What are AI interns who earn a thousand yuan per day anxious about?

A post-2000s graduate student who has not yet stepped out of campus earns a daily salary of 5,000 yuan as an intern, a figure higher than the monthly salary of many ordinary white-collar workers. If they become a full-time employee after graduation, their annual salary can even reach 3 million yuan. This is the real status quo amid the AI talent war.

According to disclosures from multiple media outlets, for core AI businesses of major tech firms including ByteDance, Tencent and Alibaba, the daily internship salary for PhD candidates ranges from 5,000 to 6,000 yuan. Even for ordinary AI intern positions, the daily pay stands at 500 to 1,000 yuan, with additional housing subsidies included.

What deserves more attention is the salary surge. Reports say that the baseline acceptable for an ordinary algorithm researcher last year was still above 1 million yuan, while this year the figure has risen to over 2 million yuan. Data from Zhaopin also shows that in the first half of 2026, the number of recruiting enterprises in the artificial intelligence industry increased by 26.3% year on year, and the number of recruited positions rose by 10.6% year on year. With demand and salary climbing simultaneously, it is not hard to understand why enterprises are willing to offer a daily salary of thousands of yuan to an intern who has not even graduated.

Who are the people getting these high salaries? We found five AI interns earning thousands of yuan per day: some, despite their major not being directly related to AI, landed core algorithm positions with their solid capabilities; some earn a daily salary of over 1,500 yuan under the talent program of a major tech firm, researching how to teach large models to complete long-horizon tasks; some have worked in the foundational model teams of multiple major tech firms, met interns with a daily salary of 2,000 yuan, and clearly seen the gap between ordinary people and geniuses; some started from non-computer science backgrounds and completed their career transition from data, algorithm to product through four internships; others who only earned 400 to 500 yuan per day in internships last year now get 1,200 yuan per day.

Some enjoy the flexible working arrangement with no clock-in requirement and no KPI, others are at the cutting edge of AI development, but most of them are questioning the same thing: how long can they stay in this "wealth creation" boom?

Below are their stories.

01. No clock-in, no meetings, but I am still very anxious

Zhang Mingming | Master's student in engineering at Tsinghua University

I am a current master's student at Tsinghua University, majoring in an engineering discipline not directly related to AI. I am currently doing an algorithm internship at a fintech company, earning over 1,000 yuan per day. In the half year since I joined, I have never attended a single meeting, written a single weekly report, and no one has ever assigned me clear KPIs.

The job hunting process was a bit tortuous. I initially applied for an algorithm position, but was transferred to the development team after joining. After communicating with my supervisor, I moved back to the algorithm post. At that time, I already had an offer from a top internet company, but after comprehensive comparison, the position and treatment at my current company are more attractive to me.

Before joining, I thought I would get a clear task sheet. The reality is that over the past half year, I have been exploring my work content entirely on my own, and the main reporting is to align progress with my mentor on a regular basis.

My job is to train some models with small parameter sizes to solve problems encountered in the company's internal business. This direction is still in the early exploration stage, the company classifies this model as a trial business, there are not many people who understand it, neither the company, my mentor nor I are fully sure that it can really work out, and how to assess my performance has also become a problem.

What I am working on is not a general model, public benchmark lists have limited reference value, there are not many objective indicators I can use, I can basically only refer to underlying indicators such as perplexity, plus an internally constructed test set. But there is a gap between these indicators and the actual business effect, so to judge whether the model is good or not, it mostly depends on subjective feelings, such as whether the response is smooth and whether the format is standardized, which you can perceive at a glance but is hard to score specifically.

Once, in order to get better model results, I tried to expand the parameter size from less than 1B to 2B, the effect did improve, but there is no clear conclusion whether further expansion can be put into practical use, and my supervisor did not say yes or no. At the same time, I have stepped into many pitfalls, for example, I did not pay enough attention to the quality of pre-training corpus in the early stage, which led to unsatisfactory model performance. But whether it succeeds or fails, I basically explore it on my own, the company will not interfere.

