AI / LLMChina
← Sticker Wall
JOURNEY · DETAIL
01.AI: The Chinese Unicorn Led by Kai-Fu Lee, Pivoting from General LLMs to Image Generation
Founded: Kai-Fu Lee · 01.AI
JOURNEY
Key Fields
FIELD STAMPSIndustryAI / LLM
RegionChina
ScaleMid-size
ChannelOther
Origin
Kai-Fu Lee founded 01.AI in 2023 with the initial goal of building a 'Chinese OpenAI,' focusing on general-purpose large models and world-class foundation models. The company secured hundreds of millions of dollars in early funding, rapidly scaling to hundreds of employees with R&D centers in Beijing, Shanghai, and Silicon Valley. However, as a domestic price war for LLMs erupted, general-purpose conversational models struggled to differentiate or achieve a commercial loop. By the second half of 2024, the company was forced to scale back, facing reports of layoffs and business restructuring. With limited resources, Lee decided to bet on AI image generation—a more vertical and commercially viable direction—to leverage existing technical assets for survival.
Milestones
2023
Inception Turning Point
Kai-Fu Lee announced the founding of 01.AI, positioning it as an AI 2.0 company, and immediately completed a tens-of-millions-dollar angel round. Starting with open-source models, the company released Yi-34B, which briefly topped the Hugging Face leaderboard, drawing industry attention. The team expanded to over 200 people within six months, leading to high R&D and labor costs that would later pressure cash flow.
2024
Rapid Expansion Growth
The company announced a new funding round, with its valuation exceeding $1 billion, becoming a star unicorn in the domestic LLM space. During this period, it released the Yi-VL multimodal model and the Yi-Lightning series, but commercial adoption remained sparse. API usage fell far below expectations, and the consumer-facing product 'Wan Zhi' failed to gain traction, leading to internal strategic disagreements.
2024
Strategic Retrenchment Failure
A full-scale price war broke out in the domestic LLM market, with ByteDance's Doubao and Alibaba's Tongyi Qianwen offering near-zero-profit pricing. 01.AI's general-purpose conversational product usage plummeted, and API revenue failed to cover compute costs. The company was forced to cut non-core R&D lines and implement layoffs, while several co-founders departed, including model training lead Huang Wenhao, who joined Alibaba.
2025
Pivot Attempt Transition
The company publicly adjusted its strategy, proposing a 'Product-Model Integration' path—deeply coupling models with products—and focusing on image generation and creative tools. It launched the AI painting product Yi-Portrait, focusing on portrait generation and stylization to acquire users quickly. Internal organizational structures were adjusted to concentrate resources on high-potential projects.
2025
Commercialization Turning Point
According to LatePost, by the first half of 2025, 01.AI no longer relied solely on external funding, with some business units achieving self-sufficiency. API services and enterprise customization projects began contributing revenue, though the overall scale remained significantly smaller than industry leaders. The team size was reduced from a peak of ~300 to around 100, and office space was downsized, described by media as 'surviving in the cracks.'
2026
Market Reshaping Growth
In 2026, 01.AI integrated its AI image generation tools into B2B services, providing customized image-to-image solutions for e-commerce, advertising, and gaming clients. Media reports indicate that monthly API usage has grown several-fold compared to pre-pivot levels, with retention rates for some vertical industry clients exceeding 70%. Kai-Fu Lee stated publicly that the company optimized its cost structure in 2025, finding a new position between foundation models and pure applications.
Turning Points
- The outbreak of the general-purpose LLM price war caused daily usage of 01.AI's conversational products to shrink, forcing management to admit that the foundation model route could not survive independently.
- The departure of multiple co-founders and core engineers, including model lead Huang Wenhao to Alibaba, forced the company to redesign its tech stack and organizational structure.
- The shift from a 'big and comprehensive' foundation model route to vertical image generation scenarios marked the company's transition from technology-driven to product- and revenue-driven.
- The 2025 establishment of the 'Product-Model Integration' strategy and the launch of Yi-Portrait gave the company its first independent product matrix for real users.
Failures & Pitfalls
- The commercialization of general-purpose LLM conversational products failed; API usage could not cover training and inference costs, proving the 'OpenAI-style' route unviable in the Chinese market.
- Strategic retrenchment in late 2024 led to large-scale layoffs and the departure of early co-founders and core technical staff, severely damaging team morale and external trust.
- Early consumer-facing image generation products suffered from low retention and lacked the viral potential of hits like Miaoya Camera, resulting in poor returns on initial marketing spend.
- Blind expansion during the early phase, with R&D centers in multiple locations, led to excessive management overhead and low decision-making efficiency, causing several parallel projects to be scrapped during the pivot.
关键成功要素
- The founder's brand effect secured massive initial funding but also burdened the team with unrealistic expectations, creating greater public pressure during strategic adjustments.
- Technical accumulation can be transferred to vertical applications after setbacks in the foundation model race, provided the team retains sufficient model training and productization capabilities.
- In a market dominated by giants, deep customization for vertical scenarios is a key lever for small and medium-sized companies to build moats.
- The speed of organizational contraction determines survival; rapidly adjusting from 300 to 100+ employees was a proactive move to protect cash flow.
- Enterprise clients show significantly higher willingness to pay for image generation tools than general consumers, making B2B cash flow more stable.
Lessons
- Startups must be wary of the 'scale illusion' brought by high financing; blind expansion of personnel and compute makes companies more vulnerable during industry winters.
- The window for technical differentiation is extremely short; general capabilities are quickly matched by giants, necessitating early design of scenario- and product-level moats.
- A stable core team is a prerequisite for successful strategic transformation; the departure of co-founders slows down the transition and shakes investor confidence.
- 'Surviving in the cracks' means accepting a smaller market segment than the giants and trading deep industry understanding and high-touch service for revenue.
Core Data
- Initial Funding：Tens of millions of USD (based on public data, not independently verified)
- Peak Valuation：Over $1 billion (based on public data, not independently verified)
- Staff Size Change：Reduced from ~300 to ~100 (based on public data, not independently verified)
- Monthly API Usage Growth：Several-fold increase compared to the 2024 trough (based on public data, not independently verified)
- Enterprise Client Retention Rate：Over 70% (based on public data, not independently verified)
- Consumer Product MAU：Not disclosed, but lower than the peak of Miaoya Camera (based on public data, not independently verified)
Competitors / Peers
01.AI faces multi-front competition in the AI image generation space: ByteDance's Jimeng leverages Douyin's traffic and massive compute; Alibaba's Tongyi Wanxiang and Tencent's Hunyuan exert price pressure in the B2B and enterprise markets; and the open-source ecosystem (Stable Diffusion, ComfyUI) continuously compresses profit margins at the tool layer. Additionally, vertical platforms like LiblibAI and Haiyi AI have built strong community stickiness and workflow moats among designers. 01.AI's current differentiation lies in its dual mastery of image generation foundation models and enterprise-grade delivery experience, though it remains orders of magnitude behind top competitors in brand awareness, user scale, and ecosystem richness.