Why imported talent is crucial in the race against Beijing

Before Donald Trump’s inauguration, Elon Musk caused a huge controversy within the MAGA movement by advocating increased high-skill immigration. As head of the Department of Government Efficiency he wanted, for example, to expand the H-1B visa program, which many Trump supporters are against.

The angry debate over the visa issue still rages on social media and both sides tend to talk past each other. The MAGA movement is against any increase in immigration, whether high- or low-skill. Musk has acknowledged that the existing H-1B program was subject to abuse by employers and especially by IT firms that rely on outsourcing: the workers they import are often no better than the Americans they replace.

But even so, Musk insists that the US needs exceptionally talented and entrepreneurial immigrants and that America must attract top brains from all over the world if it is to stand a chance of continuing to be a superpower.

Moreover, the US can’t begin to win an economic and technological competition against China without elite immigration. This is a strong statement, but I can prove it to you with a simple calculation.

Consider the fact that China has four times the population of the US. Its 18-year-olds now attend college at roughly the same rate as Americans (this has only recently become true, as China has become richer and more developed). Chinese college students are roughly twice as likely to study science, technology, engineering and math (STEM) and they graduate high school with much stronger math capabilities.

This means that each year China is producing about eight times (almost an order of magnitude) more new engineers and technologists than the US. In fact, the proper comparison is between China and Rest of World (RoW). China produces as much top talent in technology fields as the RoW in aggregate.

China does have demographic challenges due to low fertility, but the children who will enter the workforce over the next twenty years are already born, and detailed analysis shows a huge deepening of their human capital pool — a huge increase in the quality of Chinese skills and knowledge — over that period.

When you combine this with world-leading infrastructure and supply chains, it means that China can innovate simultaneously in many areas: AI, semiconductors, solar energy, batteries, EVs, aerospace, satellite and space technology, biotech, nuclear energy, ship building, advanced materials, drones… the list goes on and on. Ask an expert in any of these areas and they are likely to tell you either that China is close to parity with the West or has already surpassed it.

Take AI, the hottest and fastest-developing area in technology, and a focus of geopolitical and national security concerns. The figure here shows the flow of top research talents in AI, based on data obtained from the leading annual conference NeurIPS.

There are many fascinating and significant facts here, not least that the total number of undergrads from China with the background to pursue a PhD in AI is comparable to the rest of the world combined. (See the bars on the left of graph.)

And about half of top US researchers in AI originally come from China. The rest come from all over the world, and US-born researchers are only a fraction of the total. Most US capability in AI is the result of immigration of top brains from abroad.

The Biden administration tried to stifle China’s progress in AI and semiconductors by sanctioning the sale of advanced AI chips as well as key equipment used in the fabrication of leading-edge chips. But in fact this only helped accelerate China’s progress. It had the unintended consequence of solving a coordination problem for the Chinese semiconductor ecosystem.

Prior to the sanctions, the Chinese government had a long-standing goal to advance domestic chip capabilities. But the startups and university spin-out companies backed by the Chinese semiconductor “Big Fund” were largely a failure — they did not cap- ture significant market share from incumbent western suppliers.

It’s easy to understand why. In an industrial process as complex and difficult as semiconductor manufacturing, even Chinese fabs (chip manufacturing facilities) such as SMIC are reluctant to take a risk on a local startup supplier when the safer bet is to use already-existing equipment from a western or Japanese incumbent — i.e., the same equipment used by SMIC’s competitors such as TSMC and Samsung.

The sanctions changed all that. When confronted with aggressive US policies, Chinese chip-makers had to factor in the possibility that key components in their supply chain might be entirely cut off by the US. Suddenly the interests of small startup companies and big fabs in Beijing and Shanghai and Shenzhen became completely aligned. The big fabs had to take risks and assist in the development of domestic suppliers. Consequently, China’s semiconductor capabilities have advanced rapidly in recent years. In fact, China now has a completely indigenized supply chain in mature and legacy chips, which constitute the majority of revenue in the global chip industry. They still lag in leading-edge nodes, but have already surprised the Americans by producing chips using an innovative 7nm process. The Kirin series that appears now in many Huawei phones is a good example of this.

Even in advanced AI chips, the gap between the US leader Nvidia and Huawei is closing. Huawei’s Ascend AI chips, made using SMIC’s 7nm process, are only one generation behind what Nvidia offers. SemiAnalysis expects that more than one million of them will be shipped this year, in similar volumes to Nvidia’s latest B20 chip.

The Chinese company DeepSeek recently shocked the AI world by releasing models DeepSeek R1 and V3 that are similar in quality to the top models from OpenAI, Anthropic and Google. DeepSeek R1/V3 incorporate clever technical innovations: they required roughly ten times less compute and energy to train, operate much faster, and at perhaps thirty times lower inference cost. A sign of things to come: the DeepSeek team was almost entirely educat- ed in China, including both undergraduate and graduate school training.

My own startup company, Superfocus. ai, builds AIs for complex tasks such as advanced document analysis (for private equity firms for example) and customer service (voice and text support, food orders). We can give our AI a restaurant menu and it will take orders like a human waiter.

In the past we have used a variety of foundation models like OpenAI’s GPT-4 to build our AIs. Now we are very pleased to find that DeepSeek performs extremely well. It is a much lower cost replacement for GPT-4. Some American technologists are afraid that DeepSeek, which is released as open source, (anyone can download and use it for free) will become like Linux — the standard ecosystem within which applications are built.

Where does this leave America? Competition at the technological frontier requires infrastructure, financial capital and human capital. China has all of these from domestic resources and so, if the US wants to keep pace, it will need to import high-level talent from around the world. There’s no way of getting around it. Trump and Elon must do their utmost to forge a political consensus on immigration policy.

They must persuade the MAGA base that this is the only chance of making America great again.