The Long Bet · Article 3 of 5

What the World Actually Offers

We built the coverage matrix expecting to confirm what we already suspected. It didn't. Almost every assumption the research started with turned out to be wrong – not slightly off, but structurally wrong.

The goal was straightforward: map what 100 top universities actually offer across 31 STEM programme categories, then filter for the intersection that matters. Mathematics plus computational biology. Research-active institutions. Realistic cost. A-level entry.

The data had other ideas.

CS is Universal. AI is a Niche.

Start with the global picture. Computer Science appears at 95 of 100 universities. Physics at 94. Chemistry at 92. Mathematics at 91. These are effectively universal – if a university is in the top 100 by research output, it almost certainly offers them.

Now zoom in on the intersection this research was actually about.

Standalone AI is available at 12 universities.
CS is available at 95.
The subject everyone identifies as "the future" barely exists as an undergraduate degree at top research institutions. The gap between CS coverage and AI coverage is not marginal – it is a factor of eight. Most universities offering CS are not offering AI as a distinct programme. They are offering CS.
Programme coverage – universities offering each (out of 100)
Colour = subject cluster · Hover for exact figures · 31 canonical programmes across 100 ARWU 2025 universities

Computational Biology sits at 45 universities. Bioinformatics at 54. Real numbers, not vanishingly rare. But not what someone scanning degree titles would assume. The degree title you search for is often not the degree that exists.

One more data point: a programme literally called "CS & AI" exists at exactly 3 universities in the top 100 – UCL, Imperial College London, and City University of Hong Kong. The search term most students use to find "an AI degree" returns almost nothing. The label and the subject are not the same thing.

The China question

China has 13 universities in the top 100 by research output – 15 counting Hong Kong’s two. Fees run at $3,000–5,000 per year for international students. Several institutions – Tsinghua (#18), Peking (#23), Zhejiang (#24), SJTU (#30) – sit in the top third of the global ranking. On cost and prestige combined, China is one of the most interesting candidates in the dataset.

Two numbers tell the story.

0 → 13
Universities in the ARWU top 100  ·  2003 → 2025
China entered the ranking with zero universities in 2003. It now has 13 – more than UK (8), Germany (4), or France (4). No country has climbed faster.
100%
Electrical Engineering coverage
Every one of China's top-100 universities offers EE – vs a 72% global average. EE is foundational for AI hardware: chips, GPUs, signal processing. China isn't ignoring AI. It's building the hardware layer of it.
Programme profile – Chinese top-100 universities (incl. Hong Kong)
University App. Math Comp. Bio Bioinformatics Neuroscience Elec. Eng. Systems Eng. Materials Sci.
Coverage 7% 7% 47% 0% 100% 47% 67%
Global avg. 24% 45% 54% 35% 72% 12% 51%
Offered   Not offered

Compare that trajectory to Germany: 5 universities in the top 100 in 2003 (verified against the ARWU 2003 archive – Munich, Heidelberg, TU Munich, Göttingen, Free University of Berlin), 4 in 2025. Slowly declining for two decades. China moved from nowhere to second-largest non-US presence in the ranking in the same period. The momentum story matters.

China remains on the list. It needs a second look.

The geography no one considers

The standard international shortlist for STEM runs through the same destinations. Apply this research's filters and the picture shifts.

US and UK – the obvious starting point

The default destinations for research-active STEM at elite level. Strong programme coverage. High brand recognition. Exactly what the ranking is built to identify.

Problem: cost eliminates both without full scholarship. International fees at UK top universities run £30,000+ per year. US is higher still.

France – the country nobody mentions

4 universities in the top 100. Paris-Saclay (#13) sits in the global top 15 by research output – above Imperial, Manchester, and Edinburgh.

75%
Bioinformatics coverage across French top-100 universities (3 of 4)
Global average: 54%. Neuroscience: 75% (3 of 4) vs 35% globally. Data Science: 100% (all 4). France is not a niche option for this intersection – it is one of the strongest clusters in the dataset, though with only 4 institutions, each university carries a quarter of every percentage. In France the route is often a double licence (maths–biology or informatics–biology) feeding a specialised Master’s – a structure that suits a mathematically strong applicant. Licence-level teaching is in French, so language proficiency matters alongside the academic profile. Paris-Saclay (#13) sits above Imperial, Manchester, and Edinburgh by research output. Tuition at French public universities: ~€170/year for EU students; ~€2,770–3,770/year for non-EU internationals (often waived) – two orders of magnitude below UK or US fees.

