MiniMax's first semi-annual financial report after going public shows that the most interesting number is not its revenue.
Revenue for the first half of the year was USD 117 million, already surpassing the full-year figure for 2025. However, R&D investment was 2.5 times the revenue.
Even more interesting is the revenue structure. Revenue from the open platform and other AI enterprise services reached USD 73.93 million, accounting for 63.4% of total revenue; revenue from AI-native products was USD 42.64 million, with its share dropping to 36.6%.
A large model company that was first known to the public for its consumer products has changed the protagonist of its revenue.
MiniMax is changing the way it operates.
1. Conch AI Steps Back
MiniMax was first recognized by users thanks to Conch AI. Text, voice, and video generation, along with a series of AI-native products, gave MiniMax a very clear consumer-facing image. The business feedback for this type of venture is also straightforward: if users are willing to pay, the product is valuable.
In this financial report, revenue from AI-native products still grew by 100.9%, reaching USD 42.64 million. However, it has stepped back.
During the same period last year, revenue from AI-native products accounted for 69.7% of MiniMax's total revenue. This year, it's down to 36.6%.
On the other hand, revenue from the open platform and other AI enterprise services grew by 703.1%, from USD 9.2 million to USD 73.93 million.
AI-native products: USD 42.64 million (+100.9%), revenue share: 69.7% → 36.6%
Open platform & enterprise services: USD 73.93 million (+703.1%), revenue share: 30.3% → 63.4%
With one curve rising and the other falling, after they crossed, MiniMax's commercial focus is no longer on the side of Conch AI.
Users are still paying for products, while enterprises and developers are starting to pay for model capabilities. These are two completely different businesses, with the latter's ceiling significantly higher.
2. The B-Side Takes Over
Within a year, the share of B-side and open platform revenue has risen from 30.3% to 63.4%. The significance of this change is more worth discussing than the 703% growth rate.
Consumers buy a product. Enterprises buy a set of capabilities. Developers call models. Agents call Tokens.
When models are integrated into customer service, coding, marketing, knowledge bases, search, and Agent workflows, they are no longer just "an AI product" but become a pipeline within business processes. At this point, what clients buy is hard to define as just "software."
MiniMax is starting to gain something it rarely had before: others are using its models to create value. This means its revenue no longer depends on how good it is at making products, but on how much others need its capabilities.
This might be the most important change in this financial report.
Conch AI needs users to come to MiniMax. The open platform allows MiniMax to enter others' products. From "pulling people in" to "sending itself out."
With this step, MiniMax's identity begins to change.
3. The Model Starts Making Money
Conch sells products. API sells capabilities. Enterprise services sell the results generated after models are integrated into business.
The difference between these three layers lies in: products are one-time transactions, capabilities are continuous supply, and results follow a profit-sharing logic.
MiniMax is shifting from developing its own applications to letting others use its models to build applications.
This is also the truly noteworthy aspect of the rapid growth in open platform revenue.
An increase in API calls means models are becoming production tools for others. An increase in enterprise clients means models are entering real workflows. An increase in Agentic workloads means models are moving from answering questions to completing tasks.
Once a model is embedded in others' business processes, its commercial value is no longer calculated based on "open times," but priced based on "dependency."
How many times a user opens Conch daily is important. How many Tokens an enterprise calls daily is also important. But how much business developers build upon your model—that's the most critical thing.
The ceiling of a model company depends on how many people are willing to base their businesses on its capabilities.
The position MiniMax is vying for is to move down from the application layer and become the foundation for others' businesses.
4. Not Enough Money
Commercialization is up and running, but profitability is not.
Revenue: USD 117 million (already surpassing full-year 2025)
R&D investment: USD 297 million (approximately 2.55 times revenue)
Adjusted net loss: USD 293 million (year-on-year +111.2%)
Seeing these numbers, it's easy to wonder: they sold a lot, so why are they still burning money?
The answer lies in the cost structure of large models. For traditional internet products, after reaching a certain scale, the marginal cost of adding a new user can decline rapidly. Large models don't have it that easy. Model training costs money, inference costs money, the more users call, the greater the computational consumption; the more enterprise clients, the higher the infrastructure costs.
Therefore, revenue growth for large model companies never automatically translates into profit growth. Commercialization is running fast, but costs are also running.
What MiniMax is really facing now is no longer "whether there are clients." It's another, harder question:
Can client growth ultimately outpace model costs? The former is a sales problem, the latter is a survival problem.
5. Gross Margin Leads the Way
The good news is also here.
In the first half of the year, MiniMax's cost of sales grew by 258.1%, while revenue grew by 283.1%. Revenue is growing faster than costs. Gross profit increased from USD 3.69 million to USD 20.81 million, and gross margin improved from 12.1% to 17.9%.
The numbers are still low, but the direction is beginning to change.
This means MiniMax is beginning to enter the second checkpoint of large model commercialization: moving from proving demand to seeking economies of scale.
