大厂、哈佛和斯坦福，都在被重新定价Big Tech, Harvard, and Stanford Are All Being Repriced

When knowledge and execution keep getting cheaper, why is an elite degree still worth $300K to $400K?

Once getting the work done gets cheap, what companies pay for changes. Getting the work done used to be the product. Now it isn't enough: they also need judgment about the industry, a feel for how the work actually flows, and someone who answers for it when things go wrong.

All three share the same problem. They take time, you have to be patient with them, they can't be taught in a classroom, and they only grow inside specific companies in specific cities.

So AI shrinks the gap between people who can and can't do the work, and widens the gap between which city and which company you're in.

Part one was about how schools teach (the supply side). This one is about what the market wants (the demand side).

01 What the market is betting on

I recently listened to a YC podcast released on September 17, 2026, called The State of Startups in 2026. At the mics were YC President & CEO Garry Tan, Managing Partners Diana Hu and Jared Friedman, plus Harj Taggar.

Diana Hu said on the show that about 5% of the companies they used to accept were solo founders, and now it's 18 to 19%. Garry Tan followed up: almost one fifth of the batch. It used to be one in twenty; now it's one in five. That's YC's own data, but behind it are companies picked out of thousands of applications a year and backed with real money.

Diana Hu added that this was the highest spike they had ever seen. Her baseline is what she called the past; against that baseline, it's nearly a fourfold jump.

Diana also gave three more numbers: hard-tech companies went from 8% of the batch to 20%. The show broke that down: robotics from 1% to 6–7%, industrial and manufacturing from 4% to 10%, defense from 1.5% to 5%.

Jared Friedman added that in the current summer batch, one in six founders has a PhD, more than at any point in YC's history. Revenue is growing faster too. Median revenue at acceptance is zero; median monthly revenue by the end of the batch used to be about $8000 and is now $20K, and some companies go from zero to seven figures in three months. What they sell has changed as well: the share of companies doing full-stack, end-to-end work, or simply completing a task outright, went from 10% to more than 25%.

Why are solo founders taking off? A co-founder used to be about complementary skills, especially on the building side. Writing code, prototyping, iterating on the product, chasing bugs, testing and deploying used to take several people working together, which meant organizing them. Now one person running a few coding agents really can get it off the ground.

That doesn't mean you'll never need a team, and YC is explicit that plenty of successful solo founders add people later. What changed is the timing. You used to need the team assembled before you could start; now you start running and find people along the way.

If that were the whole story, the conclusion would be that the AI era belongs to the young, since young people clearly pick up tools faster. But YC sees the opposite.

The official description they wrote for the episode includes this line: knowing what to build is becoming more important than simply knowing how to build it. The chapter list in the same description has a section titled, simply, why experienced founders are coming back.

Picture a 20-year industry veteran. He knows the customers cold, knows where it hurts most, knows what everyone tried over the past decade and what failed. But he can't code, and hiring a full engineering team is too expensive, so all that experience never turned into a product. Now AI has arrived and fills the piece he was missing. Everything he built up over all those years is still there, and with a little learning he can patch the weak spot himself.

Once AI narrows the gap between those who can and can't do the work, the gap between those with and without experience stands out instead.

The edge young people have is energy, picking up new tools fast, the ability and willingness to embrace change; but picking up new tools is exactly the thing AI is also doing for you. People who have been through a full industry cycle know what isn't worth doing at all, and models can't do that for you, especially when real business judgment is involved. A not-to-do list matters ten thousand times more than a to-do list.

What makes experience valuable isn't the twenty years themselves. It's how many wrong options and wrong answers the potholes you hit over those twenty years have crossed off for you.

Hard tech tells a similar story, and it's just as counterintuitive. Plenty of people assumed that with AI, starting a company would get lighter and lighter, and everyone would build software and agents. But robotics, chips, and data centers all sit on top of software, algorithms, simulation, and engineering teams, and AI is stripping those costs away layer by layer. Things only big companies could afford are now within reach of small teams. Once software gets cheap, capital flows to whatever hasn't gotten cheap yet: the physical world, domain expertise, real data, and industry judgment.

YC is now publicly putting experience up front. But these numbers measure the solo-founder share, the hard-tech share, revenue, and the share of PhDs; none of them directly measures how many years founders have worked. So, strictly speaking, what they reflect is that YC's preferences have changed.

02 Execution gets repriced first

On September 18, Databricks co-founder and CEO Ali Ghodsi went on the a16z podcast with partners Martin Casado and Sarah Wang. He was also one of the original authors of Apache Spark, and he has spent years helping big companies wire AI into production.

He talked about enterprise ontology (turning how a company actually works into a map of relationships that AI can understand), and his example was small. Two people of similar ability, one who started today and one who has been at the company five years: why does the second get things done so much faster? In his words, the one who has been there five years has an ontology of how that organization works, who the people are, how you get stuff done.

That map isn't in the org chart, and it isn't in the database. Who to go to for what, who is nominally in charge but can't actually move anything, which processes say one thing on paper and run another way in practice, why certain projects got killed, what the boss really cares about. Data tells AI what happened; this map tells AI how those things relate to each other.

So companies using AI today run into an absurd situation: in theory AI has all the knowledge in the world, but none of your company's common sense, or call it your workflow.

Later he got to organizations. A company is a tree, and information travels up and down it. If you can't make a call, you pass it up to your boss; the boss has to figure out what happened and fill in the context before deciding; once the call is made, it travels back down the tree level by level. With that ontology map, he said, a big part of this can now be handed to AI.

In terms of actual people: the employee reports to the manager, the manager adds context and reports to the director, the director compresses it again, and on up to the VP and the CEO. A lot of management roles spend their day gathering what's happening below, compressing it into one page, and passing it up. Many layers in a company are, at bottom, an information relay station built out of people.

It had to work that way, because there's a limit to how much any one person can hold. But if AI knows every meeting, project, person, number, and piece of history in the company, then knowing what's going on rapidly loses its value.

It used to be that the higher your title, the more you knew. Once context is something anyone can look up, status built on information asymmetry starts getting challenged. What changes is this: who can make the call with incomplete information, who dares to kill a direction that already has money in it, who dares to take the blame when something goes wrong.

By that logic, what gets squeezed first may be management work that only gathers, compresses, and relays information. Ghodsi himself didn't spell that step out, and the data we have only shows junior hiring; whether middle management is actually shrinking can't be confirmed yet.