France is structurally strong in exactly the programmes that matter for the math-bio intersection. It is almost entirely absent from the shortlists of international families making this decision. Those two facts do not usually appear in the same sentence.

Key programme coverage by country – AI, Computational Biology, Bioinformatics, Applied Mathematics (% of country's top-100 universities offering each)
Global avg shown as reference line per programme

Switzerland adds depth: ETH Zurich (#22) and EPFL (#44), strong in systems engineering and environmental engineering, at low fees relative to their ranking. Not cheap by Swiss standards, but well below UK or US equivalents.

Canada – the same pattern

3 universities in the top 100 (Toronto #25, UBC #53, McGill #76). 100% Computational Biology coverage – +55pp above the global average. Every Canadian top-100 university offers it. Fees for international students run substantially below UK and US equivalents. Canada sits in the same category as France: structurally strong in the right programmes, almost entirely absent from the shortlists of families making this decision.

Japan – the anti-pattern

Japan's reputation in technology is global. Its programme profile in this dataset is not what that reputation suggests. Data Science coverage: 0% – against a global average of 73%. Standalone AI: also 0% – against a global average of 12%. Not below average. Absent across the board. The two Japanese top-100 universities (Tokyo and Kyoto) are strong in Astrophysics (both offer it, vs 24% globally) and Nuclear Engineering – classical, physics-based science. The computational layer that sits on top of it is missing entirely.

One caveat cuts across every country comparison above. ARWU credits research to universities – not to national research systems. France runs much of its research through CNRS; Germany through Max Planck and Fraunhofer institutes. None of that output is credited to a university, so both countries’ top-100 counts understate what the national system actually produces. For this analysis the direction of the bias matters more than its size: France’s 4 top-100 universities are, if anything, an undercount of the research environment a student would enter.

The UK paradox

Cost efficiency removes the UK from consideration early. Under a strict cost filter, it disappears from the shortlist.

+76pp
UK universities' AI coverage lift above the global average
8 of the top 100 are UK universities. 7 of the 8 offer standalone AI – 88%, against a global average of 12%. The only one that doesn’t is Cambridge. Bioinformatics: +8pp. No other country cluster comes close on AI.

This is not a recommendation. It is a finding. The cost constraint is real. But any framework that eliminates the UK on cost and then optimises freely within the remaining options is working from an incomplete picture. The trade-off is explicit and does not resolve cleanly.

The institution you know is not always the degree that exists

The coverage matrix produced a tab labelled "Surprising Absences" – universities in the top 100 missing programmes you would assume they offer.

Harvard University
ARWU #1
No standalone Mathematics – available at 91% of top-100 universities globally.
MIT
ARWU #3
No standalone Electrical Engineering – available at 72% globally.
University of Cambridge
ARWU #4
No standalone Physics or Electrical Engineering – entry is through Natural Sciences.
University of Oxford
ARWU #6
No standalone Data Science or Electrical Engineering.
UCSF
ARWU #21
5 programmes total – all biomedical. No CS Mathematics Physics Electrical Engineering.
Rockefeller University
ARWU #29
4 programmes total. No CS Mathematics Physics Chemistry Data Science Statistics.
Karolinska Institute
ARWU #50
3 programmes total. No CS Mathematics Physics Chemistry Electrical Engineering.
Programme breadth – selected universities (out of 31)

These are not failures. Cambridge's Natural Sciences structure is one of the most rigorous scientific training programmes in the world. But UCSF at #21 – ranked above Edinburgh, Manchester, and Imperial – is a medical research institution. It does not teach CS or Mathematics. Neither does Karolinska (#50), UT Southwestern (#58), or MD Anderson (#86). All four appear in the top 100 by research output. None of them are universities in the sense a STEM undergraduate would recognise.

Filter by rank and you may be researching institutions that do not offer the subject you came for. The brand and the degree structure are separate things. Confusing them produces incorrect shortlists.

The coverage matrix, the country profiles, and the surprising absences together describe the landscape as it actually exists – not as brochures present it, and not as received wisdom assumes it.

What this means for a specific student, with specific subject choices, still depends on something the coverage matrix cannot answer.

That is a different question – and a different article.

Also in this series
The Problem Statement The four-constraint optimisation problem and why every obvious answer fell apart.
How I Built It Why ARWU over QS and THE, why 100 universities, and what 1,200+ normalised pairs made possible.
The Unfashionable Four Four A-level subjects. Stable across 31 programmes, 100 universities, 19 countries.
The Floor Keeps Rising Seven scarcity cycles since the 1880s. The compression curve that makes a four-year degree longer than a full cycle.