The New Transfer Portal

Take a look at the two job profiles below. Try to guess what roles these are for (hint, you can't find either of these on LinkedIn):

On the left is the estimated value of a contract, including stock/equity, for a senior AI research scientist. On the right is the value of University of Texas quarterback Arch Manning's name, image, and likeness (NIL) deals, the financial endorsement and sponsorship agreements college athletes have been allowed to sign since 2021.

I bring this up because this is something that's been coming across my timeline a lot lately. One day, a college football player moves to a different team after receiving a massive payday. The next, a major AI lab is either acquiring a cutting edge AI company or recruiting a member of another company for an even more massive payday.

I would venture a guess that Manning doesn't think much about AI, and AI researchers probably don't spend much time on how UT's offense matches up against Ohio State in Week 2. But over the past year, these two industries have been quietly tracking each other. The talent is moving fast, the paydays keep climbing, and universities and major AI labs are willing to spend whatever it takes to stay competitive.

Which raises the question: are these institutions and labs paying to build something, or paying not to lose something?

Mobility, Money, and the New Roster Math

The transfer portal, which allows college athletes to change schools freely without sitting out a season, has rewired how university coaching staffs build rosters. According to NCAA data, in 2018, the first official year of the transfer portal, 345 Football Bowl Series (FBS) roster spots were held by players who had transferred from another four-year program. By 2024, that number was 2,450, underscoring how roster construction shifted in just a few years.

There are instances where the transfer portal is working as intended. Take Trinidad Chambliss, a Division II national champion at Ferris State who transferred to Ole Miss, took over as quarterback after the starter was injured early in the season, and led the Rebels to a 13-1 record and a run to the College Football Playoff semifinals last year. Talent that found the right fit, the portal working as advertised.

What also happened in recent years was the introduction of NIL. While the portal increased mobility, NIL has increased the flow of money. Given the payment policies that are now in place, large institutions stopped thinking about roster construction one recruiting visit at a time, and instead started building their college football teams one booster's wallet at a time.

Just take a look at the roster spends of last year's College Football Playoff teams:

That's over $700 million dollars spent on just 12 college football teams, and most of those expenses are going to paying the players. What sticks out is that most of these programs haven't won a national championship since NIL reshaped the sport's finances (or even before NIL was a thing).

To be clear, winning a championship doesn't mean you've built something. Building means you can sustain it. For a lot of these programs, building that kind of sustained success is the goal, but it's a lot harder than buying your way to one good season.

Which means that, in many cases, the spending may be less about consistently winning championships and more about paying not to lose, both on and off the field. The more achievable goal may be staying competitive in a landscape where the media deals, playoff revenue, and sponsorship money make the seat at the table worth the price of admission.

Bigger Wallets, Same Talent Moves

The AI labs building large language models have even deeper pockets than these universities, and their recruitment playbook looks surprisingly familiar.

If you just take Anthropic and OpenAI, they're commanding almost $1.8 trillion (with a T) in valuation. Google and Meta are dropping billion dollar bags left and right. Which means the researchers building these systems are working at the frontier of one of the most capital-intensive technology races in history. Just like college football programs raid the portal, the AI labs are also not shy about what they're willing to spend to stay ahead.

Meta assembled its new Superintelligence Labs almost entirely from poached researchers from OpenAI, Google DeepMind, Apple, and Anthropic, and reportedly offered packages reaching $100 million to get them there. OpenAI's chief research officer Mark Chen told his staff at the time that it felt like "someone has broken into our home and stolen something." Funnily enough, according to a Bloomberg article earlier this week, 400 former Apple employees have joined OpenAI over the last few years.

Anthropic has been pulling steadily from OpenAI and Google DeepMind researchers in return. Back in 2024, Microsoft acquired Inflection AI to rebuild its AI division in one transaction. The list goes on, and is far from complete.

Just like college football players, some of these researchers likely found a better landing spot due to increasingly available opportunities. At the same time, some of these opportunities were nearly impossible to turn down with the amount of money attached to them.

There's no national championship in the world of AI labs, but there are few companies that have the financial depth to recruit aggressively, absorb misses, and keep building regardless of what happens.

For a lot of the other labs trying to compete at the frontier, the spending looks similar on the surface. The difference is that many of them are paying to stay in a race where the talent costs, compute budgets, and potential breakthroughs make the seat worth protecting, with the hope that staying in long enough eventually turns into building something that lasts.

What The Spending Does, and Doesn't, Tell Us

In both cases, the checks clear the same way regardless of why they were written. Miami offering $10 million to a quarterback looks the same on paper as Meta offering nine figures to poach a researcher from Anthropic. Both are bets on talent. But one of those could be an organization building something, and the other could be an organization making sure a competitor doesn't get to have it.

At first glance, it's hard to tell the difference because the intent isn't always intuitive to those of us outside the programs. However, the incentive structure is pretty easy to spot — if you want to win big, you gotta pay big. But what separates intent from incentive is what happens after the money moves. The programs and labs with a system in place are compounding, each cycle feeds the next one.

Take Indiana University's head coach Curt Cignetti. He built James Madison University into a powerhouse, at the FCS and FBS level, before he ever had access to unimaginable institutional resources. When he got to Indiana, the portal and NIL money didn't replace the system, it just accelerated it.

He already knew what pieces he needed because he'd built the thing once at a smaller scale. Indiana went 11-2 in his first season as coach, then won a national championship without a single five-star recruit on the roster last year. That's what building can look like for new competitors.

What's important to remember is not everybody can afford to build and lose at the same time. Ohio State or Alabama can absorb a down year and reload. Meta can shift its strategy if an acquisition doesn't pan out and still have a deep bench of AI researchers. The spending is survivable because the institution is bigger than any single bet.

But the smaller you get or the newer you are to the big leagues, every bet has to count. That's where the system becomes the thing that matters most.

If the system pays off, you start climbing. Indiana makes the playoffs, the recruiting attention follows, the NIL money gets easier to raise, and suddenly you're playing a version of the game the big spenders were already playing. Anthropic ships products that turn heads, the next funding round gets bigger, researchers who wouldn't have taken your call six months ago are listening now, and before you know it they've jumped OpenAI in valuation.

But the leash is often shorter than it looks. One bad cycle for Ohio State or Meta is a rebuilding year, while one or two bad cycles for Indiana or Anthropic or OpenAI, or one pending lawsuit, could undo the early years of compounding. The room to reset shrinks the less you have to fall back on, and the margin only gets thinner the further down you go.

In both cases, the spending tells you who wants in. But it doesn't necessarily tell you who's building something that will last.

Systems and Cycles

Strip away the codebases and tailgates, and these industries' trajectories are following similar arcs: spend fast, recruit aggressively, and somewhere in between try to determine whether you were building something or just making sure someone else couldn't.

Given how fast things are moving, I'm not sure some of these institutions (the ones highlighted here and the many others sprouting up) have had a second to answer that question clearly for themselves. Which might be the most telling part.

The next time a headline drops about a $10 million NIL deal or a nine-figure AI researcher package, the pressing question probably isn't about the size of the check. The question is whether the institution writing that check can tell you what it's building or what it's trying not to lose.

One of those is a system and the other is a cycle. And both of them are expensive enough that you'd better know which one you're investing in.

Tags: AI / Agents, Sports