China is racing to build a self-sufficient artificial intelligence stack, from cutting edge chips to nationwide deployment, and its leaders now treat that effort with the urgency of a wartime program. A secretive extreme ultraviolet lithography project in Shenzhen, widely likened to a “Manhattan Project” for semiconductors, suggests Beijing is closer than many in the West assumed to breaking free of foreign technology choke points. If that breakthrough holds up, it could sharply narrow the gap between Chinese and Western AI capabilities, even if it does not erase the structural advantages that still favor the United States and its allies.
At the same time, China is trying to wire AI into almost every corner of its economy, betting that scale and state direction can compensate for late entry in some frontier models. That strategy carries real risks, from wasted investment to brittle systems, but it also means Western policymakers can no longer assume a comfortable lead. The contest is shifting from who has the single best model to who can combine compute, chips, and deployment at national scale.
The Shenzhen breakthrough and China’s ‘Manhattan Project’ moment
For years, export controls on advanced lithography tools were supposed to keep China at least a generation behind in the most advanced chips. That logic is now under strain as reports describe a secret facility in Shenzhen where engineers have assembled an operational extreme ultraviolet machine that insiders explicitly compare to a Chinese “Manhattan Project” for chips. The project, centered on The Shenzhen effort, is portrayed as a tightly controlled, state backed program designed to punch through the technological ceiling imposed by foreign suppliers.
Industry accounts describe this as a decisive break from the older subsidy heavy model that scattered funds across dozens of local chip ventures with mixed results. Instead, the Shenzhen initiative concentrates resources, talent, and political backing in a single, mission driven complex that some analysts say is intended to be as strategically important for China as the original Manhattan Project was for the United States. One detailed narrative of how China organized this push describes a system designed to be “immune to foreign interference,” underscoring how central technological sovereignty has become to Beijing’s AI ambitions.
Reverse engineering EUV and the ASML factor
The most sensitive piece of this story is extreme ultraviolet lithography, a technology long dominated by ASML and ring fenced by export controls. Western officials assumed that without direct access to these machines, Chinese fabs would be stuck at older process nodes that are less efficient for AI training and inference. That assumption is now being tested by reports that ex engineers from a European chip giant helped build a domestic EUV system in China, with one account noting that the machine fills nearly an entire factory floor and is already generating extreme ultraviolet light, even if it is not yet as compact or refined as ASML’s tools.
Those reports suggest a mix of reverse engineering, local innovation, and quiet recruitment of foreign expertise, with some of the specialists said to have worked under aliases to maintain secrecy. One detailed description of the project notes that the new machine was built in defiance of earlier claims that China was “way behind,” and that it is already operational enough to challenge the notion that ASML is the only game in town for EUV. Another account, shared by a user called moses_the_blue, relays how Reuters reporting described China’s EUV progress as “prudent in retrospect,” a phrase that captures how Western complacency about permanent technological dominance is starting to look misplaced.
From chip bottlenecks to AI compute at scale
Even with a working EUV prototype, China still faces a yawning gap in overall AI compute compared with the United States and its allies. Analysts who track the largest large language models note that the compute gap between leading US and Chinese systems is “staggering,” and that at the national level the US still commands a far larger pool of high end accelerators and data center capacity. One influential assessment argues that Chinese efforts are constrained not only by hardware access but also by institutional rigidities and a lack of funding mechanisms that can match the venture capital ecosystem in the United States.
Yet the same analysis concedes that compute is not static, and that breakthroughs in domestic chipmaking could gradually erode the US advantage if they are scaled and integrated into production. On the American side, experts warn that while America still leads in AI innovation, US companies currently control about 75 percent of global AI compute, a share that could slip if Washington fails to sustain investment and industrial policy. In that context, a Chinese EUV breakthrough is less about instant parity and more about changing the long term trajectory of the compute balance.
Huawei, sanctions, and the new semiconductor battlefield
Sanctions were supposed to box Chinese champions like Huawei into older chip generations, limiting their ability to build competitive AI hardware. Instead, Huawei has emerged as a central player in the country’s semiconductor push, using creative design workarounds and domestic fabs to keep advancing its processors. Reports on the Shenzhen program describe Huawei’s involvement in China’s semiconductor “Manhattan Project” as a sign that the company is now a systems integrator for a broader national effort, not just a smartphone and telecom vendor trying to survive on the margins.
That role matters because Huawei can translate breakthroughs in lithography into actual chips optimized for AI workloads, from training large models to running inference on edge devices. One detailed account of the program notes that Huawei’s participation is intended to support tasks such as AI training, directly linking the semiconductor push to the broader contest with the West. If Huawei and its partners can mass produce competitive accelerators on homegrown EUV lines, the leverage of export controls will diminish, and the AI race will shift even more decisively to software, data, and deployment.
China’s ‘AI+’ strategy and the 90 Percent ambition
Hardware is only one side of Beijing’s plan. The other is a sweeping attempt to embed AI into everyday economic activity, often described domestically as an “AI+” drive. Chinese policymakers define AI broadly, extending far beyond content generating large language models to include industrial tools, logistics optimization, and algorithmic management systems that touch factories, offices, and public services. Analysts who have examined this agenda note that Defining AI in this expansive way allows Chinese authorities to claim rapid progress across sectors, even when frontier research still lags behind the very top US labs.
