China’s trillion-parameter AI push accelerates as US export controls tighten access to rivals

Chinese AI developers are accelerating their push into massive foundation models with more than a trillion parameters, just as Washington moves to block foreign access to leading US software through unprecedented export controls. The message is blunt: when access to top-tier tooling tightens, model-building speed and scale become the escape hatch.

Parameters are the headline metric here, serving as a primary measure of an AI’s capabilities. SCMP frames the sprint as an effort to narrow the gap with leading US rivals like OpenAI and Anthropic, which continue to aggressively expand the size of their top models. In other words, this is not a side project. It is a scaling strategy aimed at competing where performance expectations increasingly track with model size.

To understand why this matters beyond the lab, you have to zoom out to how foundation models get used in the real world. “Foundation model” has become shorthand for general-purpose AI systems that can be adapted to many tasks, from customer support to coding assistance to knowledge retrieval. When developers chase trillion-parameter scale, they are trying to buy capability headroom, especially for tasks where smaller models can struggle or require more scaffolding. If the US side is pushing larger top models, the Chinese side has a simple competitive question to answer: can we reach a similar tier of performance even if we cannot lean as hard on imported US software?

That brings the export controls into focus. SCMP links the timing directly to Washington’s move to block foreign access to leading US software through export controls described as unprecedented. The operational consequence for companies in other countries is straightforward: even if the demand for frontier AI keeps rising globally, the supply of key technologies, components, or software stacks can get constrained. In these moments, strategies often converge on two paths. One is to find alternative sources of comparable capability. The other is to build it locally at a faster pace and at larger scale.

China’s trillion-parameter push reads like the second path, amplified by urgency. The source notes that Chinese companies have been looking to narrow the gap with US rivals such as OpenAI and Anthropic. Those companies are described as continuing to aggressively expand the size of their top models. So the competitive pressure is not abstract. It is measured in an ongoing cycle of “bigger model, better performance expectations,” with the United States setting a pace that rivals try to match.

This is where boards and executives need to pay attention to the second-order implications. When model size becomes a proxy for capability, resource allocation decisions get harder. Compute, data pipelines, training infrastructure, and talent all scale with ambition, and scale introduces risk: cost overruns, training instability, and the possibility that bigger is not enough if data quality, alignment, or system integration lag behind. Export controls also raise procurement uncertainty, forcing companies to redesign dependencies. That means budgets cannot just cover training. They also need buffers for what you cannot import, and for engineering work needed to replace missing software access.

There is also a portfolio implication. If your strategy depends on frontier performance, you may feel pressure to treat trillion-parameter development as a baseline, not a differentiator. Yet history in AI has shown that application wins do not come only from raw scale. They come from what happens after training: fine-tuning workflows, evaluation, safety, product integration, and the ability to turn model capability into reliable user outcomes. Still, SCMP’s framing suggests that, for now, model scale is part of the race identity, because parameters remain the primary measure of AI capabilities.

For peers facing similar constraints, the strategic stake is clear. If export controls tighten access to key US software while top American labs keep increasing model sizes, the competitive window shifts. Firms that can accelerate foundation-model development may reduce performance gaps. Firms that cannot may find themselves locked into smaller models, constrained ecosystems, or higher costs for workarounds. In the AI industry, where attention and capital follow perceived capability, falling behind on the scale narrative can quickly become a business problem, not just a technical one.