AI Agent Phones Flood the Market: Huaqin Technology Manufactures Stepfun's First Model; Five Key Hardware Segments Face Value Reshuffle

Artificial intelligence models are accelerating their migration from the cloud to the edge, with smartphones becoming the core battleground for leading companies vying for dominance in AI implementation. Chinese firms like Stepfun and ByteDance are intensively launching AI agent phones, while supply chain companies such as manufacturer Huaqin Technology (603296.SS) are deeply involved. On the hardware side, multiple segments from chips to cooling are undergoing an AI-driven value reassessment.

According to the STAR Market Daily, Stepfun is about to release its AI terminal brand, agent system, and its first AI agent smartphone. Sources indicate that this AI agent phone is being manufactured by Shanghai-listed Huaqin Technology. The relationship between Huaqin Technology and Stepfun is a deeply integrated partnership, not a simple original design manufacturer (ODM) model.

Stepfun is the large model startup in China that has invested most heavily in terminal Agent scenarios, having been the first to open-source its Step-GUI series models last year. Based on the GUI-MCP open protocol released by Stepfun, developers and terminal manufacturers can deploy an Agent-capable phone in as little as 10 minutes. Prior to this, Stepfun had already been providing model support for AI phones. Last October, ZTE (000063.SZ) launched the Nubia Z80 Ultra flagship model, equipped with Stepfun's GUI Agent model, capable of cross-application tool invocation to execute tasks based on commands.

At the upcoming 2026 World Artificial Intelligence Conference (WAIC), the first AI agent phone jointly created by ByteDance and ZTE's Nubia will also be exhibited.

From a technical perspective, AI phones are transforming from general smart hardware devices into "super carriers." Previously, the Doubao phone had already achieved integration between AI and the system, allowing Doubao to connect apps through the system, verifying the feasibility of deep binding between AI assistants and mobile operating systems. In June this year, WeChat and several phone manufacturers launched an Agent-to-Agent (A2A) assistant feature. Users can operate WeChat and send messages to designated contacts simply through the phone's AI voice assistant. WeChat customer service indicated that Honor phones have already taken the lead in implementing the A2A function, bringing a new interaction method to users.

Sinolink Securities pointed out that AI is upgrading from "a function within the phone" to "a system-level orchestrator that permeates interaction." For instance, the Doubao phone assistant bypasses the "glass wall" limitations of traditional apps, directly invoking the functions of various applications. As a super personal computing terminal, the phone carries almost all consumer-facing information flows and serves as a super entry point to the physical world. Only when AI breaks through the phone can other AI terminals potentially thrive sustainably relying on AI phones.

Rising Penetration Amid a Memory Crisis

The adoption speed of AI phones is continuously accelerating. According to statistics from China's National Development and Reform Commission's Department of Innovation and High-Tech Development, China's annual shipments of AI phones, AI PCs, and other smart terminals exceeded 100 million units last year, and are expected to maintain rapid growth this year. Sales of AI phones and AI PCs are anticipated to surpass non-AI products for the first time this year.

Overseas, OpenAI is accelerating the development of agent phones, with its latest target being mass production in the first half of 2027. A Counterpoint Research report shows that generative AI-capable models will account for 45% of global smartphone shipments in 2026, up from 36% in 2025, and are projected to reach 52% in 2027.

However, behind the bright growth expectations, the supply chain faces severe challenges. Counterpoint Research emphasized that while AI phones are rapidly proliferating, actual consumer usage demand, memory costs, and the ecosystem integration capabilities of different manufacturers will determine the next stage of market competition. Affected by the ongoing memory supply crisis, global smartphone shipments in 2026 are expected to plummet 13.9% year-over-year to 1.08 billion units, a record low.

The firm further noted that in the long term, memory remains the key factor determining whether generative AI can quickly enter the mid-to-low-priced phone market. Running AI models requires additional DRAM to store model weights, making it currently difficult to reduce the wholesale price of AI phones below $400. As memory supply pressure gradually eases and the optimization efficiency of on-device models continues to improve, generative AI features will further trickle down to more affordable smartphones.

Year	Share of Gen-AI Capable Models in Global Smartphone Shipments
2025	36%
2026	45%
2027	52%

Source: Counterpoint Research Report

Five Key Hardware Segments Face Value Reshuffle

The higher requirements of AI terminals for computing power, storage, battery life, and connectivity are driving profound changes in the hardware industry chain. Huachuang Securities stated that AI terminals are expected to boost the value of multiple hardware segments, driving a value reconstruction in chips, memory, cooling, batteries, and radio frequency (RF).

In the chip segment, System-on-Chip (SoC) design has shifted from a traditional CPU/GPU dual-core drive to a heterogeneous computing architecture with a clear division of labor among "CPU+GPU+NPU," establishing the NPU as the primary computing force for on-device AI. This means phone processors need to integrate more powerful neural processing units to efficiently run on-device large models.

Memory upgrades are equally urgent. Deploying large models locally consumes significant local memory. Taking the 7-billion-parameter LLaMA model as an example, it still requires a minimum of 3.9GB of memory under 4-bit quantization, driving phone memory evolution from the current 12GB and 16GB to higher specifications.

In the battery segment, on-device AI drives up overall power consumption. To maintain smartphone battery life, battery specifications are continuously upgrading. The penetration rate of technologies like silicon-carbon anodes and semi-solid-state batteries continues to rise, expected to drive sustained growth in battery energy density to provide greater capacity within limited space.

In the cooling segment, increased chip computing power raises cooling requirements, while still needing to meet the overall design constraints of the phone regarding weight and thickness. This is driving the penetration growth of efficient cooling solutions like graphene and vapor chamber (VC) heat spreaders.

In the RF front-end segment, AI applications demand higher throughput capacity and latency stability from network connections. Technologies like WiFi 7 and 5G-Advanced are likely to become subsequent upgrade directions to ensure on-device AI can collaborate with the cloud in real time.

From the deep cooperation between Stepfun and Huaqin Technology, to ByteDance joining Nubia at WAIC, to WeChat and phone manufacturers bridging A2A interaction, AI phones are no longer simple hardware stacks but a deep integration of operating systems, large models, and hardware ecosystems. Amid the deterministic trend of continuously rising penetration rates, when the memory supply bottleneck eases will directly determine whether generative AI can truly enter ordinary households, while the curtain on the value reassessment of hardware segments like chips, cooling, and batteries has only just risen.