In 2026, when the demand for AI computing power hits an all-time high, the three giants of mobile phone chips each made an unusual move: Qualcomm and MediaTek quietly withdrew a combined 20,000-30,000 wafer orders from TSMC; Apple, for the first time, entrusted part of its self-developed chip foundry work to Intel. TSMC's best production capacity is no longer prioritized for mobile phones. How will the giants of mobile phone chips navigate the AI era?

Over the past two years, TSMC's revenue structure has undergone a complete structural shift, with the revenue contribution from the mobile phone chip business continuously declining and the industry's focus fully shifting. In the first quarter of 2024, smartphones accounted for 38% of TSMC's revenue, while HPC (High-Performance Computing) accounted for 46%; by the first quarter of 2026, HPC's revenue share had climbed to 61%, while smartphones had plummeted to 26%. In just two years, the revenue share ceded by the mobile phone chip business has been almost entirely taken over by AI computing power-related businesses. The revenue growth driven by AI computing power far outpaces any recovery in the mobile phone chip business. With shipment volumes of mobile phone chips remaining largely flat, their share of TSMC's total revenue has been continuously diluted, and their industry standing has gradually declined.

A closer look at TSMC's revenue distribution across core process nodes reveals the direction of its capacity resource allocation and confirms the marginalization trend of the mobile phone chip business. The revenue share of the 3nm advanced process steadily rose from 9% in the first quarter of 2024 to 28% in the fourth quarter of 2025, maintaining strong growth momentum; the 5nm process has long held a central position in revenue, with these two high-end processes combined accounting for over 60% of TSMC's revenue in the first quarter of 2026. Most of the customers with access to these high-priced, advanced-process capacities are high-value computing power clients, such as NVIDIA for AI chips, Apple for self-developed chips, and cloud AI computing power chips. These clients are willing to pay a premium to secure top-tier capacity, becoming TSMC's core service targets at this stage.

TSMC's scarce high-end advanced capacity is tilting towards the AI computing power track (translated as "sector" or "field"). Mobile phone chip foundry customers now lag behind AI computing power chip customers in terms of capacity priority, supply scheduling, and resource allocation.

Qualcomm and MediaTek's reductions in foundry orders represent concessions aligned with the current capacity landscape. TSMC's high-end capacity is in short supply, and with premium (translated as "premium" or "high-quality") capacity unavailable, continuing to stockpile large volumes of advanced-process wafers would only increase financial pressure and inventory risks. How can mobile phone chip manufacturers find solutions amid capacity constraints?

The three giants of mobile phone chips each have their own approaches.

MediaTek is choosing to shift resources away from mobile SoCs. Global smartphone SoC shipments are expected to decline in 2026, with Counterpoint Research data showing an 8% year-over-year drop in global smartphone SoC shipments in the first quarter of 2026, with both Qualcomm and MediaTek experiencing double-digit declines. Counterpoint notes that MediaTek faces significant pressure in the entry-level market due to the memory chip crisis.

Facing such challenges, MediaTek is proactively scaling back its mobile business and shifting its core growth focus to new sectors. In early 2026, MediaTek explicitly stated: While smartphones remain a core business, the company is repositioning itself from a mobile phone chip manufacturer to an AI chip company. Google's next-generation TPU is set to enter mass production in the second half of 2026, with MediaTek handling part of the ASIC design and I/O subsystem. Multiple reports indicate that MediaTek's data center AI ASIC revenue will exceed $1 billion in 2026 and reach "several billion dollars" by 2027.

At MWC 2026, MediaTek showcased a range of AI applications "from smartphones, tablets, and IoT to edge servers," demonstrating a complete layout in on-device AI. In the automotive electronics sector, MediaTek is collaborating with Denso, one of the world's largest auto parts suppliers, to develop ADAS custom chips, with significant automotive revenue growth expected to begin in 2026. Combined with MediaTek's existing in-vehicle communication and infotainment solutions, this represents a gradual expansion from in-vehicle connectivity/infotainment to automated driving SoCs.

In its Q2 FY2026 financial report (January 2026–March 2026), Qualcomm lowered its mobile chip revenue guidance for the quarter to approximately $6 billion, down about 13% year-over-year, and warned that rising memory prices and supply constraints had led OEMs to reduce inventory and cut shipments of mid-to-low-end models. Counterpoint estimates that Qualcomm's smartphone SoC shipments will decline by about 8.8% year-over-year in 2026, with a slight dip in market share, though flagship and upper-mid-range segments will maintain growth. Similar to MediaTek, Qualcomm faces a structural shift of "stable high-end, shrinking mid-to-low-end."

Qualcomm management acknowledges pressure on smartphone shipments, attributing the decline in mid-to-low-end models to rising memory prices, prolonged replacement cycles, and OEM inventory reductions, and no longer strongly expects a return to high growth in overall mobile SoC shipments. Qualcomm positions flagship and upper-mid-range phones as "personal AI gateways," maintaining premium ASP and differentiation by enhancing NPU, local large model capabilities, and smart agent ecosystems; in the mid-to-low-end market, it focuses on platform reuse and cost optimization to ensure profitability.

