Summary of Key Points
StepStar (an AI company led by Yin Qi) originally focused on large-scale models as its core technology. However, it has now shifted its strategy to simultaneously develop both "large-scale models" and "agent terminals." The company is starting with smartphones, launching the STEPX Neo as its first product, and also plans to expand internationally by selling model APIs and smartphones. Its goal is to create an independent business ecosystem that integrates "models, systems, and terminals"—this approach aims to avoid direct competition with giants like ByteDance and Alibaba while not merely being a technology provider that earns modest profits. Nevertheless, this path is fraught with challenges: StepStar must compete with its existing clients (smartphone manufacturers) and simultaneously tackle the demanding tasks of model training and hardware manufacturing.
Detailed Analysis
1. Why focus on smartphones instead of large-scale models?
The business of solely developing large-scale models is becoming increasingly challenging:
- Pressure from giants: Companies like ByteDance's DouBao and Alibaba's Qianwen are rapidly gaining traction in the consumer market due to their access to massive user bases and financial resources, making it difficult for startups to compete.
- Model innovation cycle: The lead advantage of each new model generation is only short-lived, and investing billions each year may only secure a place in the competitive landscape.
- Lack of a self-sufficient ecosystem: If models are sold to smartphone or automotive manufacturers, they become just one component of their products, resulting in limited revenue and no control over pricing or user behavior.
- Low commercial potential: While customizing models for businesses (ToB) can be profitable, the Chinese ToB market is complex, making it hard to scale such services and potentially draining company resources over the long term.
Therefore, StepStar decided to develop its own terminals. By integrating models directly into smartphones, users will consume tokens associated with those models (similar to "data usage fees" for AI services). This approach allows the company to control user access points and directly monetize the value of its models.
2. Yin Qi's persistent focus on AI implementation: What lessons has he learned from previous attempts?
Yin Qi is not new to this endeavor:
- Lessons from Megvii: Early in his career, Megvii focused on computer vision (face recognition), which could only solve "perception" problems. Commercialization was either uncertain through government projects or time-consuming with autonomous driving initiatives.
- Try with Qianli Technology: Later, he joined Qianli Technology (renamed from another company) to work on "AI + automotive" applications, where StepStar provided models while Qianli developed intelligent cockpits and driving systems—a test of the "model + terminal" concept.
- StepStar's advancement: With the STEPX Neo, StepStar is extending this approach to more user-friendly devices. It has brought in experienced professionals such as Zhao Ming (former CEO of Honor) and Ni Jiayue (from Huawei/Honor) to strengthen its capabilities in hardware, supply chain, and distribution.
Yin Qi's goal is clear: to transform AI from a behind-the-scenes technology into a tangible product that users can use daily.
3. StepStar's serious approach to smartphone development:
StepStar is not just creating a concept phone:
- Separate company structure: The smartphone business is housed in a new company called Zhiyue Star, ensuring no interference with its large-scale model operations.
- Professional team recruitment: Many of the employees at Zhiyue Star come from Honor, bringing expertise in product development, supply chain management, and channel establishment (Honor quickly re-established its infrastructure after separating from Huawei).
- Advanced system: They have developed the Step AOS system, which is not a standard smartphone OS but an "agent runtime environment" that can be used in cars, PCs, and IoT devices.
- International expansion: The company plans to sell both smartphones and voice model APIs overseas, leveraging models to enhance smartphone functionality and vice versa.
4. Major challenges:
- Confronting former clients as competitors: StepStar previously collaborated with top smartphone brands like OPPO and Honor; now that it develops its own phones, these clients may become competitors. Will they continue to use its models?
- Consumer acceptance: Consumers buy smartphones based on factors like price, hardware, camera performance, and battery life, not on the concept of "model integration." StepStar lacks the brand recognition and sales network of Honor, making it difficult to attract users.
- Dual challenges: The company must invest heavily in model training and smartphone production while also managing supply chains and inventory. Both areas are costly and require sustained funding.
- International barriers: Entering foreign markets involves adapting to local communication standards and compliance requirements, which is different from selling model APIs.
5. Beyond smartphones: Building an "AI terminal empire"?
Smartphones are just the beginning for StepStar:
Its ambition is to create a complete ecosystem that spans models, systems, and terminals:
- Smartphones as a starting point: They provide direct access to AI services for users.
- Automotive applications: Qianli Technology is working on AI + automotive solutions, integrating with smartphone systems.
- Future expansion: The Step AOS system can be adapted to various devices, making AI ubiquitous.
StepStar aims to break free from relying on model rankings and take control of the entire AI service chain—from technology to products, and from users to revenue.
Conclusion
StepStar's transformation represents both a breakthrough and a risk. If successful, it could free itself from the dominance of giants and establish its own AI ecosystem. Failure could result in losing both its model and hardware businesses. However, Yin Qi is determined not to be merely another supplier of AI components for others.