Li Tao, APUS: “Don’t Block the Main Lane of Large Model Development”

Data, business models, and business logic are the true barriers.

APUS, founded in 2014, is a Chinese internet company and one of China’s earliest internet companies going global. Over its 11-year history, APUS has undergone two major transformations in the field of artificial intelligence. The first was in 2022, when the company fully committed to the more promising direction of general-purpose large models. The second was in August 2023, when APUS decided to shift focus from competing on models themselves to developing the application ecosystem around large models.

Li Tao, Chairman and CEO of APUS, stated, “When building applications and ecological barriers, you must not block the main lane of large models. Otherwise, the data, model, and technology barriers you painstakingly built over one or two years could easily be overtaken by large models.” This is an important recommendation Li Tao gives to AI companies.

Below is the transcript of Li Tao’s speech at the 2025 DEMO WORLD Enterprise Innovation & Venture Conference, organized by Chuangyebang.

2014: APUS’s Global Ambition

In 2014, I looked back at APUS. It was China’s earliest internet company going global, with one goal being to bring Chinese internet innovations to the world. At that time, China’s internet had reached an advanced stage and was becoming a global driver of internet development, comparable to the United States. Chinese internet had notable advantages in business models and applications. We aimed to extend China’s mature internet products, technology, and business models globally. At that time, of the world’s roughly 8 billion population, about 4.5 billion people had not yet accessed the internet.

After ten years of growth, APUS now serves over 2.4 billion users worldwide. In 2022, OpenAI released ChatGPT, opening a new direction for AI globally. Although AI had been explored in different ways (such as DeepMind’s matches against Go players Lee Sedol and Ke Jie), it still faced enormous challenges. In March 2022, GPT-3 was released, sparking widespread attention in China. By November 2022, ChatGPT had caused a revolutionary wave of AI awareness worldwide, including in China.

APUS had been researching AI since 2018, primarily focused on small models. While based on neural network principles, these small models differ from general-purpose large models. After observing the trajectory of general large models in 2022, we recognized the enormous potential of AI and fully invested the company into AI research.

Today, we define AI as the Fourth Industrial Revolution, shaping the next 50 to 80 years of economic development. This revolution will not only significantly improve productivity and efficiency but will also disrupt production relations. Our full-scale engagement with AI is based on this strategic consideration.

Being a globally-minded company, APUS has integrated AI with its international operations. Today, I want to share APUS’s progress in AI, particularly in application and ecosystem development under the global context.

1. Saying Goodbye to Model Competition, Focusing on Application Ecosystems

In 2022 and 2023, APUS was involved in the model “race.” On April 18, 2023, we released a hundred-billion-parameter multimodal model, becoming the first Chinese company to do so and one of the first to receive national AI model registration approval. We currently hold multiple general-purpose and industry-specific model registration certificates, the highest number in China.

However, by August 2023, we realized that competing in general-purpose models was extremely challenging. We strategically decided to shift from model competition to focusing on applications and ecosystem development.

We understood that technologies like Agent and MCP provide broader opportunities. Over the past decade, APUS expanded globally, from operating systems and browsers to security software, tools, and content applications. APUS Bible was our first flagship product.

APUS Bible goes beyond reading texts. It includes illustrated editions, AI restoration, and interactive features—all powered by large models. Unlike some AI chat or social apps that trend toward trivial content, our Bible app emphasizes serious content and social interaction, creating unique application challenges.

In healthcare, APUS developed AI Health, leveraging our proprietary medical large models. Currently, our AI doctor, deployed in top Chinese hospitals, has handled over one million cases and has been recognized as a model project by the National Health Commission in 2024 and the Cyberspace Administration of China.

Internationally, we also provide healthcare solutions for families in the U.S., combining AI capabilities with over a million real clinical cases for continuous improvement. In China, at Henan Children’s Hospital, APUS models handle complex medical imaging across surgical, internal medicine, and neurological domains.

In creative industries, we partnered with Singapore’s Changxin Media for AI screenwriting and content generation. AI is already capable of generating key film segments and handling scene-level content, showing strong practical progress.

2. Model is Data, Data is Model

A key trend in AI is that “model is data, and data is model.” Traditional systems rely on databases, but AI may eventually integrate and process data autonomously, potentially replacing conventional databases.

Currently, AI cannot fully reconstruct and process data independently but has shown potential for autonomous data structuring. Our agricultural large model, for instance, can recognize 340,000 plants and 1,500 types of pests. By integrating real-time application data and AI processing, we provide global services, supporting both overseas individual users and domestic agricultural authorities.

3. Don’t Block the Main Lane of Large Models

Foreign companies like OpenAI and Gemini, and domestic companies like DeepSeek and APUS itself, dominate the AI landscape. In August 2023, we shifted focus from general models to applications, recognizing that general-purpose large models are no longer suitable for small companies.

We emphasize: when building applications and ecosystems, don’t block the main lane of large models. Otherwise, your painstakingly built data, model, and technology barriers could be easily overtaken.

For example, APUS initially developed a product for generating global e-commerce content using our own models. However, as models like Google’s Veo3 advanced, we pivoted to integrating these superior third-party models with our own Agents and workflows, achieving better user services.

This approach led to a revamped intelligent ad generation system. Previously, global marketing required 60 designers; now, eight designers manage the system, while AI generates thousands of videos daily.

Our principle: build high barriers with your own data, business models, and commercial logic. These are your real competitive walls in the AI era. Proprietary small models can secure efficiency and confidentiality while enabling participation in AI-driven markets.

4. Global AI Market Development is Uneven

The global AI market is highly uneven. The primary engines are the U.S. and China. While China may lag slightly in general model training, it has massive industrial datasets and rich application scenarios. The U.S. remains ahead in algorithmic innovation and thinking.

Other countries’ AI development is largely negligible or reproductions of U.S. and Chinese models.

APUS sees opportunities in both C-end (consumer) and B/G-end (enterprise/government) markets. Last year, we proposed the concept of sovereign large models, helping countries along the Belt and Road, particularly in Africa, build their own AI models. We are now collaborating with South Africa, Ghana, and Uganda on sovereign models for customs, agriculture, finance, and more.

For example, in Ghana, we applied AI models to gold mining, improving recovery rates from ~80% to 92%, translating into potential growth exceeding $1 trillion globally.

We believe AI will not only enhance productivity but also reshape production relations, trade, and global economic patterns. Through AI, we aim to guide this change toward stability, peace, and development, contributing to global AI evolution.

(Source: Chuangyebang)