MiniMax Nearly Quadruples Revenue in First Post-IPO Interim Results, as Enterprise API Demand Explodes 703%

MiniMax Group, the Shanghai-based AI company that went public in Hong Kong in January, has reported its first interim results since listing — and the headline number is a shock to the upside. Revenue for the six months ended June 30, 2026 reached US$116.6 million, up 283.1% year over year from US$30.4 million. That single half-year already exceeds the company’s entire 2025 revenue of US$79.0 million, and it lands with the timing of a statement: China’s listed AI challengers are no longer pre-revenue science projects.

The results, released after the Hong Kong market close on August 26, 2026, were accompanied by a management commentary that reads more like an infrastructure manifesto than a typical earnings release. Dr. Yan Junjie, MiniMax’s co-founder and CEO, framed the company’s thesis in one line: “Intelligence can scale almost without limit; energy and compute cannot.”

The Number That Matters: 703% Enterprise Growth

The most striking figure in the release is buried in the segment breakdown. Revenue from MiniMax’s Open Platform and other AI-based enterprise services grew 703.1% year over year, from US$9.2 million to US$73.9 million. That segment now accounts for 63.4% of total revenue, up from just 30.3% in the first half of 2025 — a complete inversion of the company’s revenue mix in twelve months.

A year ago, MiniMax looked primarily like a consumer AI company, with Hailuo AI (its video-generation product) driving the story. Today it looks like an inference infrastructure business. The drivers, per the release: growth in paying individual developers and enterprise customers, rising API call volumes, and — notably — “the rapid adoption of our Token Plan,” the subscription-style inference pricing scheme that bundles token consumption for developers building on MiniMax models.

Revenue from AI-native products (the consumer side, led by Hailuo) grew a more modest 100.9%, from US$21.2 million to US$42.6 million, on higher engagement and willingness to pay.

Token Consumption Grew 20x in Six Months

The single most revealing data point in the entire release isn’t a financial metric at all. By July 2026, Yan said, token consumption on MiniMax had grown to 20 times its January level.

That is the kind of demand curve usually associated with the infrastructure buildouts of the hyperscalers, not a company that IPO’d seven months ago. It also explains the company’s strategic emphasis: MiniMax positions inference efficiency not as a cost-cutting measure but as the gating factor for continued model scaling — “Minimize the Cost, Maximize the Intelligence,” in Yan’s phrasing, is what makes the company’s stated mission of “Intelligence with Everyone” economically possible.

During the reporting period the company released MiniMax M3, its core model upgrade focused on coding, agentic workflows, and professional work. Shortly after the period closed, it followed with MiniMax H3, released with open weights for commercial video generation — a continuation of the open-weight strategy that has differentiated MiniMax from most Western frontier labs.

Margins Improving, Losses Still Wide

The unit economics are moving in the right direction, if from a low base. Gross profit improved 464.8% year over year to US$20.8 million, with gross margin rising from 12.1% to 17.9% — attributed primarily to improving infrastructure efficiency. Selling expenses actually fell 17.9% to US$27.0 million as the company leaned into organic user growth rather than paid promotion, and administrative expenses fell from 48.8% of revenue to 25.9%.

Research and development spending rose 138.8% to US$296.9 million — a large number relative to revenue, reflecting training compute costs for model iteration, but one that management was quick to note grows “significantly lower” than the 283% revenue growth rate.

The adjusted net loss was US$293.0 million for the half (versus US$138.7 million a year earlier), with the adjusted loss ratio narrowing from roughly 456% of revenue to roughly 251%. The company is still deeply unprofitable on any accounting basis. But it is not constrained: the cash balance stood at US$1,322.8 million as of June 30 — up from US$1,050.3 million at the end of 2025, courtesy of IPO proceeds — giving MiniMax one of the largest war chests among China’s listed AI companies relative to its revenue base.

Context: Behind Pace for the Year, Ahead of the Story

For all the superlatives, the South China Morning Post’s analysis struck a cautionary note: H1 revenue of US$116.6 million represents only about 32% of the US$363.77 million analysts expect for full-year 2026. Revenue growth decelerating into a back-half guidance gap is a familiar trap for high-multiple AI listings, and MiniMax trades at a valuation that already prices in steep compounding. The company did not provide revised full-year guidance in the release.

The bigger picture is competitive. MiniMax operates in a Chinese AI market where DeepSeek’s open-weight releases reset pricing expectations, Alibaba’s Qwen family distributes at enormous scale, and ByteDance’s Seed models compete directly in video generation. Globally, its Token Plan and open-weight releases put it in indirect competition with OpenAI, Anthropic, and Google for developer workloads. The 703% enterprise growth suggests that strategy — cheap, efficient, open inference — is finding real purchase with developers, with MiniMax’s models and products now serving more than 300 million users across over 200 countries and regions, plus over one million enterprises and developers across more than 100 countries.

Whether “Minimize the Cost, Maximize the Intelligence” can carry MiniMax from US$116.6 million a half to the US$5.8 billion annual revenue analysts project for 2030 is the multi-year question. But as first interim results go, this one makes the bull case concrete: demand is exploding, the mix is shifting toward enterprise, margins are climbing, and the cash to endure the losses is sitting in the bank.