Why the U.S. Needs a Leading-Edge Foundry

Semis and the AI Arms Race, Countering a Taiwan Blockade, Intel Foundry

Semis and AI Innovation

Artificial intelligence is driving the next leap forward in military technology, revolutionizing warfare across all domains—land, sea, sky, and space. Relevant examples include Anduril, Palantir, and Vannevar Labs.

AI-enhanced warfare is inextricably linked to leading-edge semiconductors, which raises national security concerns about the semiconductor supply chain.

Let’s explore from first principles.

AI Relies on Semis

Training an AI model relies on four main ingredients: talent, data, algorithms, and compute.

Talent: Developing advanced AI systems requires highly skilled machine learning engineers. The United States leads in this area, with its ecosystem of startups, universities, and research labs consistently producing top-tier AI talent.

Data: Training AI requires vast amounts of high-quality data. The U.S. benefits from abundant English language user-generated content on sites like YouTube and Reddit. High-quality labeled data is essential for model fine-tuning, and American companies like Scale AI and Labelbox provide labeling services.

Algorithms: The U.S. is a global leader in algorithmic innovation. Industry or academic researchers often publish these advancements, so they diffuse relatively quickly. The U.S. is well-positioned to maintain its leadership in AI algorithms due to its talent, research institutions, and location in the center of the AI world.

Compute: State-of-the-art AI training relies on advanced AI hardware such as Nvidia’s H100 GPUs, which rely on logic and memory chips built with leading-edge semiconductor technology. The U.S. has the leading AI system designer in Nvidia. Also, it has the largest AI compute infrastructure across tech giants like Microsoft, AWS, Google, and Meta, alongside new GPU clouds like CoreWeave.

AI Arms Race

Companies such as OpenAI, Anthropic, and Meta leverage their access to talent, data, and compute to develop leading AI models. They fiercely compete to launch increasingly intelligent models — a modern AI arms race.

A similar dynamic is likely to unfold in AI-enabled warfare.

For example, Anduril’s Lattice for Command and Control leverages AI to accelerate the "understand-decide-act" loop at speeds and scales beyond human ability.

While Anduril’s AI-powered products are only available for the U.S. and its allies, rival nations will attempt to build comparable alternatives. These nations are skilled at developing defense hardware and related software but can’t as easily replicate the the underlying AI model performance.

Put simply, the best AI will win wars.

To maintain an AI edge, the U.S. and its allies must constantly advance state-of-the-art AI models.

How can America sustain its AI lead?

More specifically, along which vectors—talent, data, algorithms, or compute—can America sustain its lead for the longest?

Talent

Although the United States currently leads in global AI talent, China also has a large pool of elite AI engineers. ChinaTalk’s translation of an interview with DeepSeek’s CEO Liang Wenfeng explains:

Waves: Jack Clark, former policy director at OpenAI and co-founder of Anthropic, said that DeepSeek hired “inscrutable wizards.” What kind of people are behind DeepSeek V2?

Liang Wenfeng: There are no wizards. We are mostly fresh graduates from top universities, PhD candidates in their fourth or fifth year, and some young people who graduated just a few years ago.

Waves: Many LLM companies are obsessed with recruiting talents from overseas, and it’s often said that the top 50 talents in this field might not even be working for Chinese companies. Where are your team members from?

Liang Wenfeng: The team behind the V2 model doesn’t include anyone returning to China from overseas — they are all local. The top 50 experts might not be in China, but perhaps we can train such talents ourselves.

Moreover, much of the U.S.’s top AI talent works in commercial companies that often deprioritize or avoid defense-related work—for example, Google’s withdrawal from Project Maven. In contrast, China’s Military-Civil Fusion policy encourages the integration of civilian technology into military development. Thus, while China may have less AI talent overall, a larger proportion may be directed toward defense.

Data

The U.S. has the data ecosystem needed to train advanced AI models like OpenAI’s GPT-4. However, China can establish a comparable system. Major Chinese firms, including Tencent, Baidu, Alibaba, and ByteDance, generate vast datasets suitable for AI training. Furthermore, the extensive surveillance by the People’s Republic of China (PRC) provides access to detailed information on citizens’ faces, communications, transactions, and behaviors, potentially giving it an edge in developing AI models tailored for military use.

Algorithms

Most algorithmic AI innovations diffuse rather quickly, whether because they are openly published or because elite AI talent can reverse engineer what’s likely happening under the hood. There’s also a culture of job hopping and tight circles in the AI community, so knowledge can informally spread. That said, it’s easier to figure out what’s going on under the hood if you can actually get your hands on the product; this can be much harder in defense.

