AI technology application and investment opportunities｜Linear View

In the current era of rapid development of AI technology, we should not only focus on the breakthroughs of the technology itself, but also think about how to apply these technologies to create real business value. I would like to share with you some of my views on the understanding of AI in the new era, the current application status, the advantages of open source large models, the comparison between China and the United States, and investment opportunities.

Double breakthrough in computing power and cost

The most notable feature of the new era of AI is the dual breakthroughs in computing power and cost. The improvement in computing power is mainly led by the United States, while the significant reduction in cost is an important contribution of China in the wave of open source big models.

In the past, the narrative logic of AI development focused on breaking through the upper limit of capabilities, but not on cost issues. However, the emergence of open source big models has completely changed this logic. It has reduced costs by 90%, not just a small drop of 30%-40%, which has enabled AI applications to move from "highbrow" to "every household".

Although the United States is still ahead in AI basic innovation, China has huge opportunities in application implementation and cost optimization. The characteristics of Chinese companies, namely "short, flat, fast, high efficiency and low cost", will be fully utilized in this round of AI application wave.

From "Why should I use it" to "Why can't I use it"

Enterprises' attitudes towards AI are undergoing a fundamental change. Last year, most enterprises were still discussing "why to use AI", but this year the focus has shifted to "why it can't be used". This change in concept is significant, indicating that enterprises have recognized the value of AI, but still face some challenges in practical application.

Key challenges include:

• Computing power bottleneck: high-end GPU cards are difficult to obtain and expensive. Although open source models have reduced costs by 90%, the cost of building a large model server is still very high. This has also prompted companies to consider using domestic computing power, driving the prosperity of the domestic computing power market.

• Data quality issues: Many traditional enterprises have accumulated a large amount of data, but the data quality varies. Data cleaning and organization are crucial, and simply purchasing hardware cannot solve application problems.

• Technical threshold: Different AI applications have different technical thresholds. Training a custom model has the highest threshold, while building a knowledge base combined with a large model has a relatively low threshold. The latter allows companies to combine internal data with large model capabilities, enabling AI to answer questions based on the company's internal situation.

Open source big model: the perfect balance between performance and cost

The emergence of open source big models brings several key advantages:

• Balance between performance and cost: The performance reaches 90-95% of the top closed-source models, but the cost is only 5-10%. This cost-effectiveness makes AI applications affordable for more enterprises.

• Open source and transparency: A fully open source strategy (including model weights, source code, and technical papers) allows companies to build their own models and ensure data security without having to worry about data leakage. Open source is divided into multiple levels: open source data, open source model weights, open source code, and open source papers. The more complete the open source, the more it can promote technological development.

• Engineering optimization: Although there is not much original innovation, the existing concepts are optimized to the extreme and effectively integrated, making the model effect far exceed expectations. This kind of engineering capability is precisely the strength of Chinese companies.

• Communication strategy: Careful design of the timing and method of release can quickly gain global attention for the company/product. In contrast, the open source strategy of some large companies is not thorough enough. They only open parameters but not code and detailed papers. What they learn is the "form" but not the "spirit".

Comparison of AI capabilities between China and the United States: Each has its own strengths

China and the United States have different development paths in the field of AI:

1. The United States: focuses on breakthroughs in cutting-edge innovation, promotes basic capabilities at all costs, and pursues breakthrough exploration of capabilities.

2. China: Focusing on cost optimization and cost-effectiveness, and striving to make AI capabilities accessible to the general public.

In terms of specific abilities:

• The United States leads in computing power and algorithms
• China has advantages in data processing and hardware supply chain

• In terms of talent, the United States has more top talents, while China has more advantages in engineering and product talents.

It is worth noting that starting this year, more high-end talents have considered returning to China to start businesses, which is a positive signal.

AI Applications: From Consultant to Executor

The efficiency improvement brought by AI applications often exceeds expectations. For example, in terms of information collection, the traditional method may take two days, but using AI Agent (tested on manus.im experience) can complete the same work in just 10 minutes, which is a significant improvement in efficiency.

More importantly, AI is changing from a simple advisory role to an executor role. In the past, large language models were mainly "language in, language out", providing advice like a consultant, but it was difficult to perform specific tasks. Today's AI can break down complex tasks into multiple steps, and call different models or tools to complete each step, and finally deliver complete results.

This capability transforms AI from a mere conversational assistant to a true task executor, greatly expanding its application scenarios.

Robotics: The next investment hotspot

In the field of AI large models, the competition landscape has become relatively stable, and opportunities for new entrants are limited. However, there are still huge opportunities in the field of robotics, which can be divided into three categories:

• Mobile robots: focus on spatial movement, such as robot dogs. The technology is relatively mature and is expected to be put into practical use this year.

• Interactive robots: focus on the interaction and transformation of robots with the physical world. This field is more technically difficult and requires processing multimodal data such as vision, text, and action. The training complexity is much higher than that of language models. It is expected to be implemented in the industrial field within 2-3 years.

• Household robots: They are used for family and elderly care. These robots are still in the demonstration stage and are expected to take 3-5 years to be put into use. More general products may take more than 5 years.

The technical challenge in the field of robotics is that it must respect the laws of the physical world and have a higher tolerance rate. The "hallucination" problem of language models can lead to serious consequences in the field of robotics, so higher requirements are placed on safety and accuracy.

Entrepreneurs in the AI ​​Era: New Qualities and New Opportunities

The requirements for entrepreneurs in the AI ​​era have changed:

• Technical understanding ability: Although AI tools have lowered the threshold for technical understanding, entrepreneurs still need to have basic technical literacy. Fortunately, AI tools can now be used to understand technical papers, lowering the learning threshold.

• Product polishing ability: The ability to transform technology into high-quality products is crucial. China has a group of "craftsmen" who are extremely persistent in their products, which is a valuable talent resource.

• Communication ability: In the era of information explosion, it is equally important to tell a good product story and design an effective communication path.

• Financing ability: AI projects usually require a lot of financial support, and entrepreneurs need to have financing capabilities.

• Younger: The main force of AI startups is those born in the 1990s and 2000s. They are not burdened by traditional experience and are more receptive to new ideas. Looking back at the founders of Hangzhou's "Six Little Dragons", most of them were born in the 1980s and 1990s when they started their businesses. If we now invest in entrepreneurs who may shine in the next 7-10 years, we should focus on those born in the 1990s and 2000s.

Investment Opportunities Summary

Opportunities in the early stage technology investment sector are mainly concentrated in:

• Domestic computing power:Domestic GPU and computing power solutions are expected to achieve a double breakthrough in cost-effectiveness and performance within a few years.
• Reasoning technology and models: AI is shifting from training-centric to reasoning-centric, and there are abundant opportunities in this area.
• Robot applications: especially in industry, home and some special scenarios (such as elderly care).
• AI+Science  (bio, material, chemical, etc) .
• Smart hardware: a hardware product that is truly intelligent, rather than a simple conceptual product.

• Autonomous driving: The new technology wave is driving the accelerated implementation of autonomous driving.

In general, China has unique advantages in AI application and cost optimization, and is expected to achieve greater success in productization and popularization in the future. For investors, paying attention to projects that can deeply integrate AI technology with actual scenarios and bring significant efficiency improvements will be the key to grasping this round of AI wave.