The computing premium stopped growing

S$5,500 was the median starting salary for a Singapore graduate in information and digital technologies in 2023.

It was S$5,500 in 2024.

It was S$5,500 again in 2025.

Three surveys, three years, the same figure to the dollar, while every other field of study moved up. Consumer prices rose 3.3% across those two years, so standing still was a 3.2% pay cut in real terms. Of the eight course clusters the Ministry of Education reports, technology is the only one that went backwards.

This is not a forecast or a vibe about the job market. Singapore runs one of the few graduate surveys on earth that publishes results down to the individual degree programme, every year, on a stable schema. The 2025 round covered 19,887 fresh graduates across all six autonomous universities, surveyed in November 2025 with a 72.4% response rate. Twelve earlier waves sit in an open dataset going back to 2013. You can check every number below yourself.

A decade of pulling away

Start at the beginning, because the flat line only means something against what came before it.

In 2013, a computing graduate earned about 4.5% more than the median graduate of every other field. That is a rounding error. Computing was a good degree, not a special one.

Then it separated. By 2016 the gap was 18%. By 2019, 20%. By 2022 it reached 37.5%, which is the widest it has ever been. In real terms the typical computing programme went from a median of S$3,752 to S$5,904 in nine years, while everything else crawled from S$3,590 to S$4,294.

At programme level the movement was steeper still. NUS Information Systems graduates earned a median of S$3,005 in 2013 and S$6,000 in 2024, a 69% rise after inflation. NUS Computer Science went up 61%. Two NTU computing degrees rose about 47%. Over the same twelve years an NUS music graduate ended up 4.6% worse off in real terms, and dental surgery fell 3.7%.

That is the run that made computing the default answer when a capable eighteen year old asked what to study.

The turn, and whether it is real

The premium peaked in 2022 and has fallen every year since: 37.5%, then 33.4%, then 25.6%.

A falling median is the kind of number that deserves suspicion before it deserves a headline. Universities kept launching computing degrees through this period, and new programmes often start lower. Add enough cheap programmes to the pool and the median drops without a single graduate being worse off. The fall could be an accounting artefact.

So test it. Take only the computing programmes that appear in every single year from 2020 to 2024, so the set cannot change, and run the same calculation. Ten programmes qualify. Eight of the ten paid less in real terms in 2024 than in 2022. NTU Computer Engineering fell 6.4%, NUS Computer Engineering 9.9%, SMU Information Systems 13.7%.

Then run the control, because a general graduate downturn would produce the same picture. Apply the identical treatment to the 83 non-computing programmes present in every year and they go the other way: up 5.6% in real terms, with only 43% falling after 2022 against 80% of computing.

It is not inflation, because every figure is deflated by the same index. It is not composition, because the programme set is held fixed. It is not the graduate market in general, because the control group rose while computing fell.

The 2025 survey then repeated the result on a completely different basis. MOE uses its own course clusters and pools graduates rather than taking a median of programme medians. Different taxonomy, different arithmetic, same answer: technology is the only cluster that lost ground.

The exception that explains it

Buried in the 2025 programme tables is the number that changes the interpretation.

While NTU Computer Engineering fell 6.4% and NUS Computer Science slipped 1.5%, NTU's double degree in Business and Computing rose 11.2%, from S$6,250 to S$6,950. It is now the best paid undergraduate programme in the country. NUS Business Analytics rose 5.6%.

Read those together and the story stops being "technology is worth less". Employers are still paying a large premium for computing ability. What they have stopped doing is paying a rising premium for computing ability on its own. The programmes that combine it with something else are still bidding up.

This is what a maturing skill looks like rather than a declining one. When a capability is scarce, having it is enough. When it becomes common, having it stops being a differentiator and the premium migrates to whatever is still scarce, which is usually the combination. The market has not decided that programming is less useful. It has decided that programming plus a second competence is what is now hard to find.

Quantity moved before price did

There is a second signal in the data, and it moved earlier and harder than pay.

Across all six universities, the share of technology graduates in full-time permanent work fell from 87.6% in 2023 to 82.7% in 2024 and 78.3% in 2025. That is 9.3 percentage points in two years, against a salary line that did not move at all.

This ordering is not an accident, and it generalises well beyond Singapore. Published salaries are sticky. An employer who has advertised a band, benchmarked against competitors and set internal equity does not cut it in a downturn, because the cost of doing so is visible and reputational. What that employer does instead is hire fewer people, hire them onto contracts, extend the probationary route, or slow the conversion to permanent. The price stays put and the quantity absorbs the shock.

If you are trying to read a labour market, the employment quality series will usually tell you what is happening before the salary series admits it.

