China Powers Up First 10,000-Chip AI Supercluster Built Entirely on Huawei Ascend Processors
Shenzhen activates China's first 10,000-card AI computing cluster built entirely with Huawei's Ascend 910C chips, delivering 11,000 petaflops and signaling that US export controls have not stopped China's AI buildout.
China Powers Up First 10,000-Chip AI Supercluster on Huawei Ascend
Shenzhen has activated China's first 10,000-card intelligent computing cluster built entirely with domestic chips. The facility, powered by Huawei's Ascend 910C AI processors, delivers 11,000 petaflops of computing capacity using what officials describe as "full-stack autonomous and controllable technology."
Combined with a 3,000-petaflop cluster activated at the same site last year, the facility now totals 14,000 petaflops. Nearly 50 organizations have signed computing power framework agreements, pushing the booking rate across both phases to 92%.
This is not a research prototype or a political showpiece. It is a production-scale AI computing facility with real commercial demand.
The Export Control Question
The cluster's existence is a direct challenge to the effectiveness of US semiconductor export controls. Since October 2022, Washington has progressively tightened restrictions on selling advanced AI chips to China, blocking NVIDIA's A100, H100, and subsequent processors from the Chinese market.
The explicit goal was to slow China's AI development by denying access to the most advanced computing hardware. Huawei's Ascend 910C — the chip powering this cluster — is Beijing's answer. While independent benchmarks comparing the Ascend 910C to NVIDIA's H100 remain limited, a 10,000-card cluster operating at 92% booking rate suggests the chip is capable enough for commercial AI workloads.
The uncomfortable reality for US policymakers: export controls may be achieving the opposite of their intended effect. Rather than slowing China's AI progress, they appear to be accelerating the development of a domestic semiconductor ecosystem that did not previously exist at this scale.
Scale in Context
To put 11,000 petaflops in perspective: this is roughly equivalent to 1,400 NVIDIA DGX H100 systems, which would cost approximately $500 million at US list prices. The actual cost of the Huawei-based cluster is not public, but China has been aggressively subsidizing domestic AI infrastructure.
The 92% booking rate is arguably more significant than the raw compute number. It demonstrates that Chinese AI companies, research institutions, and government agencies are not waiting for NVIDIA alternatives — they are already building on Huawei's platform. This creates a self-reinforcing ecosystem: more users generate more software optimization, which improves the platform's competitiveness, which attracts more users.
Huawei's Ascend Ambitions
Huawei has invested heavily in the Ascend line since US sanctions effectively cut the company off from advanced chip manufacturing equipment. The Ascend 910C represents the latest iteration, designed specifically for large-scale AI training and inference workloads.
The company faces real technical challenges. Huawei relies on SMIC for manufacturing, which does not have access to the most advanced lithography equipment. This means the Ascend 910C is likely manufactured on older process nodes than NVIDIA's latest chips, requiring more silicon area and power to achieve comparable performance.
But "comparable" may be good enough. For many AI workloads, having access to a large cluster of slightly less efficient chips is preferable to having no access to a small number of cutting-edge chips.
What Happens Next
The Shenzhen cluster is almost certainly the first of many. Multiple Chinese cities are building or planning large-scale AI computing centers, and Huawei is reportedly ramping Ascend production to meet demand.
For the global AI landscape, this means the era of US hardware dominance in AI compute is ending — not because American chips are less capable, but because China is building a parallel stack that may be good enough for most purposes.
The US export control strategy now faces a difficult question: at what point do restrictions that were designed to maintain a technological advantage instead create a competitor that would not have existed otherwise?
Sources: South China Morning Post, The Manila Times
