China’s AI lesson for the U.S.: it takes more than chips

The big picture

Washington’s export controls have squeezed China’s access to cutting-edge AI semiconductors. But chips alone don’t decide the race. China’s response — spanning data infrastructure, software optimization, and mass deployment — offers a blunt lesson: competitiveness comes from the whole stack and its adoption, not just the silicon at its core. For Western readers used to equating AI progress with GPU counts, China’s model asks a different question: how quickly can you translate compute into useful, scaled applications?

Platforms, data and industrial uptake

China’s internet giants — Baidu (百度), Alibaba (阿里巴巴), Tencent (腾讯) and ByteDance (字节跳动) — sit atop enormous consumer platforms that can distribute AI features to hundreds of millions of users overnight, from search and commerce to social video. That scale creates data feedback loops and a ready test bed for generative services. State-backed efforts, including the “East Data, West Computing” (东数西算) program to build national data centers, further push AI into government and industrial workflows such as manufacturing, logistics and public services. The result is an ecosystem where model training, application integration, compliance and go-to-market are advanced in parallel — not sequentially.

Working around sanctions

U.S. rules unveiled in 2022 and tightened in 2023 restrict China’s access to Nvidia’s A100/H100-class GPUs and later curbed China-specific variants like the A800/H800. In response, Chinese firms reportedly stockpiled older accelerators and retooled software to squeeze more from constrained hardware via quantization, sparsity and network optimizations. Huawei (华为) is pushing Ascend-series chips as domestic alternatives, and it has been reported that some large-model training runs are migrating to these platforms despite ecosystem gaps. None of this eliminates the performance hit from sanctions — but it narrows it, especially when paired with abundant data, aggressive product iteration and government procurement that can guarantee early demand.

The takeaway for Washington

The U.S. has poured billions into fabs under the CHIPS and Science Act, but fabrication is only one pillar. Sustained AI leadership also hinges on power-hungry data centers, high-bandwidth networking, talent and immigration policy, open-source tooling, and demand-side pull through federal and enterprise procurement. Chips matter. So do electricity, software, data access, and distribution. China’s approach, forged under pressure, underscores a simple point: the winner is the one that turns compute into capability — and capability into ubiquitous, trusted products — fastest.