Sports as AI Insurance

May 8. June 3. June 11.

Within five weeks this spring, three of the most sophisticated actors in and around sports finance said the same thing in public. On May 8, Axios’s Dan Primack reported that sports deals are becoming “AI protection” — investors, in his words, treating sports as “the last bulwark against AI disruption.” On June 3, Mary Callahan Erdoes, who runs JPMorgan’s asset and wealth management division, called sports “the antithesis of AI” from the stage of the Forbes Iconoclast Summit, and predicted team valuations would keep surging because of it. On June 11, PitchBook reported — citing Alastair Seaman, a partner at Arctos, the private-equity firm whose business is minority stakes in sports franchises — that PE firms are pushing deeper into leagues major and minor, “betting on the lasting power of live events as something that can’t be replaced with AI.”

Three datestamps. One thesis. The reporter who covers the deal flow, the banker who manages the money behind it, and the fund that buys the equity all converged, in the space of five weeks, on the same sentence: sports is the one media asset AI cannot synthesize, and that is why the money keeps coming.

I think the thesis is real, and mostly right. I also think the market saying it out loud has not priced the corollary. If AI cannot replace the game, it can absolutely replace the cost structure around the game — and that compression does not accrue evenly. It accrues to whoever owns the rights. Not to whoever owns the trucks, the networks, or the seventy years of middle layers that grew up between the game and the fan.

The “antithesis of AI” is about to become one of its biggest beneficiaries.

Five Weeks, One Thesis

A generative model can produce a song you have never heard, a film that was never shot, a face that was never born. It can do this on demand, at marginal cost approaching zero, in unlimited quantity. Every media asset whose value rests on the supply of finished content is exposed to that math. Scripted drama is exposed. Music catalogs are exposed. Stock photography is already gone.

A live game is different in kind, not in degree. Its value does not come from the finished artifact — the tape of a game you already know the result of is nearly worthless, which is why sports has always had the steepest decay curve in television. Its value comes from an outcome nobody knows yet, unfolding in real time, watched simultaneously by everyone who cares. A model cannot generate an unknown outcome. It cannot generate the specific Tuesday-night configuration of twenty-two athletes whose careers, contracts, and grudges are real. And it cannot generate the synchrony — millions of people watching the same uncertain thing at the same moment, which is the substrate under every rights deal, every sponsorship, and every sportsbook.

I made the pre-AI version of this argument a year ago, in a piece about what isn’t changing in live sports: as synthetic media gets abundant, the value migrates to the one thing that cannot be synthesized — scarce, scheduled, unfakeable human drama. What happened this spring is that the argument stopped being an essayist’s observation and became an asset-allocation memo. When the same sentence appears in a deal-flow newsletter, a bank executive’s conference-stage commentary, and a private-equity firm’s public positioning inside five weeks, that is not three people having the same idea. That is a thesis hardening into a price.

And a thesis that has hardened into a price deserves more scrutiny, not less.

The Money Is Acting on It

The convergence was not just rhetorical. The capital moved in the same window.

On April 28, Teamworks — which brands itself “The Operating System for Sports” and is now pitched, by itself and its investors, as an AI platform for elite sports — raised at a valuation above $1.5 billion in a round led by Hg. The same day, InStudio Ventures launched a $50 million sports-and-performance fund anchored by the firm’s own minority stakes in the Bills, the Chargers, and the Aston Martin Formula 1 team — with professional athletes making up about two-thirds of its LPs — to back sports-technology startups, its showcase portfolio company an AI muscle-imaging outfit. And private equity keeps pressing further into the sport itself — down past the major leagues and into the minors, where Arctos led a December deal to buy two minor-league baseball clubs, the Jacksonville Jumbo Shrimp and the Akron RubberDucks, in a week that saw seven MiLB teams change hands in PE-connected transactions.