Compared with top internet companies, my current job is definitely not overloaded. I arrive at the company at 10:30 every morning, take a 1.5-hour lunch break, go for a walk after dinner at around 5 p.m., and get off work at about 8 p.m. The company provides free meals, taxi fare reimbursement, and there is no clock-in requirement.

As for salary, it is calculated by day and paid monthly. When I joined, it was 1000 yuan per day, and it was raised once later. This rate is not negotiated by me, the company treats all interns equally, and the algorithm post and development post follow the same standard.

People often ask if high-paid AI interns are driven up in value by the scramble of major tech firms? My feeling may be different. Every company has its own judgment on whether one's ability matches the pay. I also tried to apply for leading large model companies, but I didn't even pass the resume screening. They value matching published papers more, which I really don't have. This just shows that high salary is not created by the scramble of major firms, but depends on whether you can meet the standards of a certain company.

However, even if I meet the standards, my anxiety has not decreased at all. The development of AI is so fast. When I did my internship last year, I still needed to write code line by line, but this year I barely need to write code myself. This speed of development makes me unclear about the upper limit of AI, and I also worry that one day it may even be able to do the work of "researching AI" on its own.

02. Daily salary over 1500 yuan, no KPI, but the threshold is getting higher year by year

Lin Chuan | PhD student in AI direction at Peking University

I am pursuing my doctorate at Peking University, majoring in intelligent science and technology, mainly researching Agentic RL, and I have published several papers at top conferences.

At first, I didn't plan to do an internship so early. Later, a HR from Tencent contacted me and asked if I would like to try the Qingyun Program, then I started to pay attention to the talent programs of various companies, and applied to several of them. JD's recruitment process was very fast, I got the offer after two rounds of interviews, so I decided to work there for a period of time first.

Before applying, I had read discussions online about the treatment of various talent programs, so my salary expectation for JD was not that high. Later, the salary offered by HR was much higher than I expected. They implement a personalized salary system for each employee, my daily salary is over 1500 yuan, and the specific figure is not convenient to disclose. In addition to salary, the company also provides free talent apartments, two interns in the talent program can share an apartment of about 100 square meters. The meal allowance follows the company's normal standard, 20 yuan per person per meal, and dinner is free.

After joining, I first discussed with my direct supervisor what research directions the department has, and then chose the one I am interested in. I am very interested in Agent, so I chose this direction.

Here, interns are mostly working on exploratory projects, which are not closely related to specific businesses. My main work is to figure out how to use reinforcement learning to make models complete long-horizon tasks. I arrive at the company at around 9 or 10 a.m. every day, my daily work is mainly to train models, do research, check experimental results, and then improve the solution according to the results.

How to promote the work specifically needs to be broken down by myself. For example, at the beginning there was no environment to run code and train models, I had to build it first. If I find that the algorithm effect is not good enough, I will adjust the algorithm, retrain the model, and then observe the experimental results. These phased progress are also the content of each report. I report to my direct supervisor once a week, and then report to a higher-level supervisor once a month. The company has not set clear KPIs for me, the arrangement is very flexible, and we are not required to work overtime. However, I still have paper publishing and graduation requirements for my PhD study, so I will take the initiative to devote more time to these goals.

When the outside world looks at AI interns, the most common misunderstanding is to take the highest salary of a very small number of people as the treatment that everyone can get. In fact, there is a big gap between the salary of ordinary AI interns and that of interns in talent programs. Even if they are both working on models and algorithms, the specific salary is determined according to the university they graduated from, paper achievements, research direction, and whether they are the talents most needed by the company at the moment. But the work they do may not be very different. Take JD as an example, theoretically everyone can do similar tasks, the difference is more reflected in the completion speed and effect, and interns in the talent program usually complete tasks faster and deliver better results.

