Key Points Summary
More than seven months after its initial public offering, MiniMax has released its first half-year financial report for 2026: revenue in the first half of the year amounted to $117 million (a year-on-year increase of 283%), exceeding the entire revenue of 2025. The most significant change is the reversal of the revenue structure. Previously, the company's revenue mainly came from C-side AI applications (such as Talkie for companionship and Hailuo for video generation), but now B-side open platforms and enterprise services account for 63.4% (a year-on-year increase of 703%). Annual recurring revenue (ARR) has exceeded $800 million, with B2B services accounting for 80% of this figure. The number of enterprise customers and developers has increased tenfold compared to the end of last year. However, the company is still in the red (adjusted net loss of $293 million, a year-on-year increase of 111%), and its research and development (R&D) expenses are 2.5 times its revenue. In the future, MiniMax needs to balance revenue growth with R&D and computing power investment to address issues related to Token profitability and customer retention.
Detailed Analysis
1. Reversal of Revenue Structure: Moving from C-side Applications to B-side Services
MiniMax previously thrived on C-side products like Talkie (AI companionship) and Hailuo (video generation), with users subscribing to or purchasing additional services. These C-side applications contributed nearly 70% of the company's revenue in the first half of last year. However, the C-side market was highly competitive, with users leaving easily, and the anticipated “super AI applications” did not materialize.
In the first half of this year, there was a complete shift: revenue from B-side open platforms and enterprise services reached $73.9 million (a year-on-year increase of 703%), accounting for 63.4% of total revenue for the first time. Management reported that as of August, annual recurring revenue from B2B customers exceeded $800 million, and the number of enterprise customers and developers had more than doubled (tenfold compared to the end of last year). Additionally, overseas revenue accounted for 60.8%, covering over 230 countries, indicating that the company's expansion is not limited to the domestic market.
In simple terms, MiniMax is now selling “AI tools” to businesses and developers, generating more and more stable revenue.
2. Tokens as a New Growth Driver: Token Usage Outpacing User Growth
What are Tokens? They can be understood as the “number of times an AI model is used.” Each time a model is called to perform a task (such as answering questions, writing code, or generating videos), it counts as a Token consumption.
MiniMax’s growth this year is not due to more people using AI, but rather an increase in the frequency of use per user. In July, Token consumption was 20 times higher than in January, indicating a rapid trend. This is because AI has entered the Agent phase: previously, chatbots only needed a few Model calls to answer a question, but now Agents can break down complex tasks into multiple steps (search → planning → execution → verification), each of which requires a Model call. For example, when an AI is asked to write a market report, it first searches for data, analyzes it, and then writes the content—each step counts as a Token consumption.
To adapt to this trend, MiniMax has been iterating on its models: in June, it enhanced the M3 model for code writing and Agent capabilities; in July, it made the H3 model open-source (free for developers to use), with 24 million downloads in three weeks, leading to the creation of over 300 new models. The founder emphasizes that the value of a model lies not in whether it is open-source, but in its effectiveness, cost, and iteration speed. This creates a positive cycle: stronger models lead to more tasks from businesses, which in turn result in more Token calls and increased revenue.
3. Rapid Growth but Still Losses: R&D Expenses Outpacing Revenue
Despite a 283% increase in revenue, MiniMax is still in the red, with an adjusted net loss of $293 million (up from $139 million last year, a 111% increase). R&D expenses amounted to $297 million (up from $124 million last year), 2.5 times the revenue.
The reason for the losses is the high cost of developing large models: hiring teams, purchasing GPUs for computing power, and building infrastructure. However, there are also positive aspects: the growth rate of R&D expenses (138.8%) is slower than revenue growth, and sales expenses have decreased by 17.9% (without significant marketing spending). The gross margin has increased from 12.1% to 17.9% due to improved infrastructure efficiency.
However, a 17.9% gross margin is still low compared to traditional software, where the cost of serving one additional customer is almost zero. Large models, on the other hand, consume resources (GPUs, electricity, bandwidth) with each use, similar to the cost of selling fuel at a gas station. Therefore, the company is working to reduce costs: the efficiency of text model computing power has increased threefold in the past two months, and the next generation, M3.1, aims to reduce inference costs to one-third of the initial level. The founder believes that lowering costs will allow for price reductions to attract more users and increase margins.
4. Intense Competition Requires Continuous Innovation
The large model industry is highly competitive, with companies like OpenAI, Google, Anthropic (internationally), DeepSeek, ByteDance, and Alibaba (domestically) updating their models every few months, competing in areas ranging from text to code, Agents, and multimodal (text + images + video). If MiniMax stops investing, it will quickly fall behind.
For example, during the development of M3, they realized the evaluation system was inadequate and rebuilt the underlying infrastructure, which has been proven effective in H3. M3.1 will further optimize these systems. B-side customers are more selective, evaluating APIs based on performance, cost, and stability; if a competitor offers better options, they may switch. To sustain investment, MiniMax has sufficient cash reserves—$1.32 billion as of the end of June, with another $2 billion raised in July, totaling over $3 billion—to maintain its competitive edge.
5. Two Critical Challenges for the Future: Token Profitability and Customer Retention
While MiniMax has found a profitable direction with B-side Tokens, it must address two key issues:
- Token Profitability: The cost per Token needs to decrease significantly to attract more customers while still making a profit (for example, if costs are reduced by one-third, a 25% price cut would result in a 12% profit margin).
- Customer Retention: Competitors are constantly improving, so MiniMax must maintain its model effectiveness, cost efficiency, and stability; otherwise, Tokens may be lost to better alternatives (for instance, if other models are cheaper and more stable).
Only by resolving these issues can the 283% growth rate be sustained; otherwise, the success may be short-lived.
In Summary
MiniMax has shifted from focusing on C-side applications to providing B-side services, relying on increased Token usage for growth. To continue this trajectory, it needs to balance cost reduction and profit generation with maintaining the competitiveness of its models.
(The company’s stock price rose 3.6% the day after the financial report was released, with its market value returning to HK$11 billion, indicating market approval of this strategic shift.)