MiniMax ARR Surpasses $800M, B2B Revenue Share Rises to 80%

In August 2026, MiniMax disclosed that its ARR had exceeded $800 million, up from just $150 million in February, expanding about 5.3x in six months. This growth is directly reflected in revenue: H1 2026 revenue reached $116.6 million, up 283.1% YoY, surpassing the full-year 2025 revenue of $79.04 million in just half a year. Q2 revenue grew 81.8% QoQ, indicating that growth momentum remains strong.

The change in revenue structure is even more critical. In H1 2026, B2B revenue share reached 63.4%, further rising to 80% by August, compared to only about 30% for the full year 2025. Specifically, revenue from the open platform and other AI enterprise services was $73.93 million, surging 703.1% YoY, with its share of total revenue rising from 32.8% in 2025 to 63.4%. Approximately three-quarters of new revenue came from B2B, indicating that the growth focus has shifted from C-end products to enterprises and developers.

The C-end business has not contracted; AI-native product revenue reached $42.64 million, up 100.9% YoY, mainly due to increased user engagement and commercialization of products like Hailuo AI. Both business lines are growing, but B2B is growing faster and has become the main source of new revenue.

The rise of the B2B business is closely related to the proliferation of Agent workflows. According to MiniMax CEO Yan Junjie in a conference call, token consumption in July reached 20x January's level. This growth is not only due to an increase in user numbers but also to higher usage intensity per user—Agent interactions with AI consume far more tokens than human chats. By May, the platform's enterprise customers and developers surpassed 2 million, 10 times the number at the end of last year.

The release of the M3 model was a key catalyst. This model has the ability to access information and operate interfaces, enabling developers to integrate it into real business scenarios with confidence. After the M3 release, existing customers continued to expand usage, and new customers accelerated their adoption. Additionally, the open-source video generation model H3, released in July, received over 24 million downloads within three weeks, spawning more than 300 derivative models, which drove developers to switch to paid API calls and further boosted official service usage.

MiniMax stated in the conference call that H3's performance changed the landscape where advanced models belong to big companies and follow a closed path. The open-source strategy not only expanded technical influence but also achieved commercial monetization through cloud service conversion, forming a positive cycle between open source and commercialization.

MiniMax's strongest M3 model is priced at $1.2 per million tokens, significantly lower than the median of $2.2 for comparable models. The company explains that this price is not purely a business strategy but because the actual compute cost for inference is lower than expected. This pricing strategy aims to reduce the unit cost of intelligence, allowing customers to run more calls and complex workflows with the same budget, thereby encouraging them to embed the model into production.

MiniMax emphasizes that high intelligence and cost-effectiveness are not mutually exclusive. The company's goal is not to trade off between intelligence and cost, but to reduce the unit cost of intelligence while driving both commercial growth and technological iteration. Commercially, more calls bring more revenue; technically, the same compute can support more post-training and experiments, accelerating model improvement and forming a positive cycle.

This cycle relies on infrastructure support. MiniMax is one of the first independent large model companies in China to establish large-scale infrastructure, with an ETTR (Effective Training Time Ratio) of 97%. Self-built compute allows for mixed scheduling of training and inference resources, using idle compute during inference troughs for reinforcement learning training, thereby reducing inference costs and improving efficiency.

Amid rapid revenue growth, financial efficiency has improved. In H1 2026, gross profit reached $20.81 million, up 464.8% YoY, with gross margin rising from 12.1% in the same period last year to 17.9%, mainly due to improved infrastructure efficiency. Sales and distribution expenses decreased 17.9% YoY to $26.97 million, reflecting reduced promotional spending due to organic user growth.

R&D investment remains high, with H1 R&D spending reaching $296.9 million, up 138.8% YoY, primarily for foundation models, multimodal capabilities, and training. Although the absolute scale has expanded, the growth rate is lower than revenue growth (283.1%), indicating relatively improved R&D efficiency. However, high R&D spending and low pricing strategy mean the company is still in a high-investment phase, with profitability pressure remaining.

In the conference call, MiniMax judged that the next phase of competition will focus on technology path selection and the efficiency of converting compute into intelligence. The company proposed the direction of 'Minimize the Inference Cost, Maximize the Intelligence,' which aligns with its current strategy.

MiniMax's growth case shows that the evaluation dimensions for large model companies are expanding. Parameter scale and benchmarks remain important, but the intelligence produced per unit cost, the number of tokens processed per unit of compute, and the stability of large-scale training and inference are becoming new key indicators. Agents and real production environments demand higher model capabilities, where every step must be accurate enough, otherwise the entire workflow may fail.

MiniMax's B2B transformation and open-source strategy provide a differentiated path for independent large model companies. By attracting developers through open source and then monetizing through API calls, while expanding market share with low pricing, this model has been initially validated with H3 and M3. However, sustaining growth will depend on model iteration speed, infrastructure efficiency, and the competitive landscape.

Currently, MiniMax's ARR growth and token consumption data point to strong demand, but high R&D spending and low gross margins remain challenges. In the future, the company needs to further optimize its cost structure while maintaining technological leadership to achieve sustainable profitability.