The context makes this more interesting, not less. The same PitchBook piece that carried the Arctos line was, in fact, headlined “AI can do a lot, including slow down PE exits” — firms are holding portfolio companies longer to retool them for AI, and the stretched timelines are dragging on net IRR, the industry’s favorite metric. Read those two facts together. The same asset class that is discounting nearly everything for AI risk is paying premiums for sports. Sports is not participating in the AI repricing. Sports is the AI repricing, run in reverse — the short leg of the trade everyone is putting on against media.

That is what insurance looks like on a balance sheet. You do not buy insurance because you expect the house to burn down. You buy it because everything else in the portfolio smells like smoke.

Here is where I part company with the people writing the memos — not on the thesis, but on where they stopped.

The Game Was Never the Whole Product

The thesis says AI cannot synthesize the game. True. But the game was never the whole product. The product — the thing the fan actually receives, the thing the industry actually employs people to make — is the game plus everything wrapped around it: cameras, trucks, directors, replay operators, graphics, commentary, highlights, dubbing, clip desks, promos, and the entire apparatus that routes a fan’s attention toward the next thing to watch. The game is unfakeable. Almost nothing wrapped around it is.

Look at what actually shipped this year, because the pattern is unambiguous.

This summer’s World Cup was the largest sports production ever mounted, and its genuine engineering story had almost nothing to do with AI — it was software. Sixteen venues backhauled to an international broadcast center in Dallas, roughly sixty high-powered servers running as a private cloud, every one of the 104 matches — around 9,000 hours of content — processed in software on commodity hardware rather than dedicated broadcast gear. That is not artificial intelligence. That is the production truck being dissolved into a data center, which is the precondition for everything AI does next: once the production stack is software, every layer of it becomes a candidate for automation.

And where AI proper did ship in production this year, it shipped as one thing: cost compression. AWS launched AI reframing that converts a horizontal live feed into vertical video automatically — Fox Sports Digital and NBCU among the early adopters — which is a task that used to be a person at an editing station. Spiideo launched AI Highlights in May. FloSports runs what it calls an “AI Producer” — a virtual camera that follows the action with no human operator on the sticks. Pixellot was on track to capture 1.5 million games in 2025, per SBJ — a company-supplied count, but the order of magnitude is the point, and the games were never going to have a crew. Now they are covered by no one.

The most instructive deployment is ESPN’s SC For You, the personalized SportsCenter inside its streaming app. Per Sports Business Journal’s reporting, the system ingests something like 16,000 game feeds a year and twenty-five tagged studio shows, with 162 writers and editors feeding the pipeline, and delivers each fan a personalized highlight show narrated by AI voice clones of four of ESPN’s own on-air personalities, built with the talent’s participation. Notice the shape of that: the game footage is untouched and irreplaceable at the center, and every layer around it — selection, assembly, sequencing, narration — has been handed to the machine. ESPN did not use AI to replace the game. It used AI to replace the apparatus around the game.

Now hold that shape up against the industry’s org chart. Where do most of the jobs in sports media sit? Not on the field. They sit in the trucks, the control rooms, the edit bays, the clip desks, the dubbing studios, the highlight shows, the promo departments. Where does most of the margin in the middle of the industry sit? Same place. The layers between the game and the fan are the industry’s employment base and its cost base at once — and every one of them is on the synthesizable side of the line the JPMorgan quote draws.

The thesis is right that the core is safe. It is silent about the fact that the core was maybe a tenth of the payroll.

The Corollary Nobody Is Pricing

So follow the compression to where it settles. If the rights are the un-synthesizable core, then everything AI cheapens around them — the truck, the edit bay, the reframing desk, the dubbing session — widens the margin of whoever owns the core. Cost compression in a value chain always accrues to the layer with the scarcity. In sports, the scarcity is the rights.

If AI cannot replace the game, it can absolutely replace the cost structure around the game — and that compression accrues to whoever owns the rights, not to whoever owns the trucks.

