China's MiniMax Sees ARR Surge Over 50% in Early 2026, Driven by 'OpenClaw' AI Ecosystem
Chinese AI unicorn MiniMax is shifting from technology validation to large-scale monetization at a blistering pace, with its commercialization efforts far exceeding expectations in the first two months of 2026. The surge is primarily fueled by deep integration with the burgeoning "OpenClaw" developer ecosystem.
According to a recent Morgan Stanley research note, MiniMax’s annualized recurring revenue (ARR) skyrocketed from 100 million in December 2025 to 150 million by the end of February 2026, marking a more than 50% increase in just two months. The investment bank maintained its "Overweight" rating on the company, citing a powerful growth trajectory.
This rapid revenue growth was accompanied by a massive spike in user engagement. Token consumption for the company's M2 model exploded, increasing sixfold in February compared to December. This surge in demand has enabled significant economies of scale, slashing the per-token inference cost by over 50% during the same period.
The dramatic acceleration in both revenue and efficiency underscores the success of MiniMax’s strategy to build a robust open platform, capturing strong demand from a global developer base. The company is now leveraging this momentum to fund its next generational leap in model capability.
ARR Soars as Commercialization Accelerates
MiniMax's impressive commercial performance is driven by the synchronized growth of multiple business lines on its Open Platform. The company’s Coding Plan has become a significant contributor to open-platform revenue and has already achieved a positive gross margin. This tool not only generates revenue but also deepens ties with developers, promoting wider adoption of MiniMax's models across multimodal scenarios.
The core drivers for the strong token growth in January and February were applications built around OpenClaw, the Coding Plan, and other cloud APIs, which heavily utilize the M2.1 and the newer M2.5 models. While the company sees significant potential to increase the average revenue per user (ARPU) for its AI-native products, its current strategic priority is on expanding its user base to build a foundation for future monetization. In a strategic move, MiniMax has paused updates for its Hailuo video generation product after version 2.3 to concentrate all resources on developing a next-generation architecture focused on end-to-end output, multimodal input, and long-form video generation.
Sixfold Token Growth Drives Down Costs
The exponential increase in token usage illustrates strong market demand and has directly translated into substantial cost optimizations on the supply side. Morgan Stanley's report highlights that the more than 50% reduction in per-token inference cost for the M2 model was achieved primarily through algorithmic optimization rather than cheaper hardware.
The report notes that M2’s inference cost could theoretically be halved again, indicating significant runway for further efficiency gains. Critically, as token volume has scaled, the company has improved its computing power load balancing and utilization, which in turn lowers marginal costs. MiniMax's large and growing demand has also given it strong bargaining power with computing power suppliers, securing favorable pricing. Management expressed clear optimism about long-term gross margin improvement, supported by a combination of enhanced pricing power from more capable models, continuous optimization, and expanding scale.
M3 Model: A Generational Leap to 'Global Top-Tier'
MiniMax's long-term ambition is anchored in advancing the intelligence of its foundational models. The company's upcoming M3 series is being designed to compete in the top tier of global AI models. This marks a strategic shift from the M2 model, which was developed under resource constraints with a focus on balancing cost-effectiveness and speed. The M3 represents a generational leap in capability, built upon the company's accumulated talent, computing power, and data.
Management emphasized that the design of each model generation—including its features, target market size, and margin structure—is planned in advance, with the M3 model explicitly set to deliver higher gross margins than its predecessors. While pushing for state-of-the-art capability, MiniMax continues to view inference speed as a key competitive differentiator and will persist in finding the optimal balance between cost and speed. Furthermore, optimizations from its large language models (LLMs) are expected to create technical synergies that boost the efficiency of its multimodal models.
Redefining the Platform: Developer Ecosystem as Core Moat
MiniMax’s definition of a "platform company" diverges sharply from traditional internet platforms. According to the Morgan Stanley note, the company’s value proposition is not in controlling internet traffic or acting as a consumer gateway, but in pushing the frontiers of artificial intelligence. In MiniMax's view, true platform value emerges when new intelligent capabilities give rise to entirely new product categories.
The strong growth momentum in the company's coding products and agent pipelines is seen as early validation of this philosophy. To build user loyalty, MiniMax is focused on rapid and reliable model iteration, operating on the principle that developers will naturally gravitate towards platforms that demonstrate the fastest and most consistent technological progress. This leads to what Morgan Stanley identifies as MiniMax's core moat: a relentless race of "model evolution speed," where sustained technical leadership is the key to locking in the developer ecosystem.