U.S-China AI talent race heats up

This report is from this week’s CNBC’s The China Connection newsletter, which brings you insights and analysis on what’s driving the world’s second-largest economy. You can subscribe here.

Chris Miller, author of the book “Chip War,” warned three years ago that the balance of modern power hinges on a semiconductor supply chain crossing geopolitical fault lines. Now, the Tufts University historian is raising a new concern: the U.S. risks losing its advantage over China in artificial intelligence talent.

When it comes to brain power, “America’s edge is deteriorating dangerously,” Miller told a U.S. Senate Foreign Relations subcommittee last week. It’s a lead that’s “fragile and much smaller” than its advantage in AI chips, he said.

Carnegie Endowment researchers echoed those concerns a day later, noting China has been producing more top-tier AI researchers over the last few years, while fewer are heading to the U.S.

Part of the difference comes from sheer scale, especially as education levels rise in China.

Its population is quadruple that of the U.S., and the same goes for the volume of science, technology, engineering and mathematics (STEM) graduates. In 2020, China produced 3.57 million STEM graduates, the most of any country, and far outpacing the 820,000 in the U.S.

The growth of China’s higher education graduates has been dramatic, increasing ninefold in just one generation. The share of adults with at least a master’s degree rose from just 0.1% at the turn of the century to nearly 0.9% two decades later.

The U.S., starting from a much higher base, has seen the same ratio climb far more steadily from 8.7% to 14.1% during that time.

The surge is reshaping companies’ pipelines. Most of Xpeng’s new hires are fresh graduates, CEO He Xiaopeng told reporters last month.

He claimed that despite a relatively small pool of AI talent in both the U.S. and China, the electric vehicle manufacturer was able to recruit 10 specialists in driver-assist technology this year.

The company’s former vice president of autonomous driving now heads Nvidia’s automotive department.

Beijing is engineering more of this momentum as it seeks tech self-sufficiency. The Ministry of Education said in August that one-fifth of higher education programs nationwide have been revamped in the last two years — with a slew of majors cut or added — to channel more students into AI and integrated circuits fields.

It’s not necessarily that China is much better at attracting global AI talent, but the ability to keep more AI experts at home “could have a major impact on talent flows,” Tufts University’s Miller pointed out to me in an email.

Meanwhile, U.S. immigration rules could make it difficult for AI researchers to work in the U.S., he said.

There’s still debate over whether volume can consistently create value.

U.S.-based OpenAI triggered the generative AI craze in 2022 with the launch of ChatGPT-3.5, and other U.S. companies, such as Anthropic, have released models that are considered global benchmarks.

China’s DeepSeek only made similar waves early this year with its claim to have beaten OpenAI at only a fraction of the cost. Google raised the bar further last month with its Gemini 3 model.

The U.S. AI models are officially unavailable in China, while Washington restricts exports of Nvidia’s most advanced chips.

Despite the restrictions, Chinese companies have increasingly found ways to use scale — whether in engineers, less advanced chips or data — to gradually build domestic AI capabilities. That’s apparent in how Chinese AI models rival OpenAI at a fraction of the cost.

Developers in China are also doing more with less.

Chinese internet leaders spent about 400 billion yuan ($56.58 billion) in capital expenditure this year — about one-tenth of their U.S. peers, while achieving “comparable” AI model performance, Wei Xiong, China internet analyst at UBS Securities, said in a report last week.

And while analysts sometimes warn of “circular financing” in the U.S. AI industry — buoyed by investor money that circulates within the same funding ecosystem rather than genuine commercial revenue — China’s leading developers, by contrast, are drawing primarily on internal cash flow.

Data abundance is another advantage. The popularity of short videos in China has given local companies a hoard of training material. Aside from two models from Google, the rest of the world’s top 15 image-to-video AI generation models came from Chinese companies, according to AI benchmarking firm Artificial Analysis.

Chinese companies have also posted the fastest growth in U.S. patent grants last year, with telecom giant Huawei ranking fifth overall among commercial firms.

The company’s global patent licensing revenue hit a record $630 million last year, with further growth expected in 2025, according to Huawei Vice President Alan Fan, who heads the company’s intellectual property rights department. He said clients are increasingly seeking not only 5G and WiFi-related patents but also audio and visual technologies.

Back home, Huawei is also partnering with universities for AI-related research and training.

All these efforts add to a growing trend.

China has ranked first in a U.S. index measuring student performance in STEM-related Olympiad competitions for several years. China’s scientific research contributions, as tracked by the publisher of the scientific journal Nature, surpassed the U.S. in 2023 — and quadrupled that lead the following year.

Yet talent still flows both ways. Many Chinese founders are still moving to Silicon Valley to launch startups, Lu Zhang, founder and managing partner of the locally based Fusion Fund, told me. She pointed to the benefits of operating within “one of the best enterprise ecosystems.”

Her bigger worry for the industry is insufficient electrical power, especially in the U.S. “I think before we run out of GPU we’re going to run out of energy,” she said.

And energy is where analysts increasingly say China has the upper hand.

In the trinity of factors that Miller identified in his Senate testimony for maintaining AI leadership —computing power, brain power, and electrical power — human capital isn’t his only worry. “America has a substantial lead in computing power, [but] China leads in electrical power.”

Kevin Liu, chief offshore China and overseas strategist at CICC, said there is still a way to go before AI demand, investments, and valuations reach the levels seen at the peak of the dotcom-era bubble.

Eswar Prasad, professor at Cornell University and former head of IMF’s China division, sees “no clear opportunity” for the U.S. and China to increase direct trade.

Paul Triolo, Partner at DGA-Albright Stonebridge Group, discussed U.S. chip policy and the prospect of the SAFE CHIPS Act becoming law amid intense chip competition with China.

Chinese markets fell Wednesday as investors parsed the latest inflation print. Hong Kong’s Hang Seng index slipped 0.56%, and the mainland CSI 300 declined 0.78% after China’s consumer prices rose 0.7% from a year earlier, its highest level since February last year.

Both benchmarks are set to break a two-week winning streak. The Hang Seng Index has dropped over 2% since Monday, while the CSI 300 is down 0.5% over the same period. For the year to date, the Hang Seng is up over 26%, while the CSI 300 has gained more than 15%.

The offshore yuan last traded at 7.0605 against the dollar, its strongest level since October 2024.

— Lee Ying Shan

Later this week: Central Economic Work Conference (expected)

Dec. 15: China to release November retail sales, industrial production and investment data

Dec. 16: Deadline for TikTok divestiture