Alibaba Proposes 10-Trillion-Parameter Model and Proprietary Silicon

Alibaba Group has mounted an aggressive campaign to dominate the global artificial intelligence landscape, unveiling ambitious plans for ultra-large foundation models alongside high-performance silicon engineered to directly counter mounting American export restrictions. At its annual Apsara conference in Hangzhou, the Chinese technology titan asserted its dominance across every critical tier of the AI supply chain—from foundational model architectures and proprietary semiconductors to high-density server infrastructure. This comprehensive strategic push signals a major pivot toward technological self-reliance, driving investor optimism and lifting the firm's Hong Kong-listed shares by over 5%.

At the core of this expansion is a relentless pursuit of artificial superintelligence. Chief Executive Officer Eddie Wu framed the current state of technology as a mere prelude to a broader revolution, comparing the emergence of machine intelligence to the transformative scale of the Industrial Revolution. Emphasizing that humanity currently accounts for the overwhelming majority of computational cognition, Wu projected a fundamental shift in which automated systems will eventually generate over a thousand times the collective analytical output of humankind. In pursuit of this vision, Alibaba’s Qwen research unit is actively developing its next-generation architecture, targeting a massive scale of 5 trillion to 10 trillion parameters. This represents a quadrupling in capacity compared to its flagship Qwen 3.8 Max, which operates at 2.4 trillion parameters. The upcoming Qwen 4 iteration and its planned successors, Qwen 4.5 and Qwen 5, are designed to tackle complex, long-horizon tasks through enhanced self-improvement algorithms, enabling models to autonomously identify functional flaws, execute optimization experiments, and generate high-quality synthetic training data with minimal human oversight.

Complementing this software ambition is a critical breakthrough in hardware independence. Alibaba’s proprietary semiconductor division, T-Head, has developed the Zhenwu V900 processor, positionings it as the nation's most formidable domestic alternative to Nvidia’s restricted hardware. Delivering a threefold performance jump over the previous-generation M890 chip introduced earlier in the year, the Zhenwu V900 is architected for massive scaling. The silicon can be clustered into clusters housing up to 500,000 unit nodes, providing the raw compute power necessary to train and execute multi-trillion-parameter models. Scheduled for full-scale mass production and commercial distribution in the first quarter of 2027, the processor is slated to drive exponential growth in Alibaba's annual chip shipments, effectively blunting the impact of Western trade barriers.

To power these massive hardware deployments, Alibaba Cloud is scaling its global physical infrastructure at an unprecedented pace. The enterprise has committed to expanding its global data center operational footprint past 20 gigawatts by 2032, deploying high-density AI supernodes at a commercial scale. Despite persistent supply chain bottlenecks that currently constrain expansion speed, enterprise demand for AI infrastructure remains exceptionally intense, accelerating revenue growth across Alibaba’s cloud computing division. Customer demand continues to dramatically exceed available capacity, underscoring the market's reliance on domestic cloud infrastructure. Rather than viewing current breakthroughs like AI-assisted coding as the peak of innovation, Alibaba regards them merely as primitive indicators of a much larger transformation—drawing a direct analogy to the early light bulb in the dawn of the electrical era. Through integrated software, custom silicon, and hyper-scale infrastructure, the company is systematically building the groundwork to lead the next computational revolution.