In much of the global debate, the race for artificial intelligence (AI) supremacy has been framed as a binary contest between the United States and China. As Washington envisions it, victory hinges on frontier models and compute scale, a battle dominated by a handful of American labs. China is cast as the challenger, constrained by export controls and dependent on catching up. That narrative is increasingly outdated. China is not pursuing a single AI strategy. It is running three large-scale experiments in parallel, each led by different clusters of firms which are shaped by distinct constraints and pointing towards a different vision of what AI leadership might mean.
The first camp is the compute maximalists, with Alibaba and ByteDance at the forefront. Their wager is that scale and performance still matter most. Alibaba has spent heavily on cloud and AI infrastructure in the past year, pushing free cash flow to a negative CNY21.8 billion, even as cloud revenue grew 34%. It now claims more than 35% of China’s AI cloud market, and its Qwen model family has spawned more than 180,000 derivative open-source models. ByteDance is the consumer-first version of this strategy. Its core products run on some of the most sophisticated recommendation systems in the world. To power these engines, ByteDance trains large proprietary models and integrates generative tools that shape everything from automated video editing to personalized feeds. It is the only Chinese firm whose applied AI already operates at an international scale.
A second camp is represented by Tencent, whose latest earnings reveal a markedly different center of gravity. While peers raised capital expenditure, Tencent’s spending fell 24% year-on-year to CNY13 billion, even as revenue grew 15% and net profit rose 19%. AI has been woven into Tencent’s mature businesses rather than used to pursue frontier training. Marketing services revenue grew 21%, while fintech and enterprise services rose 10%, supported by AI-enhanced tools. With more than a billion users across WeChat and Weixin, even small gains in engagement and monetization produce outsized returns. This is the efficiency-monetization path: China’s advantage lies not only in building the most advanced model, but in distributing capable ones across a massive, integrated consumer and enterprise footprint. Whether Tencent’s restraint becomes an enduring advantage – or whether widening performance gaps eventually force it into heavier investment – remains an open question.
A third strategy is infrastructure sovereignty, led by Baidu and Huawei. In Baidu’s latest quarterly earnings, AI-related businesses grew 50% year-on-year. AI infrastructure and subscription services are expanding faster than the legacy ad business, while its autonomous mobility arm, Apollo Go, surpassed 17 million driverless rides across 22 cities. Rather than matching Alibaba’s cloud scale or Tencent’s consumer reach, Baidu is building the rails – enterprise AI platforms, subscription services, mobility networks and model-training infrastructure – on which much of China’s AI economy could run. Huawei’s Ascend chip ecosystem, meanwhile, anchors a home-grown alternative to Nvidia’s graphics processing units (GPUs), driven by necessity as much as ambition.
These three approaches are more than corporate strategies. They are responses to China’s structural realities: uneven access to advanced GPUs, an enormous and varied domestic market and an industrial policy environment that prizes resilience alongside raw performance. Compute maximalism tries to overcome scarcity by building domestic scale. Efficiency monetization leverages China’s colossal platforms to extract value without competing at the frontier. Infrastructure sovereignty aims to build a complete AI stack able to withstand geopolitical volatility. Part of the reason these strategies coexist is that China faces a uniquely uncertain technological and geopolitical environment. Export controls might tighten. Global markets remain politically fragile for Chinese platforms. The underlying science is still in motion: frontier models continue to grow, yet efficiency-oriented architecture could eventually prove more practical. In this sense, running multiple strategies in parallel is a form of national hedging. It is an attempt to ensure that at least one pathway remains viable regardless of how the global landscape evolves.
Even so, all three draw from a shared backbone of Chinese AI capability, much of it open source. Qwen, DeepSeek, Kimi and others are forked, fine-tuned and redeployed across companies. These strategies are not rigid silos but shifting centers of gravity: three distinct ways to price risk and reward under constraint. This contrasts with the U.S., where economic and regulatory forces have pushed the industry towards consolidation. A small circle of labs dominates frontier-model training, and policymakers increasingly treat concentration as a governance asset. China, by contrast, appears to be enabling parallel experiments, each probing a different technological and commercial frontier, the South China Morning Post reports.
