- According to the latest data guidance from The Economist, there is a structural shift in the core production factors of global artificial intelligence. The proportion of Chinese researchers at NeurIPS, a top conference tracked by the Carnegie Endowment for International Peace, has risen from 29% in 2019 to nearly 50%, offsetting some of the geopolitical friction costs on the hardware computing power front.
- Data from both industry and academia confirm an accelerated return of high-level intelligence. Scientists who previously worked at institutions like IBM, Microsoft, and Alphabet have intensively moved to Westlake University and domestic tech-related stocks, boosting the expected R&D efficiency of these enterprises.
- External policy disruptions have provided significant marginal catalysts. The H-1B visa approval rate in the United States has fallen to 11.7%, coupled with a 15% decline in the probability of domestic science and engineering students pursuing doctoral studies in the US, which is expected to reshape the valuation models and human capital pricing of the technology sectors in China and the US for the long term.
Core Data Mapping and Supply-Demand Pricing
Against a macro backdrop of restricted hardware capital expenditure, the Chinese market is attempting to mitigate computing power deficits through a very high density of human capital. The latest high-frequency labor market data show that demand for AI-related positions has expanded tenfold over the past year. This surge in demand has directly reshaped the labor pricing curve, with the average monthly salary for large model algorithm engineers exceeding 60,000 RMB. More crucial preliminary indicators show an extreme distortion in the supply-demand ratio, with the supply-demand ratio for high-performance computing engineers bottoming at 0.15, indicating a fierce seven-to-one competition for a single position. This severely tilted supply-demand structure not only increases short-term operating costs for enterprises but also suggests that capital is concentrating on foundational infrastructure and computing power scheduling at an unprecedented level.
Policy Variables and Friction Costs
The marginal tightening of US immigration and technology compliance policies is substantively altering the utility functions and career paths of top global talent. The 11.7% lottery rate for H-1B work visas, combined with increasingly stringent academic scrutiny, constitutes high implicit friction costs. In terms of data, the probability of Chinese STEM students pursuing PhD studies in the US has fallen by about 15%, and their willingness to remain in the US post-graduation has also decreased by 4%. This defensive retreat caused by policy uncertainties objectively cuts off the traditional talent siphoning pipeline of Silicon Valley. If the external environment continues to maintain a high-pressure stance, multinational tech giants may face a systemic re-evaluation of their R&D resource allocation between Washington and Beijing.
R&D Capital Expenditure and Efficiency Reassessment
The valuation logic of capital markets for AI companies is subtly shifting from focusing solely on GPU reserves to evaluating the comprehensive "computing power-human capital" conversion efficiency. New R&D entities like DeepSeek provide the market with a non-consensus sample. Despite having a team size of under 150 people and an average age of about 28, DeepSeek achieves model outputs comparable to GPT-4 with only one-tenth of the R&D spending of traditional leading firms. This operation mode, relying on high-density top-tier talent and extreme engineering optimization, proves that under certain constraints, human capital can produce unexpected leverage effects. This may prompt secondary markets to reassess the long-term free cash flow discount rates of some high-energy, high-capital-expenditure tech companies.
Marginal Risks and Long-term Pricing
Although short-term data presents a prosperous picture of net talent inflows, long-term fundamental constraints remain significant. The Carnegie Endowment's sample tracking highlights an unignorable tail risk: among the top 100 Chinese researchers, a staggering 87% still choose to remain within the US system. This means that the original innovation resources at the pyramid's peak have not fully transferred. Evaluations from the Chinese Academy of Sciences suggest that the domestic research ecosystem has a comparative advantage in the commercialization amplification stage from 1 to 10, but it continues to be under pressure for breakthroughs in the foundational paradigm from 0 to 1. If future industrial policies cannot effectively correct incentives overly biased towards practical use, the relevant tech sectors may face pressure on their long-term P/E central values after experiencing valuation repairs at the application layer due to a foundational technology ceiling.