The data we can see is at the junior level. Indeed Hiring Lab data from July 2026 show that in Q1 2026, entry-level roles made up only 4.5% of software development postings, while senior roles made up 69.3%. SignalFire's June 2026 report found big tech hired about 65% fewer new grads than in 2019, and early-stage startups 76% fewer. A Stanford Digital Economy Lab paper updated in August 2026 found employment for 22- to 25-year-olds in highly AI-exposed occupations was 19% below where their peers' trajectory would predict, mainly because of fewer hires, not layoffs.

Going by the data alone, execution is what gets repriced first.

But if entry-level roles are only 4.5% of the market, where do senior engineers come from ten years from now? The people sitting in today's 69.3% of senior roles are the juniors of ten years ago who worked their way up step by step. The parts of entry-level work that look least efficient, fixing other people's bugs, reading legacy systems nobody wants to read, arguing back and forth with customers over requirements, are exactly how a person slowly comes to understand an industry.

AI may now be narrowing that path. Companies save on training costs today and owe the people ten years from now. And there's a difference here I've seen firsthand across different kinds of organizations: professional services firms and large corporates won't absorb this the same way.

Firms like BCG have plenty of levels. Not long ago I was catching up with a close friend and heard for the first time that above partner there are different grades of partner, which made my head spin. But the production line there has always been an apprenticeship: new hires work on projects, move steadily closer to clients, and build judgment and trust one engagement at a time. As AI eats the bottom layer of analysis, it may thin out that training pipeline along with it.

Corporates are different. Big companies have tens of thousands, even hundreds of thousands, of people. Cut the junior layer and how do middle and senior people get promoted? You can't parachute everyone in, can you? In theory you could, but a corporate is fundamentally a trust problem, not just a capability problem. You can parachute people in or outsource everything, but accountability and trust still have to rest with real people; someone has to hold cross-functional work and responsibility together. With far fewer junior roles, what's missing in ten years may be not just senior engineers but people the organization is willing to hand decisions to. Outside hires can fill headcount, but they can't instantly replace the credibility someone has built inside a company over years.

A company's problems are very often organizational problems, and an organization is just its people.

Who's writing the code, at least, is clear. Google's official blog said on April 22, 2026 that 75% of Google's new code is now AI-generated and reviewed by engineers, up from 50% in fall 2025; in October 2024, Pichai's words were more than a quarter. Ghodsi is blunter: more than 90% of the software at Databricks is written by AI.

Ghodsi also has a view that cuts against this piece: he doesn't think this road leads to superintelligence. He said he sees no evidence that we're heading toward superintelligence; what he observes is that training is getting longer, more brittle, and more people-intensive. Coming from someone who sells an AI platform, that carries more weight than it would from a safety researcher.

03 Skills you can write down

On July 17, Garry Tan gave a talk at the AI Engineer World's Fair titled Every company should have a Brain.

In 2013 he was an engineer at YC building an internal social network, and by his own count, only about 14 lines of usable logic per day actually stayed in the project. Today he barely writes code full-time, and by his account his output is up about 400x. Here's how he put it: the people getting 2x and the people getting 100x are using exactly the same Claude, the same weights, the same context window, the same API. So the difference isn't in the model; it's in how you connect the work. The verb he uses is wire.

Then he mapped the parts of an agent system onto the parts of a company, every last one. A skill file is an employee: it has one capability and one job, written clearly enough that someone else could follow it. The resolver table is the org chart, deciding who handles each incoming task. Filing rules are internal processes. Trigger evals are performance reviews.

It works the other way too: any part of an employee's work that can be written down clearly enough for someone else to follow can become a skill file. What people still add is concentrated in handling exceptions, updating the rules, and owning the consequences.

He mentioned that in the Winter 25 batch, a quarter of the companies had codebases that were 95% AI-generated, and that batch became the fastest-growing, most profitable batch in YC history.

But he didn't stop at writing the organization into files; he went on to a harder question. However big the context window gets, it can't hold an entire company. Your company is a library, he said, not three books spread out on a desk. What decides whether your agent is a genius or a goldfish is who is picking which three books go on its desk right now; how smart the model is comes second.

A company brain isn't as simple as setting up an internal knowledge base. Retrieval keeps getting cheaper. The hard part is what deserves to be written down, what should be thrown out, and which fact wins when two of them conflict.

What stings for a lot of companies is this: you spend your days on which model to pick and how to set up a vector database, but the most valuable knowledge your company has may never have entered any system. It's still sitting in the boss's head, the top salesperson's head, the ten-year veteran's head. When an employee leaves, the company may lose its memory. Every company has this kind of amnesia; nobody books it as a loss.

When someone with eight years at the company leaves, the company counts recruiting costs and handover time. What the map that existed only in their head was worth is hard to quantify.

His advice is concrete: never do one-off work. You ask AI to write a plan; the first draft isn't good so you fix it, the second isn't good so you fix it again, finally you're happy, and most people stop there. He says that's wrong. There's one more step: turn the method that just worked into a skill you can call again and again. If you have to teach AI the same thing a second time, the first time was a failure, because you finished the task without keeping the capability.

His words: Model quality is rented, but if you build your brain, you own that brain.

Put what YC, Ghodsi, and Garry Tan said side by side, and the market now wants three things: judgment about what to do and what not to do, the know-how that only exists inside a company, and people who will answer for it when things go wrong.

All three take time to build. A school can teach frameworks and cases for judgment, but it can't help an individual make real calls inside a real company again and again, get some of them wrong, and live with the consequences.

So why should a school charge you $300K-plus, even $400K?

04 All three live somewhere specific

A school can't give you those three things directly, but it can shape or even decide who you spend your days with and which companies' doors you can walk through after you graduate.

A city offers a density of chances to run into opportunity; whether those chances turn into experience depends on how a company organizes its work. A school can get you into a company, but whether anyone mentors you once you're in, whether you get near customers, and when you're handed real decisions determine what a degree is finally worth.

Ghodsi says the person with five years at a company has a map of how it works. That map comes from the people you run into every day, the things you overhear, the potholes you hit. And who you run into and what you overhear depends on where you are, who's in the same building with you every day, and which companies' doors you can walk through after graduation.

So to judge what a city is worth to a student, look at four things: whether there's a group of people next door or nearby who can build things right (people), whether there's funding (capital), whether there's a real industry (industry), and how many jobs there are (jobs). The full table is in the appendix at the end.

According to the Venture Monitor from PitchBook and NVCA, in 2025 the Bay Area took $185.1 billion across 2702 deals, New York $35.6 billion across 1809, followed by Los Angeles, Boston, Seattle, Philadelphia, Chicago, and Pittsburgh. And then nothing. New Haven, Charlottesville, Hanover, and Ithaca don't appear once in the full text of all six Venture Monitor reports.