Singapore's 2020 data shows how badly that can be misread. Overall graduate employment that year reached 94.0%, near the top of the whole series, and anyone quoting it would have described a strong market during a pandemic. Full-time permanent employment in the same cohort collapsed to 69.6%, the lowest in twelve years. The gap between the two measures opened to 24 percentage points against a normal four to ten. Both numbers are true. Only one of them describes what happened.

The signal arrives four years late

Here is the part that matters if you are choosing what to study.

The students who graduated into that S$5,500 in 2025 mostly chose their degree in 2021 or 2022. They chose it when the computing premium was at its all time peak of 37.5%, when every piece of available evidence said this was the highest return on four years anyone was offering. They were reading the data correctly. The data was simply about a market that would not exist by the time they arrived in it.

That lag is structural, not a mistake anyone made. A degree takes four years. The signal that guides the choice is the market at enrolment, and the outcome is the market at graduation. A cohort that responds strongly to a high premium arrives together, four years later, and increases supply exactly when they were told demand would be strongest.

There is corroboration from outside Singapore. Hacker News has run a monthly hiring thread since 2011, and counting the job advertisements in each August shows 854 in 2018 against 242 in 2026, a 72% decline. Two datasets with nothing in common, one a government survey in Southeast Asia and the other a comment thread on an American forum, pointing the same way.

The practical conclusion is not "do not study computing". Computing graduates still out-earn almost everyone at S$5,500 against S$4,400 for business and S$3,840 for arts and design. It is that a premium at its maximum is the worst available evidence about a premium four years out, because a maximum is the point from which the only available move is down.

Singapore is not an outlier

The raw Singapore figures were reported when MOE published them. What was reported was the ranking: information and digital technologies has been the best paid cluster for three years running. Coverage did not note that the number itself was identical each year, and none of it adjusted for inflation, which is where a flat line turns into a cut.

The pattern behind it has been found elsewhere, independently, and that matters more than whether Singapore's version was written up.

Revelio Labs published an analysis in May 2026 arguing that American computer science enrolment has passed its peak because the premium that drove it has shrunk. Their figure: a computer science graduate in 2016 earned about 19,000 dollars more than a mechanical engineering graduate from the same school, and by 2024 the gap was under 10,000. They also find the premium began weakening after 2022, the same year Singapore's peaked, and that enrolment tops out with the class of 2026.

The employment side matches too. Federal Reserve Bank of New York data puts recent computer science graduates at 6.1% unemployment, higher than philosophy. Handshake recorded entry-level software engineering postings down about 30% year on year in 2025. Two labour markets on opposite sides of the world, with different institutions and different survey instruments, moving the same way.

There is one genuine disagreement worth putting in front of you rather than hiding. The National Association of Colleges and Employers projects American computer science starting salaries up about 7% for the class of 2026, to roughly 81,500 dollars. That is a projection of offers rather than a measurement of outcomes, and it sits alongside separate reporting that technology workers with under two years of experience saw a second consecutive annual pay decline. Offers to the strongest candidates and outcomes across a whole cohort are different quantities, and they can move in opposite directions at the same time. Singapore's number is the second kind.

What this does not tell you

Some honest limits, because a number is only as good as what surrounds it.

This is a survey, not a tax record, so it carries the response bias that surveys carry. It measures pay roughly six months after final examinations, which is a starting salary and says nothing about what these graduates earn at thirty five. Salaries are gross monthly, the Singapore convention, and cannot be set beside annual or net figures from other countries without conversion. Singapore is a small, open, high income economy whose technology sector is unusually exposed to multinational hiring decisions taken elsewhere, so the size of the swing here is not automatically the size of the swing anywhere else.

And three years is three points. The flat line is real and it is confirmed on two independent methods, but a fourth year that turns back up would change what it means.

The number to watch

Not the median. The gap between overall employment and full-time permanent employment, which is currently 8.4 points and widening, and which moved two years before the salary did.

The 2026 survey lands in early 2027.

---

Sources. Graduate Employment Survey, conducted jointly by NUS, NTU, SMU, SUTD, SIT and SUSS and published by Singapore's Ministry of Education. Waves 2013 to 2024 from the open dataset on data.gov.sg under the Singapore Open Data Licence; the 2025 joint publication and the NUS and NTU programme level publications from MOE. Inflation adjusted using SingStat table M213801, Consumer Price Index, 2024 base year. Hacker News figures counted from top level comments in the monthly "Who is hiring?" threads via the public Algolia API.

Corroborating work cited above. Caelan Wilkie-Rogers, Computer Science Has Hit Its High Water Mark, Revelio Labs, 19 May 2026. Unemployment by major, Federal Reserve Bank of New York. Entry-level posting volumes, Handshake. Class of 2026 salary projections, National Association of Colleges and Employers.

Every Singapore figure was recomputed from the primary files rather than quoted from secondary coverage, and each one is re-derived by an assertion script kept alongside the data.