And yet, China can innovate from first principles, too:

​​This success stems from DeepSeek’s comprehensive innovation in model architecture. They proposed a novel MLA (multi-head latent attention) architecture that reduces memory usage to 5-13% of the commonly used MHA architecture. Additionally, their original DeepSeekMoESparse structure minimized computational costs, ultimately leading to reduced overall costs.

In Silicon Valley, DeepSeek is known as “the mysterious force from the East” 来自东方的神秘力量. SemiAnalysis’s chief analyst believes the DeepSeek V2 paper “may be the best one of the year.” Former OpenAI employee Andrew Carr found the paper “full of amazing wisdom” 充满惊人智慧, and applied its training setup to his own models. And Jack Clark, former policy head at OpenAI and co-founder of Anthropic, believes DeepSeek “hired a group of unfathomable geniuses” 雇佣了一批高深莫测的奇才, adding that large models made in China “will be as much of a force to be reckoned with as drones and electric cars” 将和无人机、电动汽车一样，成为不容忽视的力量.

Compute

The U.S. is the world leader in AI hardware and has a defined strategy to sustain this lead through sanctions on the sale of AI chips to China. These restrictions push the PRC to develop competitive AI accelerators domestically. Additionally, with sanctions preventing TSMC from manufacturing AI chips for China, the PRC must rely on its own semiconductor production capabilities, which currently lag two process nodes behind—a gap SemiAnalysis estimates equates to a five-year delay.

Achieving parity in leading-edge semiconductor manufacturing presents significant challenges for the PRC. Fabricating advanced chips requires leading-edge wafer fabrication equipment (WFE), which also remains under U.S. sanctions. In response, China is attempting to evade restrictions and develop a domestic WFE supply chain.

However, one tool—Extreme Ultraviolet Lithography (EUV)—remains China’s white whale and is produced solely by ASML in the Netherlands. Without EUV, the PRC’s ability to achieve parity in leading-edge chip production remains constrained, though efforts to overcome this barrier are already underway.

Once the PRC overcomes the EUV barrier, U.S. sanctions will lose their potency and the U.S. compute advantage will erode. The race for compute leadership will then depend on which ecosystem—global or Chinese—can outpace the other in developing advanced semiconductor manufacturing process nodes.

This will challenge the PRC, as the worldwide semiconductor ecosystem excels at collaboratively developing the next node across design, tooling, and manufacturing companies. In contrast, China’s nascent semiconductor industry has limited collaborative experience at the leading edge.

Blockade

If You Can’t Beat Them, Choke Them

The PRC’s inability to overcome its semiconductor deficit could lead to aggressive actions against TSMC’s home country, Taiwan. As a chokepoint in the global semiconductor ecosystem, combined with the CCP’s desire for reunification, Taiwan is a prime target for a People’s Liberation Army blockade.

This scenario prompts important questions: If Taiwan’s energy and food supplies are cut off, would TSMC’s production of AI accelerators needed for American defense technology grind to a halt? Would the research and development of the subsequent process node stall too?

Such a blockade would be an effective strategy by the Chinese Communist Party (CCP) to freeze the progress of America’s defense technology.

Countering A Blockade

The risk of a Taiwan blockade underscores the urgent need to reduce reliance on TSMC as a single point of failure in the global semiconductor supply chain. Establishing a leading-edge advanced logic semiconductor foundry on U.S. soil is critical to mitigating this vulnerability and strengthening deterrence.

TSMC recently built a modern fab in Arizona, but Taiwan doesn’t want the U.S. to have TSMC’s leading edge. Furthermore, a Taiwan blockade would destabilize TSMC’s international workforce, impacting American production.

Alternatively, a domestic leading-edge semiconductor foundry on American soil would ensure U.S. control over its advanced semiconductors, AI, and defense technology pipelines. This strategy reduces reliance on foreign supply chains, mitigates geopolitical risks, and strengthens national security. It could also reduce the probability of Taiwan being held hostage to slow the U.S. compute advantage.

Intel Foundry

Intel Foundry, with advanced fabrication facilities in the United States and Europe, has the potential to serve as a cornerstone of America’s national security and deterrence strategy.

Intel has issues, but they are mainly economic. It lacks the broad external customer base necessary to ensure financial stability. Additionally, it struggles to compete with TSMC’s cost efficiency in scaled manufacturing. Finally, Intel Foundry relies on the broader Intel organization for financial support, which forces Foundry to prioritize Intel’s internal needs over external customer demands.

Can the United States take action to ensure Intel Foundry becomes a financially sustainable business? Can Intel Foundry become a cornerstone of national security and a key player in deterring the CCP?

Stay tuned.