Now the recruitment threshold for talent programs is getting higher year by year. Based on what I know, the number of internship positions this year is less than that of last year, and the proportion of talent program interns who can be converted to full-time positions is also lower than last year. The number of positions is reduced, but the salary has not dropped, and may even be higher, because companies still have demand for top AI talents and are willing to continue to pay high prices, but they will reduce the recruitment of mid-level talents. As a result, high salaries will be concentrated on fewer people, and the threshold for getting these opportunities will be higher.

I am not too worried about being replaced by latercomers or failing to keep up with the capability requirements, because I feel that I am still making continuous progress. I am also confident about the prospect of AI development. In the future, I hope to work in a place that develops foundational models, and continue to research AGI.

03. After meeting the "AI genius", I began to question whether I should keep competing

Cheng Yuan | PhD student in computer science at a university in Beijing

I am a PhD student at a university in Beijing. Over the past few years, I have done algorithm internships in the foundational model teams of several major tech firms, and I have also been selected into the talent programs of major tech firms.

Foundational model teams are roughly divided into data, algorithm and architecture groups. I have worked on post-training, and also participated in improving model capabilities. When a task comes, such as improving a certain capability, I will prepare data, train the model, run experiments, and then adjust according to the results. Teams focusing more on research may eventually produce a paper, while teams with heavy business responsibilities will integrate the data and training results into the next version of the model.

Interns usually do not have clear KPIs. When the model has just been released and there is not much work to do, we can even write papers and do explorations on our own. Most of the time, this work is a mixture of scientific research and engineering practice.

My most direct feeling about this round of AI boom actually comes from salary.

At first, when ByteDance's TopSeed program offered interns a daily salary of 2,000 yuan with no upper limit, everyone thought it was exaggerated. At that time, ordinary algorithm interns might only earn 300 to 400 yuan per day. Now when we see a daily salary of 1,000 yuan, we already take it for granted.

But only a very small number of people can really get a daily salary of more than 2,000 yuan. I have indeed met people who "deserve this pay".

In the team where I did my internship, there was a student who completed his bachelor and doctorate degrees at Tsinghua University. He is very capable, and he really loves this work. When we finished our work during the day and wanted to play games at night, he kept working except for sleeping time, and his ideal is to "improve the level of intelligence".

The academic qualifications of people in foundational model teams are generally very competitive. The university I graduated from is also a 985 top university, but after entering such a team, I still feel that my background is not outstanding enough. In some teams, there are even more PhDs than masters. Even though everyone has gone through layers of screening, you can still clearly see the gap between ordinary people and geniuses.

This is especially true for people working on architecture and core algorithms. Those who are truly talented can see the structure and know where the problem lies at a glance. The adjustments they make are intuitive, mathematically sound, and really effective. In a foundational model team with hundreds of people, only a dozen people may really determine the development direction.

From the outside, it looks like everyone is "building AGI"; but after you step in, you will find that most people are still just ordinary workers.

So I think there was indeed some panic bidding in the market over the past year. Top talents are worth this pay at any time, but the rising salaries are not only for them. I know a person who has no published papers, only two internship experiences, and average capabilities, but finally got an annual salary package of nearly 1 million yuan.

Now the market has begun to calm down.

This year, the headcount of many teams has decreased significantly, some have even cut headcount from 100 to 50. The recruitment requirements seem to have not changed much, but the actual threshold has increased. Last year, one paper and one internship experience might be enough, but this year it also requires the papers, internship experiences and the job direction to be highly matched.

I am not too worried about not finding a job after graduation in 2028. What really makes me anxious is whether I should continue to compete with those "geniuses".

Working in the technology field is definitely fulfilling. The sense of value is very real when the model performs better because of your work, your paper is published, and you go to international conferences for exchanges. But this sense of achievement also has a price: high-intensity work, and you can hardly turn off your devices on weekends. Moreover, if I can't enter the most core team, I will also worry about what to do after I turn 35?

Another option is to