.

On the long-term view: Over one to three years, we expect practical on-device agents to change how chips are chosen. Overall design balancing memory bandwidth, latency and power will matter more than peak compute. Companies that resolve permission issues with OS providers will take the lead. We believe the data center business will determine how Qualcomm is valued.

Questions from our editors: Whose cost will mass-produce the ability to run 30B-scale models on-device. Will device makers absorb memory and advanced-node costs, or pass them on in device prices. When agents span multiple apps, who bears responsibility for privacy and payments. The division of roles among platform providers is now in question.

References

• “ 苹果续约之前，高通在夏威夷刚讲了一个更大的故事 “, by 梁卡尔 — 虎嗅网, 2026-09-26T23:00:00.000Z (ARR)
• Source URL: https://www.huxiu.com/article/4894069.html?f=rss