Yale is in New Haven, Darden in Charlottesville, Tuck in Hanover, and Cornell's main campus in Ithaca. The cities these four schools sit in have no name in the report. Not being named doesn't mean there's no investment locally; it means they don't clear the report's threshold for being named.

Industry looks similar. New Haven has exactly one Fortune 500 company, Amphenol, a connector manufacturer headquartered in Wallingford, ranked 198th on the 2026 list. Charlottesville has zero, and that zero is verified, not a gap in the search: Virginia Business magazine's August 2026 list of in-state companies has none in the Charlottesville metro. A Darden student spends two years in this city and pays $200K-plus in tuition, and in those two years never walks past a single Fortune 500 headquarters.

The employment mix of college towns tells you even more. In Ithaca, educational services plus health care and social assistance make up 52.0% of QCEW-covered jobs; Hanover 43.1%, Charlottesville 42.4%, New Haven 38.8%, with private educational services alone accounting for 30176 jobs there, Yale being the biggest employer among them. In Ithaca, one of every two jobs QCEW counts is in education, health care, or social assistance.

Whether that number is good or bad depends on where you plan to stay after graduating. Those who want to stay will find the city's employer base leans heavily on the university and the hospital. By the BLS CES count for January 2026, Ithaca has about 50500 nonfarm jobs, about 23400 of them in education and health; the other half are in other industries, just none that comes close to the university in size.

But few people stay in Ithaca anyway; Johnson graduates mostly head to New York, Boston, and Chicago. So that 52% describes what a student walks past every day for two years, not where they'll work. The city you study in decides what's around you, not necessarily where you'll work.

New York has 328950 computer jobs; Ithaca has 1280. Same state, a 257-fold difference.

At least two of these four data sets will mislead you depending on how you measure.

The first is jobs. This column has to be measured by occupation; measuring by industry will fool you. The BLS information sector classifies employers by their main line of business, so Amazon counts as retail, and software engineers at banks and hospitals don't count at all. By industry, Seattle's information sector is only 6.16% of local employment, behind San Francisco's 9.18% and San Jose's 8.29%. By occupation, Seattle has 194970 computer jobs, more than San Francisco's 156010 and San Jose's 149970.

Same Seattle, different method, and it goes from last of the three to first.

The college-town numbers have a similar problem. The percentages above use the BLS QCEW basis. CES counts public university staff as government, while QCEW counts them as education and health. The same Charlottesville is 32.6% government under CES and 42.4% education and health under QCEW; pick the wrong one and the conclusion flips.

The second is rankings, and there are three different yardsticks.

CSRankings counts papers at top conferences over a ten-year window: CMU 1st, MIT 5th, Stanford 12th, Harvard 39th, Yale 42nd. The US News graduate CS ranking is a peer-reputation survey: Stanford, MIT, and CMU tied for 1st, Berkeley 4th, Harvard 19th. And in the 2027 Best Colleges undergraduate ranking US News released on September 22, 2026, MIT is No. 1 overall for the first time, ending Princeton's run at the top that began with the 2012 edition, while also ranking first in undergraduate engineering, computer science, and economics.

The same MIT gets three different spots on three lists: 5th, tied for 1st, and 1st.

The two rankings measure different things. The US News rank comes from a survey and measures the impression faculty at other schools have; the CSRankings rank looks at how many papers were published at top conferences over ten years. One is reputation, the other is output. And output tracks headcount: on the CSRankings roster, Stanford CS has 84 people, 61 of whom published in those ten years; CMU has 274 and 184, three times Stanford. So if you're looking for an advisor or people in the same lab, look at CSRankings; if you want a title that people will recognize, look at US News.

The 2027 edition added a metric called Earnings by Major, which measures graduates' earnings four years out by major, using the Department of Education's College Scorecard data. It's worth 5% of the total score and replaced the old graduate debt metric. The yardstick for a degree went from how much students owe to how much they earn.

Princeton losing a top spot it held for more than a decade and the ranking switching yardsticks happened in the same year. How much the new metric contributed hasn't been broken out publicly.

There's another layer we haven't been able to find yet. Of the seventeen schools, only seven publish their CS department's enrollment and faculty headcount online, mostly public universities; Michigan lists 2666 undergraduates, 609 graduate students, and 96 core faculty. For the best-known private schools, the size can't be found. So for now these rankings can't be adjusted per capita.

So location matters a lot more in 2026 than it did in 2016.

The reason isn't hard to see. Once getting the work done gets cheap, what's valuable is understanding an industry, and that understanding can only be built inside a specific industry. The ontology map Ghodsi described can only grow inside a real company; you can't read it in a library. YC's hard tech going from 8% to 20% and robotics from 1% to six or seven percent point to the same thing: money is moving toward the physical world, domain expertise, and real data, and all three live somewhere specific.

And the industry doesn't have to be tech. Detroit's autos, Pittsburgh's robotics and health care, North Carolina's pharma all count. Chicago is a good example: only $2.6 billion in venture capital, the least among the top ten tech metros, yet 30 Fortune 500 headquarters, in the top tier nationally. The industry experience an MBA student wants, Chicago has, just not in tech.

But the Bay Area is still the Bay Area. In the Bay Area, a student can sit in class in the morning, drive forty minutes in the afternoon to meet a VC or look for a co-founder and build a team, then have dinner in some big tech company's building before heading back to campus. In one day they touch collaborators, capital, industry, and job opportunities.

The four data sets above all measure how much there is; they can't capture that kind of convenience. Among the cities these seventeen schools are in, I count only three where you can live a day like that: San Francisco (the Bay Area), New York, and Boston, and their strengths happen to be three different fields: software, finance and media, and biomedicine and hardware.

In October 2025, Wharton signed a long-term lease on a five-story building at 555 California in San Francisco's Financial District, about 80000 square feet, more than double its previous footprint, its first standalone campus in San Francisco. The counterexample: Northwestern decided to close its San Francisco campus in spring 2026. Kellogg's building is gone, but Kellogg's San Francisco Immersion remains; students still go to the Bay Area for the winter quarter and intern at VCs or VC-backed companies. Even the school that pulled out only saved on rent; the students themselves still have to physically go to the Bay Area.

A course catalog can be copied; this radius can't. That holds for CS153's speaker list, and it holds for Bay Area business schools too.

05 Already concentrated, or getting more concentrated?

Everyone says AI lowered the bar for starting a company, but in reality it hasn't. Opportunity hasn't spread out along with AI. Four unrelated data sets all show the top taking the lion's share.

