MiniMax: Soaring Revenue, a $1.87B Loss, and a Global AI Plan

HONG KONG – March 02, 2026 – AI foundation model company MiniMax Group today presented a financial portrait that perfectly captures the frenetic and often paradoxical nature of the artificial intelligence industry. The company announced a staggering 158.9% year-over-year revenue increase to US$79.0 million for 2025, yet simultaneously reported a colossal net loss of US$1.87 billion.

This juxtaposition of explosive growth and deep losses highlights a critical narrative in the world of high-growth tech: the divergence between operational success and the complex realities of pre-IPO accounting. For MiniMax, the story is not one of operational failure, but of a valuation soaring so high that it creates massive, non-cash accounting charges, all while the company aggressively expands its technological capabilities and global footprint.

The Unicorn's Accounting Paradox

The headline figure of a US$1.87 billion loss can be misleading without context. A deeper dive into MiniMax's financials reveals that the vast majority of this loss—approximately US$1.59 billion—is a non-cash “fair value loss on financial liabilities.” This accounting charge stems from the company's convertible redeemable preferred shares, financial instruments common in venture-backed startups. Under IFRS accounting standards, as the company's valuation increases, the liability associated with these shares also increases, resulting in a paper loss on the income statement.

In essence, MiniMax is reporting a massive loss because it has become significantly more valuable. Investors and analysts often look past these figures to metrics that better reflect a company's operational health. MiniMax's “adjusted net loss,” which excludes these non-cash items, was US$250.9 million, a slight increase from US$244.2 million in 2024. While still a substantial figure, it paints a far more stable picture of the company's underlying burn rate relative to its explosive growth.

More telling are the operational metrics. Gross profit skyrocketed by 437.2% to US$20.1 million, and the gross profit margin more than doubled from 12.2% to 25.4%. The company attributed this to improved model efficiency and better infrastructure allocation. Furthermore, MiniMax demonstrated fiscal discipline by cutting its selling and distribution expenses by 40.3%, relying instead on organic growth and user referrals. The company's cash balance also grew to over US$1 billion, shoring up its financial position for the intense R&D battles ahead.

From Models to a Platform: MiniMax's Strategic Pivot

While the financials tell one story, the company's strategic direction tells another. Co-founder and CEO Dr. Yan Junjie emphasized a fundamental shift in his comments: MiniMax is evolving from a “large-model company into a platform company for the AI era.” This strategy is built on advancing its full-modality AI capabilities, which span language, video, speech, and music.

The company has been on a technological offensive, releasing a suite of powerful models. Its M2.5 language model, released in February 2026, has reportedly achieved globally leading performance in productivity tasks like coding and tool use. The model's adoption has been rapid, with average daily token consumption for the M2 series growing six-fold since December 2025. This focus on practical, high-value applications is a core part of the company's strategy to move AI from a mere tool to a collaborative, “colleague-level partner.”

On the multimodal front, MiniMax has made significant strides. Its video model, Hailuo 2.3, has helped creators generate over 600 million videos, while its Speech 2.6 model supports over 40 languages with ultra-low latency. These powerful models are the engine for the company's product portfolio, which includes consumer apps like Talkie and Xingye, and the enterprise-focused MiniMax Agent AI-native workspace.

This transition to a platform is not just an external strategy. Internally, MiniMax has deployed “agent interns” that now support nearly 90% of its employees in tasks from software development to marketing, treating its own organization as a testing ground for AI-native workflows. By productizing these capabilities, MiniMax aims to build a robust ecosystem for its 214,000 enterprise customers and developers.

Forging a Global AI Footprint

Perhaps the most remarkable aspect of MiniMax's 2025 results is its international success. The company revealed that over 70% of its revenue was generated from markets outside of China, serving over 236 million users across more than 200 countries. This global reach is a significant differentiator in a competitive landscape dominated by US-based giants like OpenAI and Google.

MiniMax's international strategy appears to be multi-pronged. It involves offering powerful, open-source-based models that can serve as lower-cost alternatives to proprietary systems. This approach is gaining traction, with its models being integrated into major global cloud platforms like Google Vertex AI and Microsoft's Azure AI Foundry. The integration of its M2.5 model into the popular productivity app Notion as its first open-source option is a testament to this growing adoption.

This global expansion is underpinned by impressive efficiency gains. While R&D expenses grew by 33.8% to US$252.8 million—driven largely by the immense cloud computing costs required for model training—this figure was significantly outpaced by the 158.9% revenue growth. The company has actively worked to lower its costs, with the unit token inference cost for its M2 series models falling by over 50% in early 2026, demonstrating a maturing ability to scale its intelligence delivery in a financially sustainable manner. As MiniMax prepares to release its next-generation M3 model, its ability to balance intense R&D investment with a clear path to commercialization on a global scale will be its defining challenge.