Chips and chipsets: How China is rewriting the rules of the AI race

For much of the past two years, the dominant narrative around the global artificial intelligence race has centred on a single, simple metric: which country can build the most powerful large language model or the most advanced semiconductor. This framing, pushed heavily by export controls and strategic rhetoric from Washington, has cast China as a permanent laggard, perpetually chasing American technological superiority in chips. But a quiet yet profound structural shift, articulated with surprising clarity by Nvidia’s chief executive, Jensen Huang, suggests the battlefield is moving. Huang’s observation that the true value of AI lies not in training but in inference—the process of running a trained model to generate answers—and that revenue flows from deploying compute at scale, has been absorbed in Beijing with the force of strategic revelation.

China has concluded that controlling the supply of cutting-edge chips does not automatically translate into controlling the market for artificial intelligence. Instead, the country is focusing its industrial and policy apparatus on what Huang called “tokens as a commodity”: the massive, worldwide deployment of inference workloads that generate actual economic output. This is a subtle but powerful reframing of the competition. If the United States has chosen to compete in the high-end manufacturing of silicon, China has chosen to compete in the volume-driven, cost-sensitive, and rapidly growing business of inference at scale. The implications are strategic. In a world where AI models increasingly commoditise, the entity that can deliver the cheapest, most reliable, and most energy-efficient inference for countless everyday applications—from industrial automation to consumer chatbots to logistics optimisation—will capture the lion’s share of the market’s value.

China’s domestic ecosystem, already the world’s largest in terms of internet users, data generation, and manufacturing output, provides a uniquely fertile ground for this strategy. The country’s sprawling network of factories, ports, and cities offers an almost limitless number of use cases for inference-driven AI. Moreover, by focusing on inference rather than cutting-edge training, Chinese firms can rely on a broader arsenal of chips, including those that are not subject to the most stringent US export restrictions. This approach does not require a direct challenge to Nvidia’s highest-end offerings; it requires mastery of supply chains, software optimisation, and large-scale systems engineering. The result is that while Washington may be winning the battle over the most advanced silicon, Beijing is quietly building the infrastructure to win the war over who profits from AI’s global deployment. For international investors and technology leaders, the lesson is clear: the scoreboard of the AI race is no longer measured in teraflops alone, but in the practical, commercial deployment of intelligence at scale.

Why it matters:
A strategic pivot by China toward dominating AI inference rather than competing head-on for advanced chips could reshape global market dynamics. For hardware suppliers, this shift signals growing demand for cost-effective inference accelerators and optimisation software. Investors and multinational corporations must recalibrate their understanding of competitive advantage in AI, recognising that control over the most powerful chips may not guarantee control over the most lucrative markets.

Source →