The Extended Brief

Alibaba touts new AI chip, Qwen Intelligence and age of machine intelligence

Brief by The AI News AI newsroom · Sep 22, 2026, 6:02 PM EDT edition

Original reporting by Constellation Research — Larry Dignan · published Sep 22, 2026, 11:06 AM EDT

Alibaba says its new chip triples its predecessor's performance and anchors a full in-China AI stack, from GPUs to interconnects to Qwen models.

Key points

• Alibaba launched the Zhenwu V900, which it calls China's most powerful AI chip, with three times its predecessor's performance. source ↗
• A single cluster built on the V900 can support up to 500,000 cards, Alibaba said. source ↗
• CEO Eddie Wu said the Qwen team is exploring recursive self-improvement and plans a 5-to-10-trillion-parameter model. source ↗
• Wu said the planned model targets more complex, longer-horizon tasks and progress toward artificial superintelligence. source ↗
• T-Head's portfolio spans Zhenwu GPUs, Yitian CPUs, Panmai smart NICs, and ICN interconnect chips for large AI clusters. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Teams procuring AI infrastructure in China can evaluate V900-based clusters against their training workloads, testing Alibaba's claimed 3x gain over the M890 on real jobs.
• Model builders should track the planned 5-to-10-trillion-parameter Qwen model as a candidate for long-horizon, complex task workloads.
• Infrastructure architects can assess whether T-Head's combined GPU, CPU, NIC, and interconnect lineup works as a single-vendor design for very large clusters.

Context

Alibaba is a Chinese cloud and e-commerce company whose T-Head semiconductor unit designs data-center chips and whose Qwen family are its flagship AI models. The Zhenwu V900 succeeds the M890 as its GPU-class AI accelerator, and Alibaba says one cluster can scale to 500,000 cards. CEO Eddie Wu framed the combined chip-and-model stack, announced at the company's Apsara Conference in Hangzhou, as a path toward artificial superintelligence.

What to watch

• Independent benchmarks of the V900 would confirm or undercut Alibaba's 'most powerful in China' and 3x performance claims.
• A release or technical report for the 5-to-10-trillion-parameter Qwen model would show whether the RSI research yields measurable capability gains.

Related briefs

• Nathan Lambert's written Congressional testimony on the state of open models - Chinese open-weight downloads now 2x America's, >80% of OpenRouter open-model usage
• Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war
• Anthropic launches Claude Opus 5.5, gets cost, efficiency memo
• Xiaomi MiMo-V2.6-Pro 1T-A42B: the new top Open Weights model, trained for $3M

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.