5.3倍，六个月。
8月27日，杰富瑞（Jefferies）发布报告，将MINIMAX-W（00100.HK）目标价从506港元上调至533港元。上调的核心依据只有一条：8月年化收入（ARR）已突破8亿美元，高于该行此前5.8亿美元的预测。作为参照系：今年2月，另一家投行口径下的ARR还只是1.5亿美元——半年时间，约5.3倍。
增长从哪来：把曲线拆开看
拆开增速，驱动因素相当具体：7月Token消耗量较2026年初增长20倍；6月下旬M3上线后，推理速度从20 TPS提升到100 TPS，直接改善了模型可用性；7月文本模型使用量快速爬坡；8月H3发布又添了新的增量。换句话说，这不是单点爆发，而是"模型迭代→速度改善→用量放大"的链条在连续转动。
杰富瑞同时预期毛利率下半年改善、2027年仍有扩张空间，并把公司战略概括为"最大化智能、最小化成本"。即将推出的M3.1、M3 Pro与H3.1，被视作这条曲线的后续燃料。
冷水也要泼：这是投行的曲线，不是公司的报表
必须说清口径：8亿美元ARR不是MiniMax的官方审计披露，而是杰富瑞基于公开调用数据与渠道调研的推算。ARR本身是年化口径——把某个月的收入乘以12，它放大的是趋势，不是当期利润。公司确实已公布上市后首份中期业绩，但"投行推算"与"公司披露"之间，始终隔着一层需要自己验证的距离。
为什么这条曲线重要
港股AI此前缺的从来不是技术叙事，而是一条可以被逐月跟踪的增长曲线。MiniMax恰好补上了这个位置：模型迭代有节奏、调用量有数据、投行有跟踪报告。对二级市场而言，这是第一次可以用"成长股"的方法论去读一家中国大模型公司——而不是只靠"题材股"的想象力。
曲线是真的，口径是借来的。跟踪它，但别把它当成财务报表。
明天见。
5.3x, in six months.
On August 27, Jefferies raised its target price on MINIMAX-W (00100.HK) from HKD506 to HKD533. The justification came down to a single line: annualized revenue run-rate (ARR) topped $800 million in August, beating the bank's earlier $580 million forecast. For scale: last February, ARR stood at $150 million on another bank's estimate — roughly 5.3x in half a year.
Where the growth comes from
Break the curve down and the drivers are concrete: token consumption in July grew 20x since the start of 2026; after M3 launched in late June, inference speed jumped from 20 to 100 TPS, directly improving usability; text-model usage climbed steeply through July; and August's H3 release added a fresh increment. This is not a single spike — it is a chain of "model iteration → faster inference → more usage" turning continuously.
Jefferies expects gross margin to improve in the second half and expand further in 2027, summarizing the company's strategy as "maximize intelligence, minimize cost." Upcoming M3.1, M3 Pro and H3.1 are framed as the next fuel for the curve.
A bucket of cold water: the broker's curve, not the company's books
The caliber matters: the $800 million ARR is not an audited company disclosure — it is Jefferies' estimate from public usage data and channel checks. ARR is an annualized figure: one month's revenue multiplied by twelve. It amplifies the trend, not current profit. MiniMax has published its first interim results since listing, but between "broker estimate" and "company disclosure" remains a gap you have to verify yourself.
Why the curve matters
What Hong Kong AI lacked was never a technology narrative — it was a growth curve that could be tracked month by month. MiniMax fills that slot: models ship on rhythm, usage has numbers, brokers publish coverage. For public markets, this is the first time a Chinese frontier-model company can be read with a growth-stock playbook, rather than only a theme-stock imagination.
The curve is real; the caliber is borrowed. Track it — just don't mistake it for a financial statement.
See you tomorrow.
半年5.3倍的曲线足够漂亮，但它出自投行之手——跟踪它，别迷信它。
—— Dawn Vision编辑部
A 5.3x curve in six months is beautiful — but it is drawn by a broker. Track it; don't worship it.
— The Dawn Vision Editorial Desk
MiniMax · Jefferies report · ARR tops $800M · 5.3x in six months · 20x token growth · M3/H3 iteration · HK AI growth curve
Sources · 信源 Sources
本文基于 Dawn Vision 认知引擎处理当日采集信号并交叉验证 3 个核心来源后生成，经编辑部人工审核。素材来源：新浪财经7x24、华尔街见闻、AASTOCKS。ARR数据为杰富瑞报告口径，非公司官方披露。
This article was generated by the Dawn Vision cognitive engine processing collected daily signals with cross-validation of three core sources, followed by human editorial review. Sources: Sina Finance 7x24, Wallstreetcn, AASTOCKS. ARR figures reflect Jefferies' estimates, not official company disclosure.
