Alibaba's Qwen3.8-Max launch lifts cloud case as CICC sees 45% growth

Alibaba's Qwen3.8-Max, a 2.4-trillion-parameter model activating just 95 billion parameters per query, gives Alibaba Cloud a frontier anchor as CICC and two Wall Street banks project revenue growth near 45 percent.

"Alibaba Cloud is the only full-stack AI vendor in China with scale, self-developed chips and a top-tier model family," CICC said in a report dated Aug. 7, adding that it expects revenue acceleration and margin expansion.

Qwen3.8-Max, launched Aug. 3, uses a sparse mixture-of-experts design with a 1 million-token context window and ranks fifth on the Text Arena leaderboard and second on Vision Arena. Alibaba will release the model's weights next week, the first Max-class Qwen to go open-weight.

Alibaba Cloud reported fourth-quarter revenue of 41.63 billion yuan, up 38 percent year over year, with external revenue growth of 40 percent. Citi maintained a Buy rating with a $192 target and Morgan Stanley kept an Overweight rating with a $180 target, naming Alibaba its top pick in China's internet sector.

Open Weights Turn Developers Into Cloud Customers

Alibaba's decision to open-source Qwen3.8-Max is the point of the exercise. Every developer who builds on the model needs computing power, hosting and storage, and that demand flows into Alibaba Cloud, which has posted triple-digit growth in AI-related product revenue for 11 straight quarters. The company plans to charge the largest commercial users of its open Qwen models, mirroring Moonshot's revenue-sharing terms that require up to 30 percent of sales for partners with more than $20 million in annual revenue. The model's open-weight status also puts a frontier-scale system beyond the reach of export controls once downloaded, a position closed rivals such as Anthropic's Fable 5, temporarily placed under US export controls earlier this year, do not enjoy.

A Crowded Race With Cheaper Rivals

Qwen3.8-Max's 2.4 trillion parameters trail Moonshot's Kimi K3 at 2.8 trillion, and DeepSeek's V4-Flash prices at $0.14 per million input tokens, cheaper on a per-task basis than larger models. Alibaba prices Qwen3.8-Max at $2 per million input tokens and $6 per million output tokens, well below Anthropic's Fable 5 at $10 and $50. Baidu, Alibaba's clearest US-listed rival, pairs its ERNIE model family with the Qianfan Model-as-a-Service platform for enterprise contracts, sovereign deployments and its Apollo Go autonomous fleet, setting up a two-front battle in China's enterprise AI market.

Alibaba trades at 19.19 times forward earnings versus Baidu's 24.27, a cheaper multiple despite a larger AI and cloud footprint. Hedge fund ownership fell to 102 funds from 115 last quarter, and short interest stands at 1.79 percent of float versus Baidu's 3.68 percent. The open-weight release next week will test whether developer traffic converts into paid cloud usage, the bet underpinning both banks' bullish calls.

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