In 2025 the Bay Area took $185.1 billion of a national $339.4 billion. On PitchBook's 2026 ranking of MBA founders, companies founded by alumni of Harvard, Stanford, and Wharton have raised $330.6 billion in total, while the seventeen schools ranked 4th through 20th add up to $329.9 billion: the top three equal the next seventeen combined.

Endowments in fiscal 2025: Harvard $56.9 billion, Yale $44.1 billion, Stanford $40.8 billion, $141.8 billion for the three, versus $119.6 billion for the next eight combined; Harvard alone is 6.3 times Dartmouth. In July 2025, Brookings sorted 387 US metro areas into six tiers. The top tier, Superstars, has just two, San Francisco and San Jose, and the top 30 metros account for two-thirds of AI job postings nationally.

All four are concentrated at the top, just measured in different units, cities, alumni, schools, metros, which don't stack into a single map.

That said, endowment concentration was built over decades, and Harvard's $56.9 billion has nothing to do with AI. The geographic concentration of venture capital and of funding raised by alumni-founded companies looked like this before AI, too, and that is exactly the head start these schools carry into the AI era.

Only two of these have a time dimension. One is the Bay Area's share of US venture capital: 54.5% for all of 2025, 77.5% for the first half of 2026. The other is SignalFire's new-grad hiring, 65% to 76% below 2019.

The bar came down, but it didn't flatten the distribution, and on the two measures that have a time series, concentration is still rising. Whether the middle is being hollowed out, all we can see so far is the shape of the distribution; it needs more watching and checking.

The difference between doing small things and big things is worth a word. AI lets a three-month-old company reach seven figures in annualized revenue; that's YC talking. AI has also turned frontier models into a game that takes billions of dollars of compute; that's true too. Both are true. The difference is: how big a thing are you planning to do?

The number of venture deals gets closer to what you actually want to know; behind the Bay Area's 2702 deals is a crop of companies that just raised money. But PitchBook counts financing transactions, and hiring needs numbers like SignalFire's. Raising money doesn't mean hiring; the money may go to compute, acquisitions, or just staying alive, and one company can raise twice in a year.

Solo founders are on the rise, but that doesn't mean you'll never need people; the point where you need them has just moved later. You don't have to wait for the team before you start, but halfway through, you still have to find the right person. The question is how likely you are to run into them within a year, not how many CS professors the school has.

06 Measure only software, and you misread a set of schools

The four data sets above, collaborators, capital, industry, and jobs, all measure how close a city is to software. Using only that one yardstick may be a little one-dimensional.

Take Dartmouth. By the location data, it's weak on all four: Hanover doesn't even have an OEWS computer-jobs series. Yet 41% of its class of 2025 went into consulting and 27% into finance; nearly 70% took a path that has nothing to do with whether there are programmers nearby.

Second is Cornell. Ithaca's 1280 computer jobs are the fewest of all seventeen, yet likewise more than 40% of its graduates went into finance. Cornell's path leads to New York.

Third is treating tech as the same thing as CS.

AI rewrites any industry with lots of repetitive judgment and lots of data, and biomedicine has both: protein structure, drug discovery, clinical trial design. That's where AI has landed hardest these past few years.

NIH funding in fiscal 2025: Michigan $724 million, 4th nationally; Penn $723 million, 5th; Yale $680 million, 6th; Stanford $644 million, 8th; Duke $624 million, 10th; Columbia (medical) $594 million, 11th; UCLA $516 million, 15th. Further down: NYU School of Medicine $438 million, 19th; Northwestern's Chicago campus $406 million, 21st; Weill Cornell Medicine $317 million, 29th; the University of Chicago $274 million, 37th; the University of Virginia $231 million, 45th.

Yale's CS ranks 42nd on CSRankings, and only 8.3% of its class of 2025 went into tech. By CS rank it's second to last among the seventeen, ahead only of Dartmouth; by share going into tech it's last. On the NIH dimension, it's 6th in the country. Both numbers are true.

CMU, MIT, Berkeley, and Dartmouth are all on that list too, just outside the top fifty; the first three have in common that they lack their own medical school and affiliated hospital system. Harvard is a measurement trap. Harvard Medical School itself ranks 63rd with $148 million, which looks unimpressive, because Harvard's clinical funding flows through its affiliated teaching hospitals: Massachusetts General Hospital 9th at $643 million, Brigham and Women's 20th at $414 million, Boston Children's 42nd at $246 million. But NIH treats those three as separate award recipients, so adding them up and comparing the total to Johns Hopkins alone mixes different institutional boundaries; that sum can't be used as a ranking.

Ranked institution by institution, Harvard sits low; measured by what the labs around Boston are actually doing, its scale gets chopped up by this list. Add this dimension and the seventeen schools fall roughly into four groups.

On the NIH dimension, every university behind the seventeen schools has a number, but the same university often appears as several separate entities: Weill Cornell in New York and the main campus in Ithaca are two entries, and so are NYU and Northwestern. Below, I consistently use the medical entry, because this dimension measures clinical and biomedical work.

Group one: dense local industry. GSB, Haas, CBS, Stern, Anderson, HBS, Sloan, Booth, and Kellogg are in CompTIA's top ten US tech metros; the other eight are outside it. Students in this group don't have to wait until graduation; they get firsthand exposure while still in school.

Group two: thin local opportunity, but an alumni network that carries people out. Tuck and Johnson are the cleanest examples. Hanover doesn't even have a computer-jobs series and Ithaca has only 1280 jobs, yet one sends nearly 70% of its students into consulting and finance and the other sends more than 40% into finance. What they sell is a pipeline built over decades that sends students into consulting and finance; local opportunity was never the selling point.

Group three: weak on tech, strong on medicine. NIH funding reflects the size of a university's medical system: Wharton (Penn, 5th), Ross (Michigan, 4th), Yale SOM (Yale, 6th), Fuqua (Duke, 10th), plus Darden (UVA, 45th). None of them are in the top ten tech metros, but their universities are all in the national top fifty on biomedicine.

Group four: one standout strength. CMU has the best CS in the universe, with a CSRankings count of 21.0; the runner-up doesn't reach 60% of that. But CMU has no medical school and no law school, and Pittsburgh has only 34640 computer jobs.

