MiniMax-W interim revenue jumps 283% as token usage scales 20x

MiniMax-W reported interim revenue of $112 million, up 283 percent year over year and 17 percent above consensus, as token consumption scaled 20-fold since January.

"The results reflect strong demand for model APIs and token plans," CMBI analysts said in a note, raising the target price to HKD580 from HKD570 while keeping a Buy rating.

Adjusted net loss widened 111 percent to $293 million, above the $200 million market forecast, which the broker attributed to higher cloud service costs tied to model training. Gross profit rose 465 percent to $21 million, lifting gross margin 580 basis points to 17.9 percent from 12.1 percent.

Annualized recurring revenue reached $800 million in August, up from $150 million in February and $400 million in May, driven by the launch of the M3 and H3 models. Citi also raised its target price to HKD576, saying the year-end ARR target is achievable. The stock traded down 0.7 percent on the day.

The company's open platform and other AI enterprise services generated about $74 million in the first half, up 703 percent year over year, with the mix rising to 63 percent from 30 percent a year earlier. AI-native product revenue rose 101 percent to about $43 million. Overseas markets contributed more than 60 percent of total revenue.

Enterprise users topped 2 million, up 10-fold year over year, and business-to-business clients accounted for 80 percent of ARR versus about 30 percent a year ago. July token usage reached 20 times January levels, with revenue up 81.8 percent versus January.

Research and development expense rose 138 percent to about $300 million, though management noted the growth lagged the 283 percent revenue increase. Selling and distribution expense fell 18 percent to about $30 million on organic user growth and lower promotional spending. Cash and cash equivalents exceeded $3 billion after a July placement.

Management said gross margin will continue to improve in the second half and beyond 2026, with domestic chips set to account for a larger share of models in the fourth quarter, lowering per-token costs. The company is developing M3 Pro, a 3-trillion-parameter model, and said release cycles for new models will shorten as the M and H pipelines mature.

The widening adjusted loss signals heavy investment in foundation models and inference infrastructure, while the ARR trajectory points to accelerating monetization. Investors will watch the fourth-quarter domestic chip ramp and the M3 Pro launch for the next margin inflection.

This article is for informational purposes only and does not constitute investment advice.