MiniMax Group Inc. (HKEX: 00100) published its annual results for 2025 on March 2, 2026, marking its first set of financial statements following its IPO in Hong Kong in January 2026. The headline is enticing: Revenue of $79.0 million, a 158.9% year-over-year increase, surpassing Bloomberg's estimates ($71.39 million). Over 70% of the business is now international, and the company reports 236 million accumulated users.
In a market where many AI companies still depend on narratives, MiniMax is showing signs of commercial traction: "AI-native" products (Hailuo AI, MiniMax Agent, Talkie, Xingye) climbed to $53.1 million (+143.4%) and the Open Platform and enterprise services soared to $26.0 million (+197.8%), serving 214,000 enterprise customers and developers across more than 100 countries.
Now, the serious diagnosis begins where the press release turns uncomfortable: the jump in revenue coexists with a worsening loss adjustment and a staggering loss reported due to asset valuation. The business may be growing, but it has yet to prove it can be primarily self-financed through customers rather than relying on funding.
The Real Story Lies in Gross Margin and Sales Expense
The figure that piques my interest the most in these results isn't revenue growth, but the change in revenue quality. MiniMax reported a gross profit of $20.1 million, a leap of 437.2%, with a gross margin of 25.4% compared to 12.2% the previous year. The official explanation—improved model efficiency, system optimizations, and infrastructure allocation—points to what really drives the economy of a model-based company: cost per unit of useful computation delivered.
This margin improvement indicates that the team is beginning to understand the operational "game": without control over inference costs and infrastructure discipline, growth merely amplifies losses. Moving from 12% to 25% isn't stability but provides direction.
The second structural detail is the adjustment of the acquisition machine. Sales and distribution expenses fell 40.3%, from $87.0 million to $51.9 million. The company attributes this decline to organic growth from referrals and reduced promotions. Translated: the product is generating some traction without incurring as much “advertising toll.”
In consumer AI businesses, this point can often be fragile: demand may be high, but retention and reactivation costs can destroy LTV (Customer Lifetime Value). The cut in commercial expenses is a positive sign only if it doesn’t compromise future growth; here, the opposite happens: they cut costs yet still achieve 159% growth. This suggests a network effect or product positioning with enough perceived value to sustain consumption.
Still, a gross margin of 25.4% indicates the business is far from a “pure software” profile. Foundational AI resembles a capital-intensive operation requiring continuous adjustment rather than a traditional SaaS model. The internal technical question for any CFO is whether that margin can scale to levels that absorb R&D and structural costs without needing ongoing subsidies.
International Growth and Platform: Traction Exists, but the Model Remains Expensive
MiniMax emphasizes that over 70% of its revenue is generated from overseas, across more than 200 countries and regions. This early international expansion can be advantageous: it mitigates dependence on a single market and allows for capturing demand where willingness to pay is higher. It is also a sign of genuine commercial ambition, not just a laboratory experiment.
Revenue growth indicates two distinct engines. On one hand, consumer "AI-native" products with direct monetization. On the other, an Open Platform that sells capacity to enterprises and developers. Practically speaking, this is an attempt to build a "dual flywheel": consumption for distribution and brand; platform for more predictable revenues and recurring usage.
The company also reports an annual recurring revenue surpassing $150 million as of February 2026. That figure, while not broken down, suggests monetization is shifting towards contracts or repeat usage patterns. In model-based companies, this recurrence is the closest indicator of an “established product”: less dependence on sporadic launches and more continuity through integration.
Having said that, the cost of sustaining technological advancements remains a significant burden. Research and development costs rose to $252.8 million (+33.8%). While it grew more slowly than revenue, it still vastly exceeds the size of the revenue. This is not a moral critique, but a structural reading: MiniMax is still operating in “capacity-building mode” rather than “profitability-extraction mode.”
