On July 9th, StepFun, a prominent unicorn in the large model sector, hinted at the imminent launch of the world's first AI agent phone. On the same day, Ni Fei, Senior Vice President of ZTE, revealed on Weibo that Nubia, a ZTE subsidiary, would unveil the phone at the 2026 World Artificial Intelligence Conference next week.

The news comes as no surprise. Over the past six months, from ByteDance's Doubao teaming up with ZTE to launch AI phones, to OpenAI reportedly making strides in AI Agent phone development with mass production expected as early as 2027, it's been evident that large model companies are venturing into hardware.

However, it's noteworthy that, according to a Cailian Press report, 'StepFun is set to launch an AI agent phone, with A-share listed company Wingtech handling contract manufacturing. Wingtech and StepFun have forged a deeply intertwined cooperative relationship, transcending a simple OEM model.'

In fact, Tianyancha data reveals that among the investors in StepFun's latest funding round of approximately $2.5 billion, several hardware supply chain names appear, including Wingtech, Longcheer Technology, OmniVision Group, and ZTE.

The StepFun phone, produced by Nubia and manufactured by one of its investors, Wingtech, represents a unique 'investor + contract manufacturer + brand' trinity in industry history. Here, large model companies lead in defining products and experiences, contract manufacturing giants provide manufacturing prowess and capital support behind the scenes, and traditional brand manufacturers offer channels and user recognition. Each party occupies a distinct position yet is deeply interconnected.

While the large model company's foray into phone manufacturing is sure to draw significant attention to its 'crossover,' it also hints at a new industrial division of labor paradigm—software-hardware collaboration in the AI era may necessitate an organizational approach distinct from that of the mobile internet era.

I. Why Do Large Model Companies Need 'Manufacturing Allies'?

The primary motivation for large model companies to venture into hardware manufacturing is to find practical use cases.

Over the past two years, firms like StepFun, Moonshot AI, and Zhipu AI have primarily competed on parameters, benchmarks, and developer ecosystems. While model capabilities have advanced rapidly, commercialization remains a challenge—the API-based business model struggles to build user loyalty or generate revenue commensurate with R&D investment.

At this time last year, StepFun estimated its 2025 revenue target at 1 billion yuan, a figure far from sufficient to establish a foothold compared to the R&D and computing costs incurred during the same period.

As model capabilities cease to be an absolute differentiator—with gaps narrowing in language, multimodality, and reasoning—the next question naturally arises: How can model capabilities be transformed into something users interact with daily?

The answer lies in endpoints. Phones, cars, glasses, earphones, robots—any screen, microphone, or sensor could serve as an entry point for models to reach users.

Unlike the mobile internet era, where the 'application layer reaped the rewards,' AI-era model companies risk being blocked from users by hardware manufacturers and operating systems if they do not proactively penetrate endpoints. Apple's closed ecosystem and self-developed large model route is clear, and Huawei's HarmonyOS AI capabilities are rapidly evolving. For large model companies, the window of opportunity is narrower than imagined.

However, it's impractical for a large model company to build its own phone production line. Phone manufacturing is a capital-intensive, low-margin, high-turnover business, requiring years of supply chain management, quality control, and delivery cycle accumulation.

Even for a giant like Wingtech, which surpassed 170 billion yuan in revenue in 2025, the overall gross margin is less than 8%, with mobile terminal business margins even lower. This business is fundamentally scale-driven, not technology-driven. If a large model company dives in, it would not only strain cash flow but also divert attention from model R&D.

This led to the cooperative model between StepFun and Wingtech: StepFun develops the AI brain, defining the interaction logic and user experience of the agent system; Wingtech handles manufacturing, providing end-to-end contract manufacturing capabilities from design to delivery. The relationship is not merely a client-vendor arrangement—Wingtech is itself an investor in StepFun, aligning the interests of the manufacturing and model ends through capital ties. This is extremely rare in traditional ODM models.

In the past, contract manufacturers and brands engaged in strategic maneuvering (a balance of power), with brands pressuring prices and contract manufacturers controlling costs. However, when a contract manufacturer holds equity in a model company, its concern shifts beyond per-unit earnings to whether the product succeeds, achieves sustained shipments, and drives further collaborations.

This is the most noteworthy industrial signal behind the StepFun phone: A deep binding between large model companies and manufacturing giants is emerging, surpassing traditional supply chain relationships. Once this model proves viable, its scalability will far exceed self-development.

II. The ODM Giant is Crossing Eras

From Wingtech's perspective, betting on StepFun is not a mere financial investment decision.

Wingtech is China's undisputed leader in consumer electronics ODM, ranking first globally in smartphone ODM shipments for five consecutive years, with a 22.5% market share in overall consumer electronics ODM in 2024. It also leads in ODM for tablets and smart wearables, serving clients like Samsung, Xiaomi, OPPO, and Lenovo. In 2025, its annual revenue reached 171.4 billion yuan, up 56% year-on-year, with net profit attributable to shareholders at 4.05 billion yuan.

