MiniMax’s first half-year financial report since its listing highlighted more than headline revenue growth. The company reported $117 million in revenue for the first half, already above its full-year 2025 total. At the same time, research and development spending reached $297 million, about 2.55 times revenue, while adjusted net loss came to $293 million, up 111.2% year over year.
The biggest change was in the composition of that revenue. Open platform and other AI enterprise services brought in $73.93 million, making up 63.4% of total revenue. AI-native products generated $42.64 million, accounting for 36.6%.
AI-native products kept growing, but no longer led the business
MiniMax first became widely known through Hailuo AI and a set of AI-native products spanning text, voice, and video generation. Those offerings gave the company a clear consumer-facing identity.
In this report, AI-native product revenue still increased 100.9% year over year to $42.64 million. But its weight in the business dropped sharply. In the same period last year, AI-native products accounted for 69.7% of total revenue. This year, that share fell to 36.6%.
By contrast, open platform and other AI enterprise services surged 703.1%, rising from $9.2 million to $73.93 million. Their share of revenue expanded from 30.3% to 63.4%.
- AI-native products: $42.64 million, up 100.9%, revenue share down from 69.7% to 36.6%
- Open platform and enterprise services: $73.93 million, up 703.1%, revenue share up from 30.3% to 63.4%
With those lines crossing, MiniMax’s center of gravity has moved away from consumer products and toward businesses where enterprises and developers pay for model capabilities.
Enterprise demand and platform usage are taking the lead
Over one year, B2B and open platform revenue rose from 30.3% to 63.4% of the total. The original article argues that this shift matters more than the 703.1% growth rate itself.
Consumer users buy a product. Enterprises buy a set of capabilities. Developers call a model. Agents consume tokens. Once a model is inserted into customer service, coding, marketing, knowledge bases, search, and agent workflows, it stops being just an AI app and turns into a component inside business operations.
As framed in the source article, MiniMax is starting to gain something it had little of before: other companies using its models to create value. That changes the revenue logic. Income no longer depends only on how well MiniMax builds its own products, but also on how much others need what its models can do.
Hailuo requires users to come to MiniMax. The open platform lets MiniMax go into other people’s products. That move, from pulling users in to embedding outward, marks a change in the company’s role.
The model itself is turning into a business
The article draws a distinction across three layers. Hailuo sells a product. APIs sell capability. Enterprise services sell results produced after the model enters a business workflow.
The difference, in the source author’s telling, is that a product resembles a one-off transaction, capability is a continuing supply, and outcomes point toward a profit-sharing logic. MiniMax is moving from building applications itself to letting others build applications on top of its models.
Rising API call volume suggests the model is becoming part of other people’s production stack. A growing enterprise client base means the model is entering real workflows. Higher agentic workloads indicate a move from answering questions to completing tasks.
Once a model is embedded in someone else’s operations, its commercial value is no longer measured only by how often an app is opened. Pricing starts to reflect how deeply the business depends on it. The original piece argues that the ceiling for a model company depends on how many people are willing to build their own business on that capability. That, it says, is the position MiniMax is trying to reach.
Revenue is rising, but losses and R&D spending remain high
Commercialization is advancing, but profitability is still distant. The report laid out three headline figures:
- Revenue: $117 million, already above full-year 2025
- R&D spending: $297 million, around 2.55 times revenue
- Adjusted net loss: $293 million, up 111.2% year over year
The source article notes that large-model economics differ from those of traditional internet products. Model training costs money. Inference costs money. More usage means more compute consumption, and more enterprise customers mean higher infrastructure expense. Revenue growth does not automatically become profit growth in this business. Commercial traction can accelerate while costs rise at the same time.
Under that logic, the harder question for MiniMax is no longer whether it has customers, but whether customer growth can eventually outpace model costs.
Gross margin improved from 12.1% to 17.9%
There was also a more constructive signal in the numbers. In the first half, cost of sales grew 258.1%, while revenue grew 283.1%. Revenue increased faster than costs.
Gross profit rose from $3.69 million to $20.81 million, and gross margin improved from 12.1% to 17.9%. The level is still modest, but the direction changed.
The article describes this as the next stage of large-model commercialization: moving from proving demand to searching for scale economics. Strong model performance can answer the first question. The next one is whether the same $1 of revenue can be produced more cheaply over time.
The source points to training efficiency, inference efficiency, model routing, chip adaptation, system scheduling, caching, compression, and token efficiency at the product layer as the levers that now matter in cost control. MiniMax said in the report that it plans to improve training and inference efficiency through full-stack co-design across models, infrastructure, systems, and products.
In the article’s phrasing, model rankings decide who gets attention, while the cost curve decides who lasts longer.
More than 60% of revenue came from outside mainland China
Geographic revenue mix was another standout feature. In the first half, MiniMax generated $70.83 million outside mainland China, representing 60.8% of total revenue. Revenue from mainland China was $45.75 million, or 39.2%. Its products and services now cover more than 230 countries and regions.
The source article says that is a notable structure for a Chinese large-model company. AI products can be distributed globally by design. One model API can serve developers in multiple countries, and AI-native products can directly reach overseas users.
Unlike traditional internet companies expanding abroad through channels, localization, supply chains, and offline infrastructure, the article characterizes the AI route as one where models are developed in China, products are distributed globally, and APIs plug directly into overseas developer ecosystems.
At the same time, overseas revenue introduces extra complexity in exchange rates, compliance, payments, and local markets. Even so, the current revenue mix already shows what the article calls a pattern of “developed in China, monetized globally.”
After the valuation surge, the market is looking for profits
The article says MiniMax’s market value jumped after listing and has since fallen back to around HK$100 billion. In the author’s view, the market has not abandoned the company, but wants to see something beyond the broad AI growth story.
Revenue has already demonstrated part of MiniMax’s commercialization ability. B2B revenue making up more than 60% shows that enterprises and developers are willing to use its models. Its listing and follow-on financing have also improved cash reserves, according to the article.
The next question is narrower and harder: can this business make money? The article argues that stronger models do not necessarily bring higher profit, and more customers do not necessarily produce better cash flow. What matters is whether revenue growth can gradually separate from cost growth.
A 17.9% gross margin is a clear improvement from 12.1%, but still far from the profit structure of a mature software business. The article sums up MiniMax’s current position this way: the commercialization curve is running, while the profitability curve is still near the starting point.
The next hurdle for Chinese large-model companies
The piece closes by placing MiniMax within the broader Chinese large-model sector. In its framing, the first stage of competition centered on model parameters, benchmarks, inference ability, and release speed. The second stage focused on applications: who could turn a model into a product people actually use.
According to the article, MiniMax has already cleared both stages. Hailuo demonstrated consumer product capability. The open platform demonstrated demand from enterprises and developers. What comes next is a different contest, built around customers, channels, tokens, inference cost, business model, and globalization.
The source concludes that model capability is no longer the finish line but the starting point. A strong model earns entry. A product with users earns the second card. Enterprises and developers willing to pay earn the third. The last card is profit.
Looking back at the report, the article highlights three figures together:
- $117 million in revenue, showing model capability is translating into scalable commercial demand
- 63.4% of revenue from B2B and open platform services, showing a shift from an AI product company to a model-capability company
- $293 million in adjusted net loss, showing the commercialization process is still unfinished
The article ends with a blunt assessment: the hardest step for a large-model company is not building the model itself, but turning intelligence into a business. MiniMax has shown that people are willing to buy. The next step is proving that the business can support itself.
The original article was published via the WeChat public account “版面之外” and written by “版君.”