Beijing has set a particularly bold target to Integrate AI Into 90 Percent of Its Economy by 2030, a goal that reflects both ambition and political signaling. One critical assessment of this plan, titled “China Wants to Integrate AI Into 90 Percent of Its Economy by 2030. It Won’t Work,” argues that the slogan “China Wants” to wire AI into “90 Percent” of activity glosses over deep structural obstacles, from uneven digital infrastructure to misaligned incentives in state owned enterprises. The same analysis contends that the push to embed AI into such a large Percent of Its Economy risks top down campaigns that prioritize headline numbers over genuine productivity gains, and concludes bluntly that “It Won” Work as advertised.
Performance gap: Chinese models are catching up
Despite those caveats, Chinese AI systems are closing the performance gap with Western models faster than many expected. Benchmark tests show that in some categories, Chinese models now match or nearly match US systems, particularly in language understanding and coding tasks. One widely cited set of evaluations found that Chinese systems narrowed that gap to 0.3% in 2024 on certain leaderboards, a margin that is effectively within the noise for many real world applications.
The same testing showed that In General Reasoning, measured by the MMMU benchmark, the U.S. advantage shrank from a double digit lead to 13.5%, underscoring how quickly Chinese labs are learning to scale and fine tune large models. Some Chinese systems now outperform any American made model on specific tasks, even if the overall ecosystem still tilts toward US players. For policymakers in Washington and European capitals, those numbers are a warning that the AI race is no longer a comfortable blowout but a contest in which incremental gains can flip the narrative in a few training cycles.
The ‘Silicon Curtain’ and decoupling from the West
As China builds its own EUV tools and AI chips, a new metaphor has entered the debate: the “Silicon Curtain.” The phrase captures a world in which semiconductor and AI ecosystems split into partially separate blocs, with limited cross border technology transfer and parallel standards. The Shenzhen project, explicitly described as China’s “Manhattan Project” for chips, is a prime example of this trend, since it is designed to function without reliance on Western suppliers or intellectual property. One detailed account of the program notes that the Manhattan Project style initiative positions the Chinese state as the central systems integrator, coordinating everything from optics to control software.
For the West, this raises the prospect that export controls could accelerate, rather than prevent, the emergence of a rival AI industrial base that is largely insulated from external pressure. Once China can produce advanced chips at scale, the leverage of tools like ASML’s export licenses will diminish, and the focus will shift to standards, security, and market access. Some analysts already speak of a future in which Chinese and Western AI systems operate on different clouds, use different encryption and identity frameworks, and are trained on divergent data regimes, creating a digital divide that is as much political as it is technical. The Shenzhen program, described by one report as a Manhattan Project that “shatters global export controls,” is an early glimpse of that world.
Why the West’s lead still matters
None of this means that the United States and its allies have already lost their edge. The US still hosts the most advanced frontier labs, the deepest pools of AI talent, and the largest commercial platforms deploying these systems at scale. Analysts who warn about complacency also emphasize that America currently enjoys a dominant share of global AI compute, with US firms controlling about 75 percent of the world’s capacity for training and running large models. That concentration gives Washington and its partners significant leverage over standards, safety practices, and the direction of research, at least for now.
At the same time, the structural weaknesses that hamper China, from capital allocation to institutional constraints, are not easily fixed by a single breakthrough in Shenzhen. One detailed critique of Beijing’s strategy argues that At the national level, China’s political economy makes it hard to sustain the kind of open ended, high risk research culture that produced many of the breakthroughs in deep learning. That analysis, which underscores the “staggering” compute gap and a lack of funding mechanisms, suggests that even a successful “Manhattan Project” for chips will not automatically translate into a dominant AI ecosystem. The West’s challenge is to treat China’s surge as a spur to invest and reform, rather than as an excuse for fatalism.
What a narrower gap means for global AI governance
A world in which China is only a step behind the West in AI capabilities will be very different from one in which the United States enjoys an unassailable lead. For one thing, it will be harder to set global norms and safety standards if Beijing feels it can simply go its own way with comparable technology. The “AI+” strategy, which aims to diffuse AI across China’s economy and society, already reflects a governance model that prioritizes state control and surveillance in ways that clash with liberal democratic values. Analysts who have examined this approach warn that Chinese deployment choices could become a wake up call for Europe and other regions that have been slower to integrate AI into public services.
At the same time, a narrower gap could create incentives for cooperation in areas like safety research and incident reporting, since both sides would have a shared interest in preventing catastrophic misuse or accidents. Yet that kind of collaboration will be hard to sustain if the underlying competition is framed as a zero sum race for economic and military dominance. The emerging “Silicon Curtain,” the Shenzhen “Manhattan Project,” and the push to Integrate AI Into 90 Percent of Its Economy all point toward a future in which AI is deeply entangled with national power. Managing that reality will require Western governments to invest in their own capabilities, understand the details of China’s strategy, and avoid both complacency and panic as the gap narrows.
More from MorningOverview