China closed the salary gap. The United States never competed on salary, and the one thing it did compete on is now the hardest part of the offer to price.
The American research system buys its bench at a deep discount. A computer science doctorate who takes an industry job expects a median $180,000 in the first year. The same doctorate who takes a postdoc expects $70,000. That $110,000 spread is not a wage gap, it is a premium paid up front for a later American career, and 58.3% of the people paying it hold temporary visas. On September 15, the legal basis for that career stops being open-ended.
Every talent system runs on either cash or a promise. China has spent two decades trying to buy scientists with cash and mostly getting the second quartile. The United States has spent seventy years buying them with a promise, at prices no employer could otherwise get away with. The story worth watching is not that Beijing raised its bid. It is that Washington keeps repricing the promise.
What Happened
The New York Times published a piece on Monday by Li Yuan arguing that a combination of funding cuts, immigration restrictions and suspicion of Chinese-born scientists is eroding America’s advantage in scientific talent while China turns into a credible alternative.
The reporting is strong. Two of this year’s four Fields Medal winners were born in China, left around age twenty, and built their careers in the West. Omar Yaghi, who shared the 2025 Nobel in Chemistry, left Berkeley for Tsinghua. Mainland and Hong Kong universities have recruited at least eight prominent mathematicians from Western institutions since last year. Zhilin Yang took a Carnegie Mellon doctorate back to Beijing and built Moonshot AI, whose Kimi K3 now trades blows with American frontier models. Terence Tao had his own federal funding suspended and restored after litigation, and warns that abrupt disruption damages researchers’ trust in the system.
Two hard policy facts sit underneath the feature. On July 17, the Department of Homeland Security published a final rule replacing duration of status for F and J visa holders with a fixed admission period of no more than four years, effective September 15, 2026. And State Department data show F-1 visa issuance fell roughly 36% between May and August 2025 against the prior year, about 97,000 fewer visas, with India down about 60%.
The Times frames this as a competition for talent. It is more useful to read it as a compensation problem.
The Backstory
The United States does not recruit most of its scientists. It manufactures them, cheaply, out of people who arrive on their own initiative and pay for the privilege.
In 2024, American institutions awarded 45,929 science and engineering doctorates. Temporary visa holders earned 16,738 of them, 38% of the total, including 61% of doctorates in computer and information sciences, 54% in engineering and 51% in mathematics and statistics. On the bench behind them sit 69,877 postdocs, the highest count the National Science Foundation has ever recorded, of whom 40,746 hold temporary visas.
That is the actual structure of the American research enterprise: a majority-foreign junior labor force, working at stipends the federal government sets by notice. The NIH floor for a predoctoral trainee in fiscal 2026 is $29,364 a year. A postdoc with no prior experience gets $63,480.
Nobody takes that deal for the money. They take it because the terminal value has been extraordinary and reliable. Expected stay rates for temporary visa holders finishing S&E doctorates hit 85% in 2024, up from 75% in 2004, and reach 90% in biology and computer science. Measured retention runs lower than intent, which is worth flagging: NSF tracked the 2017 to 2019 cohorts and found 73% still in the country roughly five years out. The Times cites an 87% figure, but that is stated intent from a cohort that graduated between 2005 and 2015. Intent and residence are different measurements and they should not be quoted as the same number.
Either way, the trade held. Six lean years, then a green card, a lab, a company, a life. The lean years are the price. The option on the other side is the product.
The Plan
The current administration is not cutting the money as much as the headlines suggest. It is cutting the certainty, which is the input that actually clears this market.
On funding, the attempts mostly failed. The NIH plan to cap indirect cost recovery at 15% was permanently enjoined in April 2025, affirmed by the First Circuit in January 2026, and left to die when the government declined to petition the Supreme Court. Congress rejected the proposed cuts and wrote protections into the fiscal 2026 appropriations, freezing indirect rates at prior negotiated levels across NIH, NSF, NASA, Commerce and the Department of Energy. NIH landed near $47.5 billion.
On immigration, the changes are structural and they are landing. The new rule admits F students for the length of the program on their I-20, capped at four years, plus 30 days on each end. Students who need longer file for an extension of stay. The rule also restricts program changes and blocks a foreign graduate of a US degree from starting a second program at the same level.
Then there is the H-1B fee, which is the cleanest illustration of the problem. A $100,000 payment on new petitions for beneficiaries abroad took effect on September 21, 2025. A federal court in Washington upheld it in December. A federal court in Massachusetts vacated it on June 8, 2026 as a tax Congress never authorized. The same judge stayed his own order four days later, putting the fee back. The First Circuit declined to stay the vacatur on July 24. The underlying proclamation expires by its own terms on September 21, 2026 unless extended.
Five status changes in eleven months, on the instrument that converts a doctorate into a career. That is not a policy, it is a volatility surface.
The Business Model Angle
Three things follow, and they matter well beyond universities.
One. This is deferred compensation, and deferred compensation is priced on credibility.