Technology can grow locally while decision-making power and capital stay somewhere else. Argo AI grew out of CMU's talent pool; Ford and Volkswagen put in billions of dollars, it peaked at a $12.4 billion valuation, and in 2022 both stopped investing and the company shut down. Of its two thousand people, the city couldn't absorb many locally. On September 30, 2026, Citadel founder Ken Griffin announced a $3 billion gift commitment to CMU: $1 billion for CMU's Pittsburgh campus and $2 billion to build a new CMU Miami campus. Miami is where Citadel moved its global headquarters in 2022, and CMU's explanation is that information technology and finance are more concentrated there, which can bring faculty and students different partners. Laurence Ales, interim dean of the Tepper School of Business, said you would think a venture capitalist can invest anywhere, but geographic distance plays a role; it's a relationship business.

Darden shows best what this dimension changes. Charlottesville has zero Fortune 500 companies, no name in the venture reports, and 4040 computer jobs; by the first four measures it's the weakest of the seventeen. Yet UVA's NIH funding is $231 million, 45th in the country.

Zoom out, and four of the seventeen aren't full universities. MIT and CMU have no medical school and no law school, Berkeley has no medical school, Dartmouth has no law school. The other thirteen, Stanford, Harvard, Yale, Penn, Columbia, NYU, Northwestern, Chicago, Duke, Cornell, Michigan, UCLA, and Virginia, have all five, business, law, medicine, CS, and engineering, though Chicago's engineering school, the Pritzker School of Molecular Engineering, only arrived in 2019, starting late and narrow.

AI won't rewrite just one industry; it will rewrite them one after another, and a school with broad coverage gets another opening each time.

But there's a ready counterexample: the 2027 undergraduate ranking mentioned in the previous section. MIT is No. 1 overall, and it has no medical school and no law school. Beyond those three firsts, it also ranks first in the combined category of biological computing, bioinformatics, and biotechnology, tied with CMU.

MIT took a different road. It didn't spread itself across every field; it made the few it has cross over with each other. With no medical school, it still claimed the computational half of biology. So having fewer fields isn't necessarily a disadvantage, but it takes cross-disciplinary work between them to make up for the gaps.

There's one more dimension. Dartmouth's undergraduate college is a traditional Wall Street target school: banks come to recruit on campus every fall, the school runs treks to New York, and its alumni network is famously tight even by Ivy League standards. Tuck's MBA plugs into that network. It has nothing to do with CS rankings, nothing to do with how many programmers are nearby, nothing to do with NIH funding. It was built over decades by people picking up each other's calls, and whether an alum picks up your call depends on whether they see you as one of their own.

Bourdieu called this social capital. It takes time, can't be bought, and can't be copied in a semester. No public data set measures it, but not being measurable doesn't mean it doesn't exist. It may be the dimension where these seventeen schools differ most, and the one AI will find hardest to move.

Where these seventeen schools land in the AI era depends on which industry each one has bet on, and when that industry's turn comes.

07 What alumni can do for you

Beyond schools, cities, and money, there's one group that hasn't been counted: alumni. Countries, organizations, universities, companies: all of them rise and fall on people.

A top school, especially HYPSM (Harvard, Yale, Princeton, Stanford, MIT) and the Ivies, doesn't stand alone. Behind it are the CEOs, partners, founders, professors, and trustees who walked out of a history more than a century long. For them, their alma mater isn't just the past tense; it's a credential still in circulation. The stronger the school, the more that line on their résumé is worth; if the school keeps declining, that credential gets marked down with it. Beyond that self-interest, there's identity, reputation, and the wish to leave something for the next generation.

Right now they sit on both sides of the table. On one side, they donate, provide compute, endow chairs, bring their companies' real projects into the classroom, and sit on boards telling the school what's happening in industry. On the other side, they're buyers: they offer internships and jobs, keep campus recruiting going, become customers for student startups, and keep trusting their alma mater's screening when hiring standards get murky.

Funders, advisers, channels, and customers are the same group of people, and that's the hardest thing to copy about a top school. Open courses and AI tutors can copy the curriculum, but they can't copy this flywheel that has been turning for more than a century: the school has a reputation, so it attracts better students; graduates rise to higher positions; those people send back money, information, and opportunities; the school uses those resources to train the next class; the reputation keeps growing. Alumni helping each other works the same way VCs pull each other into club deals.

What AI may be choking off is the middle link: graduates rising to higher positions.

But you have to separate things here, because these seventeen are business schools, not CS departments. On the tech path, entry-level roles are only 4.5% of software development jobs, and the people in senior roles today worked their way up from the entry-level cohort of ten years ago. On the consulting and banking path, the evidence points the other way: Bain is up 25%, McKinsey up 12%, and eight investment banks aren't cutting analysts.

So the loop looks different depending on where graduates go. The end that sends students into tech is narrowing; the end that sends them into consulting and banking isn't, for now. Which end matters more for where a business school stands in ten years, we can't tell yet.

What today's alumni can do is use donations, hiring, and industry resources to temporarily patch the narrowing end. If it works, the top schools' advantage grows further. If it doesn't, if the school relies only on older alumni without producing a new generation of winners, the loop will slowly lose speed over ten or twenty years. So top schools, like big companies, aren't safe because of this; they just have one more chance than others to save themselves. Whether they can depends on the following.

First, willingness. Everyone with a degree holds that credential, a big tech VP, an investment bank MD, a founder, a trustee, and none of them want their alma mater's degree to be worth less ten years from now.

Alumni really only have four things to give: money, jobs and internships, time and connections, and the endorsement of their reputation. And what schools are short on right now is people who can assess students, and a good location.

Money can be turned into people who assess students, but at a bad exchange rate. The Berkeley money from part one is a reference point: $252 million to start, plus three more gifts totaling $75 million, over $300 million in all, most of which became a building, with two new faculty positions tied to it in the official materials. Money can endow chairs, but a permanently endowed chair locks up a large principal that can only ever fund that one seat. The $500 million gift commitment CMU's School of Computer Science received in September lists directions including redesigning how computing is taught, GPUs and other compute, AI tools, and support for students. How much goes to each isn't specified, and how much of it will become teaching and assessment staff isn't said either.

Jobs are difficult in a very specific way. An alum who is a VP at a big tech company used to help their alma mater most concretely by pushing for a few more new-grad headcount each year. If big tech sharply cuts new-grad hiring, the slots that alum can steer shrink with it. The thing alumni most want to give, and can most easily give, is exactly the thing AI may be squeezing.

Going forward, the scarcest thing alumni can give students, beyond a job, is a real apprenticeship: time with customers, handling exceptions, sitting in on post-mortems, and owning a small piece of the outcome.

There's another mismatch: the way people give doesn't match what schools are short of. A building can carry a name; a TA's salary can't. Oral exam time, proctoring staff, in-person questioning: all of it is recurring spending, every year, every month, cash flow with no ribbon-cutting, and that operating budget is exactly what schools lack. Buildings and equipment are easier to see, name, and count, and they stand out more in press coverage.

