Summary of Key Points
Kimi (the Dark Side of the Moon) has completed four rounds of financing in half a year, raising a total of $3.9 billion. Its pre-investment valuation has soared from $4.3 billion to $31.5 billion. Behind this is not just a simple valuation bubble; it reflects a fundamental shift in the underlying logic of the large-model industry: these models have become capital-intensive businesses that require continuous investment to maintain and upgrade their computational capabilities. Kimi has also demonstrated a viable commercial path with annual recurring revenue (ARR) of $300 million, attracting industrial capital and sovereign funds. These long-term investors are interested in the strategic value of AI as the next-generation infrastructure, rather than short-term returns. However, Kimi faces challenges: the higher the valuation, the higher the threshold for its next round of financing, and it must quickly increase its ARR to a level that can support a market value of tens of billions of dollars; otherwise, it may encounter a funding crisis.
Detailed Analysis
1. Large Models as Capital-Intensive Businesses
Large-model companies are not like internet platforms (such as WeChat), where serving more users hardly increases costs. They are more akin to building power plants: they need to continuously invest in equipment, maintenance, and electricity generation.
- Costs: High-end GPUs (costing hundreds of thousands each), expensive computing power (OpenAI spent $10.5 billion on computing resources for Microsoft in one year), salaries for research staff (with many doctors earning millions per year), and data cleaning and training processes.
- Examples: OpenAI’s revenue in 2025 was $13 billion, but its total costs were $34 billion, resulting in a loss of $20.9 billion; Meta and Google doubled their capital expenditures in 2026, all focused on AI servers and data centers.
- Kimi’s Expenses: Kimi uses the money to purchase GPUs, build data centers, pay for electricity, and support its team, all to prepare for the next round of model iteration. These investments do not disappear but become tangible hardware assets. If stopped, it would fall behind competitors.
2. $300 Million in ARR is More Valuable than a $31.5 Billion Valuation: Commercialization is Finally Taking Shape
The $31.5 billion valuation is more of an aspiration, while the $300 million in ARR represents actual revenue. This is the first time a domestic large model has supported its valuation with real earnings, rather than relying on a “Chinese version of GPT” narrative.
- What is ARR?: Annual recurring revenue, which is money that can be consistently generated each year (e.g., from API calls or subscriptions). Kimi’s ARR increased from $100 million to $300 million in just three months, with 70% coming from API developers. The growth in overseas users by 400% indicates that there is a real demand for its products.
- Comparison with Other Companies: Zhipu generates revenue through local government and enterprise deployments, while MiniMax relies on consumer subscriptions. Kimi follows the Anthropic approach (focusing on APIs and international markets), which has greater potential for scalability (Anthropics’ ARR increased from $1 billion to $47 billion in just over a year).
- Industry Divide: Domestic large models are now divided into two categories: those still competing on technical parameters and those with stable revenue. Among the companies with revenue, Kimi’s approach to international APIs is most similar to leading global players.
3. Change in Investors: From VC to Industrial Capital
Previously, AI financing mainly came from VCs (venture capital), but now it has shifted to industrial capital (such as China Mobile and Meituan), state-owned funds, and sovereign funds (e.g., Temasek in Singapore and MGFX in Abu Dhabi).
- Reason for the Change: VCs seek short-term returns (exit within 3-5 years), while AI requires long-term investment. Industrial capital and sovereign funds are interested in gaining a foothold in the “next-generation digital infrastructure.” AI, like electricity or 5G, is crucial for controlling the future of the digital economy.
- Examples: Tencent’s investment in Kimi is not about quick profits but about securing a position in the underlying industry chain; national funds invested $50 billion in DeepSeek, treating AI as a strategic asset for the country.
- Valuation Logic: Valuations now consider both financial returns and industrial value as well as geopolitical strategy. The $31.5 billion valuation is not the result of market bidding but a joint “pricing” by multiple strategic investors.
4. Contrary to Internet Logic: More Users Mean Higher Costs
The logic of internet platforms is that marginal costs approach zero: serving 1,000 or 1 million people incurs similar costs. However, AI is the opposite:
- Cost Formula: Total computing cost = Number of active users × Number of calls per user × Cost per unit of computing power. The more users and frequent calls, the higher the cost.
- Example: ChatGPT’s initial daily cost was $7 million; now that its user base has increased significantly, the daily cost is even higher. AI is more like a “public utility”: the more people use it, the more resources are consumed, and the greater the potential losses.
- Consequences: Leading AI companies must continuously raise funds to survive. As revenue grows, so do capital expenditures (for example, Kimi may need to spend $2 for every additional dollar in revenue).
5. Kimi’s Deadline: Higher Valuations Bring Greater Pressure
The $31.5 billion pre-investment valuation is not a safety net but a timer:
- Next Round Financing Threshold: A valuation of at least $40 billion is required to attract investors for the next round, but there are few institutions that can afford this.
- Time Window: Yang Zhilin needs to increase Kimi’s ARR from $300 million to $3 billion or even $1 billion before exhausting the funds from the primary market. Only then will it be able to support a market value of tens of billions of dollars upon listing.
- Two Possible Outcomes: If successful, listing will mark its “crowning”; if not, it will result in failure (the secondary market focuses on short-term metrics and does not wait for long-term growth).
In Conclusion
Kimi’s financing story represents a turning point in the AI industry, from focusing on narratives to demonstrating tangible capabilities. This capability includes not only technology but also the ability to continuously invest in computing power and generate revenue. In this arms race of computational resources, the company that can sustain itself will become the “infrastructure operator” of the next-generation digital world.