At MWC 2026, Qualcomm showcased the Wearable Platform Supreme Edition, X105 modem, and WiFi 8 chips, all centered around AI and connectivity upgrades, expanding the Snapdragon brand across multiple device forms, including wearables, PCs, XR, and automotive.

Similar to MediaTek, Qualcomm is also ramping up its AI chip efforts. In 2026, Qualcomm confirmed securing a custom chip order from its first hyperscale cloud customer, entering the cloud AI accelerator field. Qualcomm stated that such custom chip orders offer significantly higher ASP and margins than traditional mobile SoCs, becoming a key growth driver to hedge against smartphone cycle volatility.

Qualcomm has accumulated substantial platform capabilities in automotive chips, making it an important growth market in 2026. Qualcomm claims to have secured over $30 billion in Design-Win, expecting automotive revenue to maintain double-digit growth in the coming years to offset volatility in smartphones and IoT. It has repeatedly cited automotive as a key growth engine in financial reports.

Unlike MediaTek, Qualcomm is heavily betting on data connectivity and interaction between personal AI products, with efforts in wearables, PCs, and other products. Qualcomm CEO Cristiano Amon predicts that the market size for "wearable AI products" could exceed 100 million units in the coming years, with the potential to reach billion-unit levels.

Apple was once TSMC's most important customer, with its A-series/M-series chips long locking in TSMC's most advanced process nodes. Now, for the first time, Apple is explicitly entrusting some of its chip production to Intel. Offloading part of its capacity to Intel represents Apple's risk diversification amid TSMC's tightening supply; to amortize its increasingly expensive chip costs, Apple's solution is cross-category reuse. Supply chain data reveals that the A20 and A20 Pro chips in the iPhone 18 Pro series are estimated to cost $280 each, a significant increase from the previous-generation A18 and A19, coupled with memory cost increases exceeding 200%, placing the high-end iPhone 18 under its most severe BOM cost pressure in recent years.

The A-series chips' cross-device compatibility relies on the shared development ecosystem of iOS and macOS, ensuring highly unified software and hardware adaptation. Apple uses the A18 Pro mobile chip in its MacBook Neo model, with the same SoC powering both smartphones and ultraportable laptops. The MacBook Neo not only streamlines chip R&D expenditures but also showcases Apple's chip ecosystem barrier: leveraging unified chip designs across multiple device types, spreading R&D costs across massive overall shipments, and maximizing the utilization of TSMC's foundry capacity.

In 2026, mainstream smartphone manufacturers' new AI initiatives are primarily focused on "AI agents," "system-level large model integration," and "device-cloud collaboration," directly raising the bar for mobile phone chips in terms of computing power, energy efficiency, memory bandwidth, and AI engine architecture.

Qualcomm CEO repeatedly emphasized the same judgment at Web Summit 2026 and the Davos Forum: The AI race will be decided at the "edge"—where devices, data, and users converge. When AI models can process requests locally on your phone rather than routing to cloud servers, the result is lower latency, better privacy protection, and no cloud bills.

Counterpoint Research predicts that shipments of smartphone SoCs supporting Agentic AI will grow at a 281% CAGR (2025-2027), with penetration rising to 32% by 2027, up from just 4% currently. Counterpoint Vice President Peter Richardson stated: "We expect one in three smartphones shipped in 2027 to feature Agentic AI capabilities, covering high-end (over $600) and upper-mid-range ($250-$600) price segments." Analyst Shivani Parashar added that MediaTek has taken the lead with its Dimensity 8400, 8450, and 8500 chips supporting Agentic AI, gaining an early advantage in expanding to mid-range models.

OpenAI is collaborating with Qualcomm and MediaTek to develop an AI agent-centric smartphone, targeting annual shipments of 300-400 million units by 2028. The device's architecture is designed to handle lightweight tasks (context awareness, memory management, small AI models) locally, while offloading complex reasoning tasks to the cloud. This device will be redesigned from the ground up to support continuous, low-power local AI inference, rather than simply adding an NPU to existing chips.

Mobile SoCs are stepping out of the spotlight in the semiconductor industry's core narrative, but smartphones remain the world's largest-volume intelligent terminals, with growing demand for on-device AI inference. How can mobile phone chip companies find new growth? The answer may lie in the development trajectory of AI agents. While AI has pushed mobile phone chips down TSMC's priority list, AI agents require local processing, ultimately bringing the search for computing power back to smartphones. Counterpoint reports that penetration of Agent AI-capable phone chips was just 4% at the end of 2025 but is expected to reach 1 in 3 phones by 2027, with penetration exceeding 80% in high-end models.

AI phones represent an upgrade from single-performance selling points to AI platform + ecosystem products. What AI has taken away in this round may return in another form.