However, as mentioned earlier, the industry's overall gross margin is less than 8%, with mobile terminal business margins declining as smartphone contract manufacturing accounts for a larger share. Despite its scale, profitability relies heavily on industry trends, leaving limited autonomy.

Meanwhile, although its AI server business grew rapidly, with annual revenue exceeding 40 billion yuan, core components like GPUs still require external procurement, facing the same dilemma of 'high turnover, low margins.' Operating cash flow even turned negative in 2025. The company's asset-liability ratio reached 72.62% by the end of 2025, highlighting the need for a business model breakthrough.

This is not unique to Wingtech but inherent to the entire ODM industry. With smartphones entering a mature market, replacement cycles lengthening, and hardware innovation margins diminishing, brand manufacturers wield increasing pricing power, squeezing contract manufacturers' negotiation room. To escape the low-margin cycle of 'earning a few yuan per phone,' Wingtech needs new value growth drivers.

AI offers this opportunity. If large models truly penetrate phones, PCs, cars, and IoT devices, the value of endpoints will shift from standardized hardware to personalized, continuously evolving intelligent services. For contract manufacturers, this means upgrading from 'helping brands build devices' to 'helping brands build intelligent devices.' The former derives value from manufacturing capabilities and supply chain efficiency; the latter from software-hardware synergy and AI capability integration.

Wingtech's investment in StepFun is essentially a proactive move to secure a foothold in this transformation. It's not simply about selling more AI phones but redefining its industrial role—from a pure hardware contract manufacturer to an infrastructure provider for AI terminal ecosystems. If this role solidifies, its profit structure could shift upstream from contract manufacturing fees.

From this perspective, Wingtech's AI industry布局 (layout) is not scattered financial investments but a systematic strategy.

Investing in StepFun secures access to large model capabilities. Investing in OmniVision Semiconductor, GigaDevice, and Montage Technology secures core components like image sensors, memory chips, and memory interfaces. Investing in Victory Giant Technology secures AI server PCBs. It has even ventured into humanoid robots and embodied AI.

The logic behind this portfolio is clear: When the AI terminal wave arrives, Wingtech will not merely be an executioner of contract manufacturing but a critical node deeply embedded in every link of the industrial chain.

III. A New Paradigm of Industrial Division of Labor is Emerging

From a broader industrial perspective, the emergence of the StepFun phone may signal that AI terminals are entering an era of manufacturing implementation, albeit with differences from traditional contract manufacturing models.

In the mobile internet era, the division of labor in the phone industry was clear: Chip manufacturers provided underlying computing power, operating system vendors offered software platforms, brand manufacturers defined products and integrated supply chains, and ODM contract manufacturers handled manufacturing. Brand owners served as the hub of the entire value chain.

However, in the AI terminal era, the hub may shift. Large models define the core experience of agent systems—users perceive not 'how fast the Snapdragon runs' but 'whether the AI understands me, remembers my habits, schedules my itinerary, and operates apps for me.' This capability is defined not by brand owners but by large model companies.

When large model companies become the definers of experience, the central role of brand owners in the value chain is shaken. Brand owners remain crucial—channels, user recognition, after-sales service, and quality endorsement are capabilities large model companies cannot quickly replicate. However, brand owners are no longer the sole dominators. A new industrial division of labor has emerged:

Large model companies output intelligent capabilities, brand manufacturers provide market access and user trust, and ODM contract manufacturers offer manufacturing and supply chains. Each party plays an irreplaceable role, but who occupies the hub position in the value chain depends on who grasps the core decision-making factors for users.

StepFun's approach takes this new division of labor further. It does not merely pre-install an AI assistant on Nubia phones but creates a standalone AI terminal brand, treating the phone as a physical shell for the agent system.

For users, who manufactures the phone or what chips it uses may become less important; what matters is whose agent it houses and what it can do. If this logic holds, large model companies will wield far greater influence in the industrial chain than any app developer today, potentially redefining the power dynamics between hardware and software vendors.

Of course, from an industrial history perspective, whether this model can scale remains to be seen. The existence of Huawei's HarmonyOS AI and Apple's closed ecosystem means large model companies will never secure the top brand access points. Current agent phones are more of an attempt to open these entry points. If, in this experimental field, AI phones can deliver differentiated experiences, the model of 'large models defining experiences, ODMs handling manufacturing, and brands providing channels' could expand across the industry.

Ultimately, the question returns to the industry: In the agent era, can large model companies, contract manufacturers, and brand owners find a more efficient cooperative model than traditional supply chain relationships? If the answer is yes, then today's StepFun and Wingtech may be just the first dominoes in a larger restructuring.

Source: Songguo Finance