Amid the artificial intelligence wave, compensation for graduates of elite universities is surging upward.

Recently, several individuals familiar with the employment situation of Peking University doctoral graduates told reporters that PhD holders in AI-related fields are now commonly receiving first-year salaries reaching several million yuan, with some graduates commanding even higher package deals. As AI capabilities grow stronger, it is also creating a divide among university graduates, with those from non-elite institutions bearing pressure in this round of technological transformation.

Who Is Competing for Top Talent

Recently, Wang Hao, founder of an AI startup, told reporters that PhD graduates from top laboratories at Peking University and Tsinghua University can earn annual salaries reaching several million yuan if they join model research and development projects at major tech companies. The startup's co-founders include several current Peking University doctoral students. According to Wang Hao, the doctoral salaries mentioned here include both cash compensation and incentives such as stock and options. Meanwhile, some master's graduates from Peking University also command daily wages reaching several thousand yuan. Currently, the companies offering top-tier pay to Peking University and Tsinghua University graduates are major firms like Alibaba and ByteDance. ByteDance enjoys a proximity advantage, with its Beijing headquarters located less than 10 kilometers in a straight line from Peking University, making it a significant destination for Peking University graduates. According to Yang Ling, assistant professor at Peking University's International Machine Learning Research Center, many current Peking University students begin interning at ByteDance early on. Take ByteDance's 2026 "Top Seed" large model elite talent campus recruitment program as an example—the project plans to recruit approximately 30 outstanding doctoral students. The program focuses on frontier AI fields, covering large language models, machine learning algorithms and systems, and more. ByteDance hopes to use this program to find the top 5% in the large model field to accomplish what 95% of people cannot. Tencent and Alibaba are both competing with ByteDance for talent. Tencent has launched the "Qingyun Program" targeting top technical students globally. The program claims to provide customized training plans for graduates, open up work opportunities in core businesses, emphasize unlocking forward-looking technical topics while offering "highly competitive compensation." Salary standards leaked online for the Qingyun Program state "1 million to 3 million+, with no salary cap." AI is transforming every industry, and leading companies can find suitable talent from various schools at Peking University. In Yang Ling's view, Peking University has dense talent resources—undergraduates, master's students, and doctoral students are all extremely valuable talent resources, and the youngest talent is exactly what is most needed for developing AI. They may come from the School of Computer Science, the School of Intelligence, interdisciplinary schools, the School of Mathematical Sciences, or even the College of Engineering or the Health Science Center. Yang Ling received his doctorate from Peking University and was selected for ByteDance's "Top Seed" program during his doctoral studies. According to him, some doctoral students are already able to deeply participate in ByteDance research projects during their studies, and ByteDance also provides some computing power support for university projects. "Many interns on ByteDance teams now take on very important roles, and may even directly participate in model training tasks."

Human resources at major tech companies are also turning their attention to the high school stage. According to industry insiders, some companies now hope to discover talent earlier, so they have begun reaching out to high school students who have participated in the Informatics Olympiad or Mathematical Olympiad and won gold medals or been admitted to Peking University or Tsinghua University.

The Most Valuable People of This Era

What do these major tech companies hope to get from PhD graduates of Peking University and Tsinghua University? In Wang Hao's view, the hundreds of thousands of computing power cards held by Qwen represent an investment of nearly 10 billion US dollars. If a doctoral graduate can help a project optimize chip usage efficiency, then an annual salary of several million yuan seems inexpensive in the face of a 10 billion US dollar computing power investment. Qwen is a large model project developed by Alibaba. In mid-2025, Alibaba announced plans to invest 380 billion yuan over the next three years in AI and related cloud computing. Subsequently, Alibaba CEO Wu Yongming stated that capital expenditure would be increased on top of the 380 billion yuan. "Large models are now the most valuable track, and they are the most valuable people of this era," Wang Hao said. Yang Ling holds a similar view. "I think it's not a loss at all for ByteDance to pay several million a year, and it's not even a loss to pay some core researchers 20 million. Because running one experiment takes at least several hundred cards, or as many as several thousand cards, and the cost of one run could range from several million to tens of millions—the cost of a single mistake could cost hundreds or tens of millions," Yang Ling said. A top researcher can reduce multiple ineffective model training runs. ByteDance is also one of the most generous AI investors, with market rumors that the company's AI investment budget for 2026 is close to 160 billion yuan. ByteDance's short video platform Douyin has over 900 million monthly active users, and precisely pushing content that interests users is its core competitiveness; its AI application Doubao has over 300 million monthly active users and needs to provide rapid responses to user questions; ByteDance's Volcano Engine provides cloud services to enterprises. These important businesses all depend on AI and top talent. The doctoral graduates who receive high salaries cannot rest easy, however. "Many PhD students from Peking University and Tsinghua now even go directly to (major tech companies) to work on core tasks, directly replacing the previous generation of tech talent. This is also very dangerous: you go in now to do core tasks, and in two years someone smarter comes in and replaces you," Yang Ling said. This is also one reason he later chose to do postdoctoral research at Princeton University and then return to Peking University. Now Yang Ling both mentors doctoral students at Peking University and promotes entrepreneurial projects. Yang Ling predicts that five years from now, AI talent will not be as scarce as it is today. On one hand, the supply of university talent will continue to increase, and model companies will also cultivate their own batch of talent; on the other hand, industry development may also hit a ceiling, and demand for talent will not keep expanding indefinitely. AI is also creating a divide among university graduates, and the salary gap between graduates of top universities and those of mid-tier and lower-tier universities may widen. "Personally, I think in AI, there will eventually be no place for graduates from Top 20 or Top 15 universities. What these people can do, Codex can probably do too," Yang Ling said. Codex is an AI programming platform launched by OpenAI. Its strongest competitor is the Claude programming tool launched by Anthropic. The AI programming business has supported Anthropic's rapid revenue growth and ultra-high valuation, but it has also made the work of junior programmers dispensable. Yang Ling believes that AI's division of talent will become increasingly extreme. "This thing will gradually compress forward, compressing to the very top tier of people—only they can truly create high-value things, and others are completely unnecessary." This means universities also need to change. "Peking University is also changing, all schools are changing, and ultimately they must integrate with industry. If university laboratories don't integrate with industry and have no research funding, then students can't do things or learn anything," Wang Hao said. Ordinary schools must face and actively respond to the challenge of a downward spiral.