Hello! I'm your financial analysis assistant. The news about Zhipu AI receiving a huge amount of funding is incredibly informative, not only mentioning a massive $5 billion investment but also revealing the underlying logic of the current AI industry's transition from a model of "burning money to attract users" to one focused on "core technology and infrastructure development."
To help you understand this easily, I will first summarize the key points in one sentence and then break down the information into five key dimensions for a detailed analysis.
📝 Summary of Key Points
Zhipu AI has once again received a significant $5 billion investment in just half a month. This money will be primarily used for three purposes: developing the next generation of "fully self-trained" models, building its own computing infrastructure, and optimizing the efficiency of both software and hardware. This marks a fundamental shift in the competitive landscape of the AI industry, where the focus is no longer on subsidizing end-users to gain market share but on establishing long-term technological barriers and cost advantages through control over underlying computing power and automated training capabilities.
---
🔍 In-Depth Analysis: Five Key Dimensions to Understand Zhipu AI's Strategy
1. Where the Money Is Being Spent: Betting on "Fully Self-Trained" Models
The most notable concept in this funding round is "fully self-trained" models.
- Simple Explanation: Traditional AI training is like a teacher teaching a student, where human engineers provide data, set rules, and check the results. The goal of "fully self-trained" models is to enable AI to evolve on its own. In other words, Model A trains Model B, which then iterates to create Model C, forming a recursive cycle that does not require frequent human intervention.
- Current Status and Challenges: No company in the world has yet achieved true "full self-training." The current state is more like "assisted driving," where humans still need to monitor the process. Zhipu plans to invest billions to overcome this challenge, allowing AI to decide when to stop training and correct errors on its own.
- Why It's Important: If successful, this would significantly reduce the labor costs associated with training new models and speed up the iteration process. Although humans will still be needed in the short term, this is seen as a necessary step towards achieving superintelligence.
2. Computing Power: From "Buying Services" to "Building Factories"
The news highlights that Zhipu has made "computing infrastructure" one of its three main investment areas, which is crucial.
- Shift in Role: Previously, model companies (like Zhipu) were mainly consumers of computing power, renting from cloud service providers. Now, Zhipu is becoming both the owner and builder of its own computing resources.
- Specific Actions:
- Building a Data Center: Zhipu has already constructed a 1GW (gigawatt) data center using domestically produced AI chips.
- Combining Hardware and Software: It has acquired a software company (Zhongke Jiahe) to optimize code and algorithms, making the same domestic chips more efficient and cost-effective.
- Strategic Intent:
- Ensuring Supply: With the global shortage of computing power, having its own resources ensures a steady supply for model training.
- Cost Savings: By managing its own computing power and optimizing software, Zhipu has increased the "computing power multiplier" (the revenue generated per dollar invested) by 14 times. This means more value can be created with the same amount of money.
- Competitive Advantage: Computing power is the "oil" of the AI era; controlling the supply gives Zhipu a significant advantage in the long run.
3. Industry Trend Indicator: From "Burn Money to Attract Users" to "Core Infrastructure Development"
Zhipu's approach reflects a major shift in the way the AI industry spends its money.
- Past Approach (First Half of the Game): Similar to the mobile internet era, companies heavily subsidized end-users, offering free services and promotions to gain user numbers. However, this strategy led to users leaving as subsidies stopped, and the resulting data did not form a sustainable barrier.
- Current Approach (Second Half of the Game):
- Focusing on Business (B2B) Over Consumers (B2C): Companies realize that businesses (such as those using AI for programming or office work) are more willing to pay for AI services. Therefore, funds are being invested in practical B2B applications rather than consumer subsidies.
- Investing in Infrastructure: Money is no longer spent on advertising but on purchasing chips, building data centers, and automating training processes.
- Investor Focus: Investors now value a company's "long-term vision" and its ability to implement technology, rather than short-term user growth. They are willing to provide substantial funding to leading companies because they believe it will lead to genuine technological breakthroughs and future cash flows, rather than creating bubbles.
4. Is It a Bubble or Real Value? Capital is Betting on a "Long-Term Cycle"
There are concerns about whether the AI industry is a bubble, but this funding round signals a more rational approach.
- Reasons It Doesn't Seem Like a Bubble:
- Practical Use of Funds: Money is being directed towards tangible assets and technologies like computing power and chips, rather than marketing or inefficient subsidies.
- Resource Lock-UP: Leading companies are taking advantage of easy financing to convert capital into computing power, electricity, and talent, preparing for potential funding challenges in the future.
- New Evaluation Criteria: The focus is no longer on the number of users but on whether a company can establish a positive cycle of "high investment → new models → more revenue → further investment." Investors carefully assess how each dollar invested leads to effective model iteration and real revenue generation.
5. Zhipu's Evolution: From a Model Company to a Comprehensive Entity
Finally, let's look at the qualitative changes within Zhipu itself.
- Previous Role: Zhipu was a pure model company, with a simple business model: consuming computing power to generate tokens (APIs) for profit.
- Current Role: Zhipu is evolving into a comprehensive entity that combines computing power, engineering, and modeling:
- Computing Power: It owns its own data center and controls the supply of computing resources.
- Engineering: It optimizes software to enhance chip performance.
- Modeling: It uses self-training techniques to involve models in data generation and validation, improving the efficiency of computing power utilization.
- Competitive Advantage: Zhipu believes that true competitiveness lies in the product of "computing power supply × engineering efficiency × training methods."
- Future Vision: Zhipu may even start providing computing power services to others, transforming from a typical AI startup into a key player in the AI infrastructure sector.
💡 Insights for the General Public
1. AI Competition Enters a More Complex Phase: The focus is no longer on app downloads but on the solidity of underlying technologies and the autonomy of computing power.
2. Opportunities for Domestic Chips: Zhipu's extensive use of domestic chips and performance optimizations indicate that domestic AI hardware is moving from being "usable" to "highly efficient," making the related industry chain worth paying attention to.
3. Rational View of Funding: Huge investments do not necessarily indicate a bubble. The key is whether the money is being invested in areas that create long-term value, such as computing power and automation. If a company is still subsidizing users, that's a cause for concern; if it is building data centers and developing self-training technologies, it is investing in the foundation for future growth.
In summary, the AI industry is entering a more sophisticated phase, where competition revolves around core technologies and infrastructure. The adoption of domestic chips and the strategic focus on long-term value suggest that the industry is moving in a positive direction.