$BABA: Alibaba's Qwen4-Max turns the AI race into a unit-economics game — and Citi still sees 109% upside. Buy, $193 target vs $92.21 cited (16 Nov).

Qwen4-Max launched 16 Nov: 2.4T total params, 71B active per token, 1M-token context, native vision; #6 on Code Arena for web dev.

vs Qwen3.8-Max, token consumption falls 24–49% across benchmarks and average task cost drops ~58% — RMB 26.58 to RMB 11.16. API priced at RMB 10/30 in China, $1.50/$4.50 overseas per 1M input/output tokens.

Citi's read: Alibaba pivots from raw parameter expansion to enterprise productivity and agentic workflows — dev, manufacturing, finance, legal — lifting MaaS adoption, inference demand and Cloud revenue.

Valuation: $193 SOTP — 10x P/E FY27E ecommerce profit, 7.0x P/S AI Cloud, 5.0x P/S AI Labs & Apps, 30% discount to net cash.

Unhidea take: the 58% cost cut is the real product — cheaper tokens mean more inference volume, and Alibaba monetizes the volume through Cloud, not the model.
Unhidea take: watch the cloud line, not the benchmark — Qwen4-Max only pays if AI Cloud revenue accelerates; Alibaba reports earnings 19 Nov (earnings calendar), the first check.

Risks: AI spend and margin pressure, retail execution, traffic/GMV slowdown, US–China trade overhang. Source: Citi Research, 16 Nov 2026. Not financial advice.