As the first public AI companies report their earnings, we’re getting insights into what kind of businesses AI model companies really are.
MiniMax, the Shanghai-based AI startup that listed on the Hong Kong Stock Exchange on January 9, 2026, has published its first earnings report as a public company — and the numbers paint a picture of a business scaling fast, improving its unit economics, but still burning through cash as it races to compete at the frontier of AI development.
Revenue Growth Outpaces Expectations
MiniMax reported full-year 2025 revenue of $79 million, up 158.9% year over year from $30.5 million in 2024. That kind of growth rate is eye-catching by any standard, but what’s particularly notable is the geographic composition: more than 70% of revenue came from international markets. For a Chinese AI company, that’s a remarkable achievement. The company’s products and services are now deployed across more than 200 countries and regions, and it has accumulated 236 million cumulative users and 214,000 enterprise customers and developers worldwide.
Revenue breaks down into two segments: AI-native products (consumer-facing apps like Hailuo AI, Talkie, and its new MiniMax Agent workspace) contributed $53.1 million, or 67% of the total, growing 143% year over year. The Open Platform and enterprise services segment grew even faster — up 198% to $26 million — as more developers and businesses plugged into MiniMax’s model APIs.
Margins Are the Real Story
If there’s one figure that deserves a closer look, it’s the gross margin expansion. MiniMax improved its gross margin from 12.2% in 2024 to 25.4% in 2025 — more than doubling in a single year. Gross profit grew 437% to $20.1 million, far outpacing revenue growth.
The company credits this to improved model and system efficiency and better infrastructure allocation — in other words, the same compute is generating more revenue. This is the leverage that AI model companies have been promising investors, and MiniMax is beginning to demonstrate it in practice. The trend echoes what has been reported across the broader AI industry: as models mature and inference becomes more efficient, the economics of serving AI at scale improve materially.
The Real Loss — and the Accounting One
Here’s where the earnings report requires some unpacking. MiniMax posted a headline net loss of $1.87 billion for 2025, up from $465 million in 2024 — a number that looks alarming at first glance. But $1.59 billion of that loss came from a single line item: fair value losses on financial liabilities, specifically the remeasurement of convertible preferred shares as MiniMax’s valuation surged ahead of its IPO. This is a non-cash accounting item that reflects how much more valuable the company became, not money leaving the door.
Strip that out — along with $24 million in share-based compensation and $6.9 million in one-time listing expenses — and the adjusted net loss was $250.9 million, barely changed from $244.2 million in 2024. In other words, the underlying cash burn was essentially flat year over year, even as the company more than doubled its revenue. That’s a meaningful signal about operational leverage.
Research and development spending grew 33.8% to $252.8 million, driven largely by increased cloud computing costs for model training. But notably, R&D growth ran at less than a quarter the pace of revenue growth — another early sign that the model is becoming more efficient to build and run.
What MiniMax Actually Makes
It’s worth pausing on what MiniMax is, exactly, because it sits at an interesting intersection. Unlike pure-play API providers, MiniMax runs consumer products at scale. Hailuo AI, its video generation platform, had helped creators generate more than 600 million videos cumulatively by year-end. Its speech model covered more than 200 countries, with users generating over 200 million hours of audio in total. Music models, an AI agent workspace, and a social AI app called Talkie round out the portfolio.
On the model side, MiniMax released its M2 series in Q4 2025, claiming it was the first Chinese model on OpenRouter to exceed 50 billion tokens in daily consumption. Its latest model, M2.5, released in February 2026, set a new industry record on the SWE-Bench Verified coding benchmark and reduced costs dramatically — the company claims running an agent continuously for a full year costs just $10,000.
Going Concern, IPO, and What Comes Next
MiniMax’s balance sheet carries a technical going concern note: the company had net liabilities of $2.65 billion at year-end. But this is almost entirely attributable to the preferred shares that automatically converted to equity at IPO. With $1.05 billion in cash and equivalents on hand as of December 31, 2025, and fresh IPO proceeds of roughly HK$5.3 billion (~$680 million) raised in January 2026, the company is well-capitalized for the near term.
Looking ahead, MiniMax is framing its next phase as a transition “from a large-model company into a platform company for the AI era.” The company is betting on three areas: AI coding agents (targeting L4-L5 intelligence levels), workplace productivity, and long-form multimodal content creation. The ambition is to be the infrastructure layer that other applications and businesses build on — a familiar aspiration in the AI industry, but one that MiniMax, with its unusual combination of consumer scale and model capability, is now publicly accountable for delivering.
For investors and analysts watching the AI sector, MiniMax’s results offer a useful early data point: AI model businesses can grow fast, can improve margins significantly as they scale, and can reach global audiences from day one. The question is whether they can get to profitability before the compute arms race demands another round of capital. At MiniMax’s current trajectory, the answer is not yet clear — but the direction of travel seems to be encouraging.