Twelve of the seventeen schools are named after a single donor or their foundation: Wharton, Booth, Kellogg, Sloan, Haas, Stern, Anderson, Ross, Tuck, Johnson, Fuqua, Tepper. The price of a name also tells you something about time: J.B. Fuqua gave $10 million in 1980 and Duke's business school became Fuqua; Tepper gave $55 million in 2004; that same year Ross gave $100 million. In twenty-four years, the price of a name went up tenfold. The naming money didn't all go into buildings, either: Ross gave another $50 million in 2017, and the school's published breakdown put only $10 million into campus construction, $16 million into faculty support, and $8 million into a student investment fund.

The names of business schools are a rich list from the past.

What's harder than money is that a university receives resources from individuals, and to really use them, they have to become something the school can keep doing year after year, and that's the step that most often gets stuck. Inviting a CEO to give a talk is easy; embedding their company's real project into a course, staffing it with TAs, designing the assessment, and rerunning it every year is another matter. Receiving a gift isn't the end; turning it into faculty, compute, assessment, and opportunities for students is the beginning. And many gifts come with restrictions, so the school may not be able to redirect them to what most needs doing.

Garry Tan said the same thing earlier. An organization that can't turn what worked into something it can call on again and again wakes up every morning with amnesia, no matter how good the model is. Universities are the oldest example of that rule. When a lecture ends, whatever that CEO brought leaves with them; only the part written into courses, internship pipelines, and recruiting relationships stays.

The last generation's winners, coming back to help, may end up preserving exactly the last generation's model of success. How much of the experience of someone who rose on a twenty-year-old path still applies today, nobody can say. Alumni can push a school to change, or be the old system's most solid supporters. And they tend to care more about rankings, naming, and reputation, while what schools really need to change is curriculum, assessment, and faculty incentives, none of which come with a ribbon-cutting.

The same goodwill pays out differently in different places. Alumni of Bay Area schools work at Bay Area companies, so the job, internship, or coffee they offer pays off locally on the spot. Alumni of schools in weaker locations naturally have less to offer locally, because there just aren't as many companies there; the jobs they offer are in other cities, and introductions may happen elsewhere too. That helps students; it does nothing for the local industry.

So alumni won't make money and opportunity any less concentrated. The 6.3x gap between Harvard's $56.9 billion and Dartmouth's $9 billion was compounded over decades from donated principal and long-term investment returns. With enough money, you really can do a lot of what you want.

There's a lot alumni can do: endow chairs, build research centers, fund startup funds, give scholarships, bring company courses in, set up data partnerships. Two of those relate to location: bring industry here, or send students out. One of them has always worked, and it happens to be what schools in weaker locations do.

The first is hard. Argo AI is the textbook case: the technology grew in Pittsburgh, the decision-making power and capital were somewhere else, and once the two automakers pulled their money, the company was gone. But Pittsburgh's robotics companies didn't vanish with it; the cluster around NREC is still there. So what Argo proves is that a single company can't survive outside capital pulling out, not that an industry can't be brought in.

The second has always worked. Tuck's class of 2025 sent 41% into consulting and 27% into finance, and Johnson sent more than 40% into finance, while Hanover and Ithaca turn up zero times in the six venture reports. That shows these two schools have built a cross-regional employment pipeline. The pipeline moves people out; it doesn't move industry in. A third path is the one CMU's Miami campus is about to try: the school itself goes to another city. Under the current plan, pending regulatory approval, it will admit graduate students first in 2028 and undergraduates four years later. Whether that path works will depend on whether it can fill the nearly 300 faculty positions it has planned.

When judging whether a school is safe, the number of famous alumni is the easiest thing to count and the least informative. I think the key is whether alumni goodwill turns into things the school actually does. Is that alum CEO showing up once or staying involved? Did the money go into a building or into faculty and courses? Are alumni giving advice, or real projects and jobs? Has the school changed how it trains and assesses students in response to industry feedback, and does any of that show up in where graduates end up?

This argument has a condition that would overturn it. If in the next three years some school in a weaker location uses its alumni to anchor a real industry locally, enough to actually move that city's venture deal count and Fortune 500 headquarters count (a donated building doesn't count), then this may need to be reassessed. The CMU Miami campus can't test this, since it's a new campus somewhere else; it says nothing about whether industry can be anchored around CMU's Pittsburgh campus. No such example has appeared yet.

People who give to their alma mater are rarely making an investment decision. Mostly they're paying back what they feel they owe, or trying to spare today's students some of the detours they took. But what schools need right now can't be bought with donations. The largest single gift by an individual in the history of American higher education came from someone who isn't a CMU alum. Griffin went to Harvard, has funded CMU through Citadel since 2015, and CNN described the gift as part of his bet on Florida. He will also join CMU's board of trustees.

AI has lowered the barrier to knowledge, but it may raise the barrier of trust and relationships. The teaching value of a school is being repriced, and the alumni networks of the top schools may be worth more because of it. Only one question remains: whether schools can actually put these alumni resources to work.

08 What it means for you

If you work at a big company, don't just look at how many AI tools your company has deployed; look at whether the layers have moved. Layers not moving doesn't mean nothing is happening; it may still be stuck at organizational inertia, at the level of efficiency, quality, or delivery time. Which roles have stopped shuttling information, who holds final sign-off, which teams have turned something that worked once into something they can call on again and again: these are what matter more. You can also check yourself: how much of your day goes to gathering what's happening below, compressing it into one page, and passing it up? The more it is, the more at risk you may be.

If you're in finance: writing memos faster won't protect you; the model is faster than you. What matters is how close you are to the trades, the clients, and the risk-taking. The eight investment banks told Reuters they aren't cutting analysts because AI-generated models and pitchbooks can't yet be handed directly to a CFO; the whole weight of that sentence rests on the word handed. Without a chance to pull the trigger, even the most elegant analysis can't become judgment.

If you're building a company, look first at the customer's budget and cash flow. Only when someone is already paying for that piece of work do you have a chance to take over the whole task; YC's end-to-end share going from 10% to more than 25% is about exactly this. Your contract also has to spell out who's responsible when something breaks, which now matters far more than which model you pick. And the hard-tech number going from 8% to 20% deserves a serious look: if your industry experience grew somewhere with a physical world, regulation, and real data, that's exactly where people are most needed right now.

09 Final thoughts

The question we left open in part one was which kind of proof the market is buying. It doesn't really buy either. It buys what sits outside the proof: judgment, context, and someone who takes responsibility when things go wrong.

