China’s government has a plan to cut Nvidia out of its AI ecosystem. The execution is proving considerably harder than the directive.
In mid-September 2025, China’s Cyberspace Administration (CAC) issued guidance barring major technology firms from purchasing Nvidia AI chips, pushing them toward domestic alternatives instead. Companies like ByteDance and Alibaba were told to suspend acquisitions of Nvidia hardware, including the RTX Pro 6000D, as Beijing doubled down on its push for technological self-reliance.
A ban that is easier to write than to enforce
Huawei’s Ascend chip series represents the most prominent domestic contender, and it has made genuine progress, but engineers and developers working with it consistently describe a product that trails Nvidia’s hardware on raw performance and, more critically, on software maturity.
Software ecosystems are the quiet moat around Nvidia’s dominance. CUDA, the programming framework that makes Nvidia GPUs so useful for AI workloads, has been built up over nearly two decades, with thousands of optimized libraries, tools, and developer integrations layered on top. Replacing a chip is hard. Replacing the entire scaffolding of code that runs on it is a different problem entirely.
The workaround some firms are exploring involves optimizing inference workloads to run more efficiently on lower-powered or mid-range local hardware. DeepSeek, Baidu, and Alibaba are among the companies reportedly adapting their operations to what is available domestically.
Nvidia’s China revenue goes to zero
Even after the US cleared the H200 chip for sale to China, no Chinese firms have purchased it. Beijing’s restrictions have been comprehensive enough to block purchases even when Washington gave its approval.
As of mid-2026, Nvidia has reported zero revenue from H200 chip sales to China. For context, China was historically one of Nvidia’s largest markets for data center hardware before successive rounds of US export controls began tightening access in 2022 and 2023.
State money flows toward domestic chips
State-funded data center projects have been retooled to require domestically produced chips, with the requirement applied retroactively to ongoing builds. That kind of mandated procurement creates a captive market for Chinese chip suppliers, even when their products are not yet fully competitive on performance.
What Beijing has essentially done is run two parallel tracks. The first track is a hard political directive to reduce reliance on foreign chips. The second track is a development program to make domestic chips good enough to carry that directive without causing the AI sector to fall behind. Right now, track one is moving faster than track two.
What to watch from here
DeepSeek’s emergence as a globally competitive model trained on constrained hardware was a demonstration that hardware disadvantage is not necessarily fatal to AI progress.
For Nvidia, zero revenue from approved H200 sales is a clear signal that compliance-oriented chip design for China is not producing results under the current political climate.
Huawei’s Ascend roadmap will be closely watched by developers who are currently making do with what is available and waiting to see if something meaningfully better arrives.