Summary of Key Points
MiniMax’s revenue in the first half of 2026 soared by nearly three times, reaching $120 million, which is more than 1.5 times the annual revenue of 2025. However, the net loss still amounted to three times the revenue ($358 million), although the loss decreased by 11% year-on-year. The company has shifted its focus from serving individual users (ToC) to corporate clients (ToB), accounting for 80% of its business, with overseas revenue accounting for 60% of total sales. The annual recurring revenue (ARR) exceeded $800 million, indicating accelerated commercialization. MiniMax is prioritizing investments in text-based models and adapting to domestic chips to reduce costs. The long-term competitive focus lies in the choice of technical approaches and the definition of problems to be solved.
1. Revenue: Rapid Growth and Shift from Individual to Corporate Clients
MiniMax’s revenue increased significantly in the first half of the year, nearly tripling compared to the same period last year and even exceeding half of the annual revenue in 2025. The most important change is the source of revenue: 70% of the revenue previously came from individual users (selling AI tools to individuals), while now 80% comes from corporate clients (providing AI services to companies, such as offering open platforms for them to utilize AI interfaces). Corporate clients are more stable in their payments, making “open platforms + enterprise services” the primary source of revenue, which has increased by more than seven times year-on-year. Additionally, the overseas market contributed 60.8% of the revenue, showing success in international business.
2. Losses: More Revenue, but Still Not Enough to Cover Costs
Despite the rapid revenue growth, the company incurred a loss of $358 million in the first half of the year, three times its revenue. This represents a 11% reduction in losses compared to the same period last year, which is a positive development. The reason for the high losses is that large-scale AI companies are still in the “money-burning” phase: they need to train models (which requires substantial computing power), purchase computing resources (similar to the costs of electricity and equipment in a factory), and expand their product offerings. These investments have not yet generated returns, but as the number of clients increases, the costs will be spread out, potentially leading to profitability in the future.
3. ARR Exceeds $800 Million: Accelerated Commercialization
The CEO mentioned that the ARR exceeded $800 million in August. This figure represents the stable annual revenue that can be expected based on the current business scale. For example, if a company signs a long-term contract with MiniMax and makes regular payments each month, this amount will contribute to the annual ARR. The rapid increase in ARR indicates that more companies are willing to use MiniMax’s services on a long-term basis, suggesting accelerated commercialization and a secure source of revenue in the future.
4. Computing Power: Cost Savings and Optimization are Key
Computing power is the “lifeline” for large-scale AI companies (just like kitchen equipment for a restaurant). MiniMax is investing most of its computing resources in text-based models (four times more than in video models) because text applications are more mature and have a larger user base. The company is also working to adapt to domestic chips, with domestic computing clusters expected to be launched soon, which will reduce costs and reduce reliance on imported chips. Additionally, MiniMax aims to reduce the “inference cost” (the computational cost of AI answering questions) by one-third. Lowering costs will allow for price reductions, attracting more users, and more users will further reduce costs, creating a positive cycle that will improve the gross profit margin.
5. Long-Term Competition: It’s About Choosing the Right Direction
The CEO believes that the competition among large-scale AI companies will not be about who has the most computing power or users, but about who can “define the right problems” and “choose the right technical approach.” Some companies focus on text-based models, while others work on multimodal models (text + images + video); some use one training method, while others use another. These choices will lead companies down different paths, and ultimately determine which ones can establish a foothold in the market. MiniMax has chosen high-value areas such as multimodal productivity and cybersecurity, as these tasks are more valuable and easier to monetize.
By breaking down the information in this way, even non-financial professionals can easily understand MiniMax’s current situation and future direction.