Chip wars escalate: Taiwan raids, china’s tungsten gambit, and the race for AI hardware independence

Taiwan’s recent crackdown on alleged Nvidia chip smuggling to China, particularly through the high-profile raids on Super Micro Computer and its affiliates, underscores the escalating enforcement of US-led export controls targeting AI-capable semiconductor hardware. These actions not only highlight Taiwan’s pivotal role in policing semiconductor supply chains amid intensifying US-China tech rivalry but also reflect the broader geopolitical stakes in controlling advanced AI chip technology. By aggressively investigating and disrupting illicit chip transfers, Taiwan is reshaping the global tech power map and signaling a firm stance in the ongoing AI arms race.

China's strategic export controls on tungsten, a critical material for AI chip production, have become a pivotal lever in the intensifying US-China semiconductor rivalry. By imposing a tungsten export ban, China has disrupted global AI chip supply chains, forcing manufacturers worldwide to scramble for alternative critical minerals and accelerate chemical substitutions that could fundamentally reshape semiconductor production by 2028. This weaponization of tungsten not only exposes a hidden choke point in AI chip manufacturing but also escalates the geopolitical battle over control of essential materials, as soaring tungsten hexafluoride prices reveal the fragility of current supply chains and the strategic importance of securing these inputs.

Beyond export controls, China’s aggressive semiconductor subsidies and industrial policies have triggered massive overcapacity in critical chip chemicals, significantly altering global supply chain dynamics. This overcapacity not only tightens the geopolitical contest for technological independence but also challenges competitors by flooding the market with essential inputs, thereby reshaping pricing and availability. Meanwhile, China's integrated semiconductor supply chain strategy is advancing at a pace that starkly outstrips the US CHIPS Act, positioning China to dominate the global chip race amid surging AI demand and mounting industrial tensions, underscoring Beijing’s commitment to building a self-sufficient and resilient semiconductor ecosystem.

Chinese tech giants such as Huawei, Baidu, and Meituan are rapidly advancing large-scale AI training on domestically developed chips, marking a decisive move toward hardware self-reliance. Huawei’s training of its 718 billion-parameter Pangu Ultra MoE model on 6,000 Ascend NPUs and Baidu’s use of Kunlunxin P800 accelerators for ERNIE 5.1 exemplify this trend, while Meituan’s training of a 1.6 trillion-parameter model on 50,000 domestic AI accelerator cards underscores the scale and maturity of China’s indigenous AI hardware ecosystem. This shift reflects a broader strategy to build a sovereign AI stack that can operate independently amid tightening US export controls.

Emerging AI firms like DeepSeek and Zhipu AI are spearheading the development of custom AI inference chips, aiming to reduce dependence on Nvidia and Huawei hardware while achieving greater hardware-software synergy and lowering operational costs. DeepSeek’s quiet expansion of its chip design team and plans to raise up to $7 billion for semiconductor ventures, alongside Zhipu AI’s pursuit of bespoke ASICs tailored to its GLM model family, reflect a strategic push to control the inference stack—a critical frontier where AI models transition into scalable commercial products. This mirrors a global trend among leading AI labs but carries heightened urgency in China due to restricted access to cutting-edge US technology.

The domestic AI chip design ecosystem in China has expanded significantly, with companies like Cambricon Technologies, Biren Technology, and Hygon Information Technology gaining traction as Chinese firms plan to allocate nearly half of their AI accelerator budgets to local suppliers within the next year, up from 30% currently. Supported by a massive government-led $294 billion investment in AI infrastructure and data centers, this growing supplier base is poised to benefit major tech firms such as Tencent, Alibaba, and Huawei, reinforcing China’s ambition to achieve at least 80% domestic supply of core AI technologies. However, ongoing global memory chip shortages remain a bottleneck, potentially constraining growth despite these advances.

China’s vast talent pool—three times larger than that of the US—and its enormous domestic market provide a formidable advantage in AI chip innovation, enabling exploration of diverse research paths that could challenge Nvidia’s dominance. As experts note, China’s rapid progress in sectors like electric vehicles and life sciences parallels its AI ambitions, although long-term demographic challenges such as declining fertility rates may slow innovation beyond the 2030s. Moreover, sustained global cooperation remains critical, as complete decoupling due to export controls would ultimately harm both China and the US, underscoring the need for strategic openness despite geopolitical tensions.

Amid escalating US-China AI hardware competition and stringent export controls, China has pragmatically permitted limited imports of Nvidia's high-end H200 AI chips to alleviate critical compute shortages faced by leading domestic AI firms such as Alibaba, ByteDance, and DeepSeek. This cautious reopening reflects Beijing's recognition that despite its long-term ambition for semiconductor self-reliance, immediate AI development demands necessitate selective easing of restrictions to maintain competitiveness in training larger AI models.

China’s approach embodies a strategic compromise that balances urgent hardware needs with its broader goal of nurturing domestic chip capabilities. While allowing only a very limited number of H200 chips under strict conditions, Beijing continues to prioritize homegrown solutions, anticipating advancements from companies like Huawei, whose next-generation AI chips are expected to rival Nvidia’s offerings by next year. This dual-track strategy underscores China’s intent to reduce foreign dependency over time while managing geopolitical and governance risks posed by US export controls.

Beyond chip imports, China’s semiconductor ambitions require substantial expansion in manufacturing capacity, memory supply, and trusted software ecosystems to support sustainable AI growth. The limited reopening to Nvidia’s H200 chips thus serves as a temporary bridge addressing immediate compute shortages, while China invests heavily to build a resilient, domestically controlled AI hardware supply chain capable of competing globally amid the intensifying US-China technology rivalry.