Investors, company leaders, and a top accelerator, from three different positions, are saying the same thing. Once the gap in who can do the work shrinks, the gap in who has experience shows.

As for whether management layers or entry-level roles get cut first, the data we have points to entry-level roles. The work that used to train new people may be going to AI. That's bad news for new grads and good news for people with twenty years in, and neither group happens to be the person pictured in business school and CS department brochures.

Over the next two or three years, I think four numbers will tell whether this argument holds:

First, whether the solo-founder share levels off around 20% or keeps climbing. If it keeps climbing, the idea that you don't need a full team to start is still advancing. Second, whether entry-level roles' share of software development jobs recovers from 4.5%. This is the most consequential number in the whole piece; if it doesn't recover, the senior engineers of ten years from now become a debt nobody claims. Third, whether the Bay Area's share of US venture capital falls back from 77.5%. It has to keep falling across several funding cycles to shake the Bay Area's geographic advantage; over just a quarter or two, a drop may only mean those few mega-rounds dropped out of the base. Fourth, whether a school outside the Bay Area can show that online access delivers opportunities of the same quality as being there in person.

One more thing matters a lot, and there's no room for it here, so the next piece is about it: when one middle manager can run ten agents, and the work new hires used to learn on goes to machines, will companies get smaller and flatter, or will power just concentrate in fewer hands?

What Berkeley protects with pen and paper and what YC pays for are the same thing: judgment that can't be replaced, and trust that can only be built over time.

These two pieces counted a lot of tables: course catalogs, rankings, funding, jobs, metro areas. But under every cell is a person. A teacher who decided to put exams back on paper, a department chair who wouldn't expand their department, a sixty-year-old still writing checks to their alma mater, a new grad who just found out the path their seniors took is gone. AI replaces the tools; the ones running this system, taking part in it, and getting swept up by it are, from start to finish, people.

One last line from Garry Tan: the model's ability is rented; your own brain is yours. That holds for companies, and it holds for people; and which city that brain grows up in matters more than which school.

If you're at any of these seventeen schools, or making this choice for someone, I'd love to know which of these you'd look at first.

Data appendix

Venture capital in 2025 (sum of four quarterly PitchBook-NVCA Venture Monitor reports; totals are my calculation): Bay Area $185.1 billion / 2702 deals, New York $35.6 billion / 1809 deals, Los Angeles $17.5 billion, Boston $15.5 billion, Seattle $6.5 billion, Philadelphia $4.6 billion / 575 deals, Chicago $2.6 billion / 311 deals (Crain's citing the PitchBook-NVCA basis), Pittsburgh $2.29 billion (of which institutional venture capital was $2.06 billion / 63 deals, from the fourteenth annual Innovation Works and EY report). New Haven, Charlottesville, Hanover, Ithaca, Ann Arbor, and Detroit appear zero times across the six reports.

Fortune 500 headquarters (the official metro breakdown was discontinued in 2026; the figures below come from seven regional sources with seven different methodologies, so read them only for order of magnitude): Chicago 30, New York 49 (counting Connecticut separately; 62 for the larger combined statistical area), San Jose 21, San Francisco 14, Boston 14, Pittsburgh 10, Philadelphia 9, New Haven 1, Charlottesville 0.

Computer jobs (BLS OEWS, occupation basis, reference period May 2025): New York 328950, Seattle 194970, San Francisco 156010, San Jose 149970, Chicago 131320, Boston 128180, Philadelphia 89100, Pittsburgh 34640, Durham 25070, Ann Arbor 8410, New Haven 5300, Charlottesville 4040, Ithaca 1280. Hanover isn't in any metropolitan statistical area and has no series; the nearest available figure is Manchester's 9210.

CSRankings US institution ranks (all areas, 2016 to 2026 window): CMU 1, MIT 5, Cornell and Berkeley tied 7, Michigan 10, Stanford 12, NYU 14, Penn 18, Columbia 19, UCLA 20, Chicago 24, Northwestern 26, UVA 27, Duke 30, Harvard 39, Yale 42, Dartmouth 63.

Educational services plus health care and social assistance as a share of QCEW-covered jobs (BLS QCEW, March 2026; counts jobs, not people, so multiple jobholders are counted for each job): Ithaca 52.0%, Hanover 43.1%, Charlottesville 42.4%, New Haven 38.8% (including 30176 jobs in private educational services).

NIH funding FY2025 (BRIMR institution table, Award column, in hundreds of millions of dollars, national rank in parentheses): Michigan 7.24 (4), Penn 7.23 (5), Yale 6.80 (6), Stanford 6.44 (8), Duke 6.24 (10), Columbia medical 5.94 (11), UCLA 5.16 (15), NYU School of Medicine 4.38 (19), Northwestern Chicago campus 4.06 (21), Weill Cornell 3.17 (29), Chicago 2.74 (37), UVA 2.31 (45). Harvard Medical School 1.48 (63), with its teaching hospitals counted separately: Massachusetts General 6.43 (9), Brigham and Women's 4.14 (20), Boston Children's 2.46 (42). Berkeley 1.46 (66), MIT 1.34 (70), Dartmouth 1.04 (81), CMU 0.35 (176).

Endowments FY2025 (NACUBO-Commonfund basis): Harvard $56.9 billion, Yale $44.1 billion, Stanford $40.8 billion, Penn $24.8 billion, Michigan just over $20 billion, Columbia $15.9 billion, Northwestern $15.2 billion, Duke $12.3 billion, Cornell $11.8 billion, Chicago $10.6 billion, Dartmouth $9 billion.