Every founder who has ever paid in equity instead of cash understands the mechanism. You are not selling a job, you are selling a claim on a future state, and the discount you can charge today is a function of how believable that future state is. Raise the perceived risk of the claim and the discount collapses, which means you have to pay cash you do not have.
America has been running the largest deferred compensation scheme in the knowledge economy, and paying for it with something that never appeared on any budget line. The four-year admission cap does not shorten anyone’s stay. Median time from graduate school entry to a science or engineering doctorate ran between 6.3 and 7.9 years across broad fields in NSF’s own data. The rule inserts a mandatory renewal application into the middle of the median science PhD. That is a small administrative fact and a large pricing fact.
Two. China is not buying the option. It is paying cash, because cash is all it can offer.
Chinese universities now match European salaries and add relocation packages, which the Times documents through researcher Dong Jielin. The Qiming program, the successor to Thousand Talents run out of the Ministry of Industry and Information Technology, was reported to offer signing bonuses of 3 to 5 million yuan, roughly $420,000 to $700,000, plus home purchase subsidies. National R&D spending reached 3.93 trillion yuan, about $550 billion, in 2025.
Cash buys mid-career scientists with a track record to protect. It is much worse at buying the twenty-two-year-old who has not proved anything yet, which is exactly the population the American system has always harvested. That is the asymmetry the Fields Medal story illustrates and the recruiting story does not.
Three. The numbers everyone is quoting describe a stock, not a flow.
Record postdocs. Record doctoral enrollment. Record stay rates. All true, and all misleading, because a doctoral pipeline is deferred revenue: the people counted this year signed up three to seven years ago. The bookings line is first-time enrollment, and it broke early. First-time, full-time enrollment of temporary visa holders in science and engineering master’s programs fell 21.2% in 2024, from 77,702 to 61,263. That happened a full year before the visa collapse, and master’s programs are the feeder for a large share of doctoral admissions.
Any operator who has watched a subscription business report record revenue while new bookings quietly halved knows how this ends and how long it takes to show up. On the current structure, the American research system will post excellent numbers for years after the decision has turned against it.
The Risk
The strongest counterargument is that this is a narrative running ahead of the data, and it deserves a real hearing.
Congress refused the cuts. The courts killed the indirect cost caps. Stay rates are at record highs. Postdoc counts are at record highs. Full-time doctoral enrollment among temporary visa holders rose 5.3% in 2024. The system that is supposedly collapsing keeps printing records, and one bad summer of consular scheduling is not a structural break.
China’s own constraints are also real and under-reported. Basic research is 7.08% of Chinese R&D spending, a record share and still a fraction of what a Fields Medal pipeline requires. Stanford’s analysis of the Young Thousand Talents program found it recruited high-caliber but not top-caliber researchers: those who declined offers were more productive, far more likely to hold overseas faculty positions, and drew much larger grants. Returnees outperformed after arriving, though the advantage shrank considerably once funding and team size were controlled for. That is what buying with cash gets you. It buys availability, not the top of the distribution.
The sharper risk is domestic and it is not about China at all. The marginal alternative for a talented 22-year-old is not Beijing. It is Toronto, London, Zurich, Singapore or Hong Kong, none of which need to beat American science to win the trade. They only need to be more predictable than it for a decade. And there is a second-order risk in the numbers above: definite postgraduation commitments fell across every science and engineering field between 2023 and 2024 while postdoc rates rose. More new doctorates are being routed into the lowest-paid, most visa-dependent tier of the system at precisely the moment that tier’s legal footing became renewable by application.
Quick Questions
Does the new rule actually shorten how long students can stay? No. It caps each admission at four years and requires an extension application to continue. The effect is procedural risk in the middle of a degree, not a shorter degree.
Is federal research funding actually down? Not by much at the top line. Congress rejected the proposed cuts for fiscal 2026 and courts blocked the indirect cost caps. The disruption came from freezes, terminations and litigation rather than from appropriations.
Why does a $110,000 pay gap matter if postdocs have always been underpaid? Because the gap is only sustainable while the payoff is credible. It is compensation deferred into a residency outcome, and it is the cheapest research labor in the developed world for exactly that reason.
What is the single number to watch? First-time, full-time enrollment of temporary visa holders. It leads the doctoral and postdoc counts by several years and it turned negative before the visa data did.
Is China winning this? On money, it has closed most of the gap. On the institutional patience that lets an unproven 25-year-old fail productively for three years, the evidence in its own media coverage says otherwise.
The Business Model Analyst Take
Every business that pays below market for talent is selling something other than money, and most founders never write down what it is. Equity. Access. A reference. A shot. The moment your people stop believing in the thing you are actually paying with, you are competing on cash against people who have more of it.
America built the best talent business in the world by paying in a currency it never had to print. The current policy mix does not spend that currency, it debases it, and the debasement shows up in a place no quarterly report will catch for another five years. The pipeline is always cheaper than the spot market. When you throttle the pipeline, you do not save money. You buy at spot, and by then the other bidders have learned the price.