CEO Dr. Yan Junjie stated that in 2025, they built "full-mode" R&D capacity and are transitioning from a model enterprise to a platform firm. The strategy makes sense: a platform captures distribution and technical lock-in through APIs, tools, and community. The risk is that “platform” becomes a label to justify internal complexity without corresponding margins and cash flows.
Adjusted Losses Show No Improvement and Reported Losses Distort Diagnosis
Two MiniMax entities coexist within the same financial statements. The first is operational: growing, improving gross margins, reducing commercial expenses, and seemingly gaining efficiency from scale. The second is financial-accounting: reporting a net loss of $1.87 billion, compared to $465.2 million in 2024, primarily explained by unrealized losses on financial liabilities.
This second component may be “non-operational,” but it has a real effect: it introduces volatility and can affect market perception, future funding capability, and strategic flexibility. For a management team, the key is to separate the signal from the noise without denying it: the business may be improving, but the financial instrument causing these variations is also part of the incentive system and cost of capital.
Looking at the operational side without embellishment, the significant figure is the adjusted net loss of $250.9 million, slightly worse than the $244.2 million from 2024. Adjusted excludes stock compensation expenses, valuation losses, and IPO-related costs. In other words: even after cleaning up the accounting and events, the company continues to consume capital.
Here, the improvement in gross margin and the reduction in commercial expenses should start reflecting in the adjusted results. That has not yet occurred. A plausible explanation is that the expansion in R&D and growth in structure consume the progress. Indeed, administrative expenses increased by 155.9% to $36.8 million, due to higher personnel costs, stock-based payments, and listing-related expenses.
Operationally, this illustrates a classic tension in AI companies: the team is attempting to run two races at once. One, maintaining model competitiveness across multiple modalities (language, video, voice, music). Another, building a global commercial engine with enterprise support, compliance, billing, and customer care. Each layer adds fixed cost and coordination.
The discipline here isn’t a “cost-saving plan,” but rather a decision of focus: which modalities truly monetize, which products drive distribution and which bring in cash, and which parts of the research translate into efficiency of cost per token and paid retention.
Post-IPO Cash Buys Time, Not a Sustainable Model
MiniMax closed 2025 with $1.050.3 billion in cash and equivalents (including equivalents, financial assets, restricted cash, and fixed deposits), up from $880.6 million the previous year. Following an IPO of HK$4.8 billion in January 2026, the company has a cushion to act.
However, this cushion has a cold interpretation: it buys time to find the correct unit economics. If the adjusted loss hovers around $250 million per year, the cash can finance several product cycles, but does not eliminate the need to convert growth into margin.
The product narrative is compelling: models M2, M2.1, M2-her, the video model Hailuo 2.3, Speech 2.6, and Music 2.0/2.5; and as of February 2026, M2.5 with efficiency improvements in programming and a strong increase in token consumption. All this indicates the capacity for iteration.
But in the modeling business, iteration isn’t enough. The indicator determining the winner isn’t the number of launches but whether each generation reduces marginal costs, increases willingness to pay, and sustains a distribution channel independent of commercial expenditure. MiniMax begins to show pieces of that puzzle: rising gross margins and international sales. What’s missing is the closing piece: that the adjusted results start to visibly improve.
The Technical Direction is Correct, but the Ultimate Test is Accounting
MiniMax achieved a real operational advance in 2025: strong growth, doubled gross margin, lower commercial expenditure, and international expansion. It also made clear that it will continue investing heavily in R&D to sustain multimodality and product offerings.
As the Editor-in-Chief, my reading is pragmatic: the company is buying market share and capacity in a growth-focused market, but still doesn’t demonstrate that its revenues can sustain its structure without relying on funding. The market can tolerate early-stage losses if it sees a clear path towards margin and cash flow.
The next phase will be less about model announcements and more about basic accounting: consistently rising gross margins, operating expenses growing slower than revenue, and adjusted losses decreasing quarter by quarter. That will be the proof that the platform is maturing as a business and not merely as a laboratory.