Strong model capabilities only solve the first problem. The next problem is: can the same dollar of revenue be produced cheaper and cheaper? The first problem tests technology; the second tests engineering, systems, and operations.
What matters behind this is no longer just model parameters. Training efficiency, inference efficiency, model routing, chip adaptation, system scheduling, caching, compression, and Token usage efficiency at the product level will all enter the cost ledger.
MiniMax mentions in its financial report that it will improve training and inference efficiency through full-stack co-design of models, infrastructure, systems, and products.
This direction is crucial. Because the real cost competition in the large model industry is shifting from whose model is stronger to who can deliver the same intelligence at a lower cost.
Model leaderboards determine who gets attention. Cost curves determine who survives longer.
6. Global Revenue
MiniMax has another distinct feature: its commercialization has not been locked into the Chinese market from the start.
In the first half of the year, revenue from outside mainland China was USD 70.83 million, accounting for 60.8% of total revenue. Revenue from mainland China was USD 45.75 million, accounting for 39.2%. Currently, MiniMax's products and services cover over 230 countries and regions.
This structure is interesting among Chinese large model companies.
AI products inherently have global distribution capabilities. A model API can serve developers in different countries; an AI-native product can directly reach overseas users.
Traditional internet companies going global require channels, localization, supply chains, and a lot of offline infrastructure. The globalization path for AI companies is shorter: models are developed in China, products are distributed globally, APIs are directly integrated into overseas developer ecosystems. You can essentially have cross-border business without spending a dime on plane tickets.
MiniMax has already embarked on this path. Of course, global revenue also brings new complexities like exchange rates, compliance, payments, and local markets.
But at least from the revenue structure, it already possesses a rare characteristic: developed in China, monetized globally.
If this path continues to be successful, its commercial ceiling will also open up accordingly.
7. After a Trillion
The capital market has given MiniMax sufficiently high expectations.
After going public, the company's market capitalization once surged, but has now retreated to around a trillion Hong Kong dollars. The market is not bearish on it; it's that after looking at it, they want to see something else.
This means that simply telling the story of "the AI space is huge" can no longer easily become a new valuation story.
Revenue has proven part of its commercialization capability. The B-side revenue share exceeding 60% also proves that enterprises and developers are willing to use MiniMax's models. The IPO and subsequent financing have also significantly improved the company's cash reserves.
The next card only leaves one question: can this business make money?
This question is much harder than growing revenue. Because a stronger model doesn't necessarily mean higher profits; more clients don't necessarily mean better cash flow. What truly determines whether this business can be established is whether a gap can gradually open between revenue growth and cost growth.
A 17.9% gross margin is already much better than 12.1%. But there's still a long way to go compared to the profit structure of mature software businesses.
MiniMax is now in a very special position: the commercialization curve is already up and running, while the profit curve is still lying flat at the origin. The distance between these two lines is the gap it needs to fill next.
8. One Final Hurdle Remaining
Placing MiniMax back into the Chinese large model industry makes the changes even clearer.
In the first stage, everyone competed on models. Model parameters, benchmarks, reasoning capabilities, and release speed were the focus of competition. In the second stage, everyone competed on applications. Whoever could turn models into products actually used by people gained the first batch of real users.
MiniMax has already completed these two steps. Conch proved its consumer product capability; the open platform proved enterprise and developer demand.
Next, what large model companies truly need to compete on is changing: clients, channels, Tokens, inference costs, business models, globalization.
Model capability is shifting from the end goal to the starting point.
A strong enough model gets you the entry ticket; a product that people use gets you the second card; enterprises and developers willing to pay gets you the third card. Finally, you need profit.
MiniMax already has the first three cards. The last one is still on the way.
9. MiniMax, Where Are You Now?
Looking back at this financial report, three numbers are most worth considering together.
USD 117 million revenue: Model capabilities are beginning to form scaled commercial demand.
63.4% B-side & open platform revenue share: Shifting from an AI product company to a model capability company.
USD 293 million adjusted net loss: Reminding everyone that commercialization is far from over.
Putting these three numbers together reveals today's MiniMax.
It has already passed the stage of "does anyone use it?" and is currently passing the stage of "is anyone willing to pay for the model?" The question it truly needs to answer next becomes very specific: can the money others are willing to pay for it ultimately sustain itself?
This might also be a signal that the Chinese large model industry is truly entering its next phase. Revenue can get higher and higher, clients can become more and more numerous, but in the end, everything returns to the most basic income statement:
How much money is actually retained for every additional unit of revenue generated.
A Note from "Beyond the Page":
The hardest step for a large model company has never been developing the model. The test is how to turn intelligence itself into a business.
MiniMax has already proven that people are willing to buy.
The next step is to prove it can sustain itself.
Between someone buying and being able to sustain itself lies the entire deep-water zone of large model industry commercialization.
This article is from the WeChat public account "Beyond the Page," author: Ban Jun