Sources

YC: Lightcone podcast, The State of Startups in 2026, Y Combinator official YouTube, 2026-09-17, https://www.youtube.com/watch?v=yslXlV2BP_Y . The line knowing what to build is becoming more important than simply knowing how to build it, and the section title about experienced founders coming back, come from the episode's official description and chapter list, not from any one guest speaking. All other quotes come from the episode. Titles for Diana Hu and Jared Friedman per https://www.ycombinator.com/people and the YC blog, 2026-06-11. Ali Ghodsi: a16z podcast with Martin Casado and Sarah Wang, 2026-09-18, https://www.youtube.com/watch?v=GzEtpAKYRvE . The video circulates under two titles; go by the link and date. Garry Tan: Every company should have a Brain, AI Engineer World's Fair, AI Engineer official channel, 2026-07-17, https://www.youtube.com/watch?v=eBUyTS7SzV4 . Note that his 2026-08-06 talk at YC Startup School overlaps in content but differs in wording; everything quoted here is from the July talk. Macro data: Indeed Hiring Lab 2026-07-23; SignalFire State of Talent 2026-06-22; Stanford Digital Economy Lab, Canaries in the Coal Mine, revised 2026-08-12; FRED series IHLIDXUSTPSOFTDEVE through 2026-09-11; blog.google 2026-04-22. Pittsburgh: TechCrunch 2022-10-26 and Smart Cities Dive (Argo AI's shutdown, Ford and Volkswagen pulling investment, about two thousand employees, $12.4 billion peak valuation); The Robot Report citing Aurora CEO Chris Urmson; CMU Tepper 2022-23 MBA employment report, geographic breakdown. Campuses: Poets and Quants 2025-10-23 (Wharton's new campus at 555 California) and 2025-05-02 (Northwestern closing its San Francisco campus, Kellogg SF Immersion continuing); executivemba.wharton.upenn.edu San Francisco campus page; kellogg.northwestern.edu San Francisco Winter Quarter page. Finances: NACUBO-Commonfund Study of Endowments FY2025 (FY2025 market value, $944.3 billion across 657 participating institutions); individual FY2025 endowments via the NACUBO basis; NSF NCSES Higher Education Research and Development Survey FY2024 (US higher education R&D spending of $117.5 billion, 55% federal, about 25% institutional); Stanford FY2025 sponsored research revenue and federal share from facts.stanford.edu; NIH funding FY2025 from the Blue Ridge Institute for Medical Research (BRIMR) 2025 institution table, titled All Funded Institutions including R&D Contracts, which includes R&D contracts and is built on NIH RePORT FY2025 year-end data, Award column, https://brimr.org/brimr-rankings-of-nih-funding-in-2025/ , file Institution_2025.xlsx, downloaded and checked 2026-09-27. The same university can appear as several entities in the table; this piece uses the medical entry and lists affiliated teaching hospitals separately. Location: CSRankings (all areas / US only, 2016–2026 window, DBLP September 2026 data, read live in a browser 2026-09-22); US News Best Graduate Schools 2026, Best Computer Science Schools (peer-reputation survey, graduate level); US News Best Colleges 2027 undergraduate ranking, released 2026-09-22 (MIT No. 1 overall for the first time; new Earnings by Major metric worth 5% of the total, replacing the graduate debt metric, sourced from the Department of Education's College Scorecard); PitchBook-NVCA Venture Monitor, four 2025 editions plus 2026 Q1–Q2; Innovation Works and EY fourteenth annual Pittsburgh report, 2026-03; CED 2025 NC Venture Report, 2026-02-24; Fortune 500 Explorer 2026-06-03 plus Philadelphia Inquirer 2026-06-04, Pittsburgh Business Times 2026, AdvanceCT 2026-06-04, Virginia Business 2026-08-02; BLS QCEW March 2026, CES August 2026 preliminary, OEWS May 2025; CompTIA State of the Tech Workforce 2026; Brookings, America's AI boom is concentrating in a few metro areas, 2025-07 (six-tier classification of 387 metros); PitchBook University Rankings 2026 (official site returned 403; used the Poets and Quants 2026-09-08 reprint, checked line by line). Alumni and naming: Bloomberg Philanthropies and washingtonpost.com 2018-11-18 (JHU $1.8 billion), forbes.com 2024-07-08 ($1 billion for medical school tuition); news.stanford.edu 2016-02-23 (Knight $400 million); news.umich.edu 2013 and Poets and Quants 2017-09-20 (the three Ross gifts); cmu.edu 2004-03-19 and cmu.edu/homepage 2013 (the two Tepper gifts); fuqua.bulletins.duke.edu school history (J.B. Fuqua $10 million, named 1980); trustees.duke.edu (Tim Cook is a Duke trustee and Fuqua class of 1988; no major gift from him to Duke was found). Naming for all twelve schools per each school's official history page; this piece gives documented amounts and years only for the three above. CMU and Griffin: the $3 billion gift commitment, the split between two cities, the size and academic design of CMU Miami, the uses and naming of the School of Computer Science's $500 million, and the funding relationship since 2015, from CMU's 2026-09-30 announcement https://www.cmu.edu/news/stories/archives/2026/september/carnegie-mellon-university-announces-historic-3-billion-gift-from-ken-griffin-pioneering-a-new-model , the School of Computer Science announcement https://www.cmu.edu/news/stories/archives/2026/september/500m-gift-will-fuel-next-era-of-computer-science-at-carnegie-mellon , the CMU Miami academic design page https://www.cmu.edu/news/stories/archives/2026/september/higher-education-gets-a-new-model-cmu-miami , and the feature on Griffin and CMU https://www.cmu.edu/news/stories/archives/2026/september/griffins-landmark-gift-will-shape-the-future-of-higher-education ; the reasons for choosing Miami and Laurence Ales's remarks from CMU's two-cities feature https://www.cmu.edu/news/stories/archives/2026/september/two-cities-one-cmu ; graduate students first in 2028 and undergraduates four years later from Reuters 2026-09-30, reprinted by whtc.com https://whtc.com/2026/09/30/citadels-griffin-donates-3-billion-to-carnegie-mellon-with-plans-for-miami-campus/ ; Citadel's 2022 decision to move its global headquarters to Miami from Griffin's memo to employees, printed in full by fortune.com 2022-06-23 https://fortune.com/2022/06/23/read-the-memo-ken-griffin-sent-to-citadels-employees-outlining-the-companys-plan-to-leave-chicago-for-miami/amp ; Griffin's 1989 Harvard undergraduate degree from https://alumni.harvard.edu/stories/legacy-for-financial-aid ; the largest single gift by an individual in US higher education from Bloomberg 2026-09-30 https://www.bloomberg.com/news/articles/2026-09-30/ken-griffin-s-3b-carnegie-mellon-donation-is-biggest-single-gift-ever ; bet on Florida is CNN's phrasing in its 2026-09-30 headline https://www.cnn.com/2026/09/30/business/ken-griffin-carnegie-mellon-donation-miami , not CMU's or Griffin's. Enrollment and hiring data in the charts: National Student Clearinghouse 2026-08 (undergraduate CS enrollment at four-year schools down 8.4% year over year, total undergraduate enrollment up 1.3%); classes.berkeley.edu (CS 61A enrollment 1269 in fall 2025, 1469 in fall 2026, capacity 1750); Karat 2026 (share of employers allowing AI in technical interviews: 38% in the US, 68% in China). The case against: Poets and Quants 2025-11-25, 2025-10-16, 2025-09-15; cityam.com 2026-06-01; Reuters Breakingviews 2026-06-29.