China’s AI Chip Effort Sparks Global Concern

In the rapidly intensifying global race for artificial intelligence (AI), the spotlight has long been on semiconductor giants such as U.S.-based NVIDIA and Taiwan’s TSMC, whose factories produce the critical GPU chips underpinning data centres worldwide. Memory chip suppliers such as Samsung, Micron, and SK Hynix also play a crucial role. More recently, Google has emerged as a formidable contender with its in-house TPU chips.

However, much less widely known is the Dutch company ASML, which produces cutting-edge lithography machines—the backbone of modern semiconductor manufacturing. ASML is currently the only provider of Extreme Ultraviolet (EUV) lithography systems, essential for producing the most advanced chips.

Here lies the strategic tension: U.S. sanctions prevent ASML from selling these advanced machines to China, while restrictions have also been imposed on the export of NVIDIA GPUs and critical software, aiming to curb China’s rise in the AI race.

Despite these barriers, China has continued to advance. By stockpiling older chips and, in some cases, acquiring components through illicit channels, Chinese firms have successfully developed world-class AI models such as DeepSpeed-IC and CogentAI. Huawei, with its Kirin chip series, and various domestic semiconductor startups have further accelerated China’s capabilities, raising international concern.

Chinese scientists have reportedly developed a prototype lithography machine, currently at the experimental stage, capable of generating EUV light. The project, often referred to as China’s “Manhattan Project,” involves former ASML engineers working under high security and with pseudonyms. Their goal is to produce fully functional advanced chips by 2030, potentially surpassing existing predictions of China’s technological capacity.

Huawei’s involvement is pivotal. Despite being blacklisted and restricted from using Intel or Qualcomm processors, the company has advanced its HarmonyOS, Kirin chips, and Ascend AI chips, outperforming NVIDIA’s equivalents in certain AI models. Its SuperPod Connect platform links thousands of computing units for AI training, multiplying processing power.

China has also patented a technique called Quadruple Patterning, allowing production of 2-nanometre chips using older DUV machines instead of EUV systems. While the technological breakthrough is significant, large-scale industrial production remains a formidable challenge.

China’s chip initiative underscores the country’s determination to achieve self-reliance in semiconductors, even in the face of international restrictions, and marks a potentially transformative moment in the global AI landscape.