Alibaba says it built the ‘most powerful AI chip in China’ as the country races to catch up with the U.S.

Alibaba unveiled its Zhenwu V900, which CEO Eddie Wu called China’s most powerful AI chip and said delivers 3x the performance of the prior Zhenwu M890. The company plans a 5-10 trillion-parameter next-generation model, versus 2.4 trillion parameters for Qwen3.8-Max, and targets more than 20GW of AI computing capacity by 2032. The announcement underscores China’s push for AI-chip self-reliance amid U.S. export restrictions, although data-center supply-chain shortages may constrain Alibaba’s infrastructure expansion pace.

Analysis

The investable signal is less the claimed chip-performance leap than Alibaba’s willingness to internalize the AI stack. If deployment is real, BABA can reduce exposure to constrained imported accelerators and package lower-cost compute with its cloud services, pressuring Tencent (0700 HK) and Baidu (BIDU) in China enterprise AI. The second-order bottleneck shifts to power delivery, cooling, networking, advanced packaging and HBM; domestic chip design progress does not by itself solve foundry yield, memory bandwidth or cluster-interconnect reliability.

The economic trade-off is substantial: a multi-gigawatt buildout implies capital needs that can run into tens of billions of dollars annually at industry-standard data-center costs. Over the next 1-3 quarters, investors should focus on AI-cloud revenue growth, utilization and capex-to-revenue rather than model parameter counts; accelerating capex without corresponding cloud monetization would compress BABA FCF and re-rate it as an infrastructure spender rather than an asset-light platform. Company benchmark claims should be discounted until third-party performance-per-watt and customer adoption data emerge.

For NVDA, the near-term risk is policy rather than immediate revenue displacement: visible Chinese substitution strengthens the case for tighter U.S. controls, while also shrinking the future addressable market for permitted China-specific products. Consensus may be too quick to read domestic Chinese chips as an NVDA-equivalent threat: frontier clusters remain constrained by memory, software tooling and manufacturing yields. A bilateral trade détente or export-license relaxation would be the clearest reversal of the China-substitution narrative within days; sustained evidence of BABA cloud margin resilience is the 6-18 month validation.

Key Decisions for Investors

• Initiate a 3-6 month defined-risk bullish BABA position via call spreads, sized modestly until the next earnings release validates AI-cloud revenue and capex discipline. Target 2:1 upside/downside; exit if management raises infrastructure spend without cloud growth or FCF guidance support.
• Use BABA versus BIDU as the cleaner China-AI relative-value expression: long BABA / short BIDU over 3-6 months, with a 8-10% relative stop. Alibaba has a larger distribution and enterprise-cloud channel, while BIDU is more exposed to AI monetization expectations without equivalent ecosystem leverage.
• Do not short NVDA solely on this development. Instead, establish an alert around any U.S. export-control announcement or China revenue/guidance revision; a confirmed tightening cycle would favor a tactical NVDA underweight for 1-3 months, while license relief invalidates the thesis.
• Watch 0981 HK (SMIC) and Chinese data-center power/cooling suppliers rather than chasing chip-design headlines. Upgrade only after independent evidence of advanced-node yield improvement and AI accelerator volume shipments; absent that data, the supply-chain claim is not yet investable.