You can already watch this play out in the most distressed corner of the American sports economy: local rights. Main Street Sports — the largest of the old regional-sports-network operators — wound down in April, leaving thirteen NBA teams and seven NHL teams as local-television free agents for the 2026-27 season. I spent a quarter mapping this market in a research report earlier this year, and the pattern since has only sharpened. The bids now on the table, per Sports Business Journal’s reporting, are a fraction of the old RSN checks: Fubo offering in the range of $8 to $15 million per team; Gray, Scripps, and Nexstar below $10 million; DAZN — which announced in April it would absorb the ViewLift streaming platform — courting orphaned teams with rights-fee haircuts of as much as 40 percent. Scripps took the Pistons in May at a reported $8 to 10 million. Gray extended the Suns and Mercury for four years — a long-term re-up of what had been the first modern NBA deal to leave an RSN for over-the-air television.

Ask the obvious question about those numbers: what makes a 40-percent-haircut rights deal survivable for the buyer? Only one thing — a production stack dramatically cheaper than the one the RSNs carried. And the AI tooling that shipped this year is precisely that stack. Automated cameras for the games a crew never covered. AI highlights instead of a clip desk. Automated vertical reframing instead of a social-video team. The local-rights market is the first place in sports where AI production economics are not a slide in a vendor deck — they are the difference between a bid that closes and a bid that doesn’t. Nobody involved is calling it AI insurance. It is the same trade, run from the other side.

The leagues have noticed where this leads, even if the memos haven’t. MLB absorbed nine more clubs’ local production this year and has said publicly it wants all thirty clubs’ local rights in hand by the end of the 2028 season, packaged as one national in-market product. The Angels bought their regional network outright and now sell a direct subscription at $74.99 a season. A league becoming its own production company is the corollary in institutional form: pull the rights to the center, compress the stack underneath them, keep the spread. And the quietest league is the most instructive — the NHL, which has the smallest local-media revenues in American major-league sports and therefore the most to gain from compression, held its first league-wide analytics summit this spring, roughly ninety staffers from thirty-one franchises, with something like two hundred data staff now working across the league’s thirty-two clubs. Keith Pelley, who runs MLSE, put it flatly in theScore’s reporting: “AI is massive. It is changing our business.” The league whose economics were most damaged by the RSN collapse is the one industrializing the cheap stack fastest. That is not a coincidence. That is the corollary being executed by the people who need it most.

Meanwhile the people the compression works against — the production houses, the truck operators, the regional broadcasters whose margin was the production, the networks whose value-add was the apparatus — are on the wrong side of the same line, holding cost structures built for a world where wrapping the game was expensive and therefore defensible. It was the expense that made it a business. AI does not have to touch the game to take that business apart.

The Measurement Footnote

One more thing happened this year that will move more rights money than any model, and it barely made the AI conversation at all.

In February, Nielsen’s Big Data + Panel methodology made its debut as the currency on the biggest stage in American television — Super Bowl LX, measured at 124.9 million average viewers and later revised upward to 125.6 million. Under the hood, that is a change in how the number that underwrites every rights negotiation gets manufactured: panel measurement fused with set-top-box and smart-TV data at population scale. Every carriage fight, every upfront, every league’s internal model of what its rights are worth runs on that number. Re-base the currency and you re-base the market — quietly, in a methodology appendix, with no press cycle about disruption. With eMarketer projecting US sports TV ad spend past $20 billion in 2027, small percentage shifts in the measured number move real money.

I include this as a discipline check on everything above. For all the AI language now wrapped around sports finance, there is no evidence that any actual rights deal this cycle was priced by a model — deals are still priced the way they have always been priced, by reach and desperation. The genuinely consequential quantitative event of the year for rights values was not artificial intelligence. It was an accounting change in how reach gets counted. Anyone underwriting the “sports as AI insurance” trade while ignoring the measurement regime change is watching the wrong instrument panel.

The Moat and the Fire

A year ago, I argued that live sports was the thing that wasn’t changing — that in an age where everything synthetic becomes abundant, the scarce, scheduled, communal, unfakeable event becomes the most valuable real estate in media. Liveness was the moat.

I underwrote that argument, at the time, as a case for resilience. Sports would endure. What this spring’s convergence — the reporter, the banker, the fund, five weeks, one sentence — made clear is that endurance was the timid version of the claim. The moat is not just holding. Everything outside the moat is on fire, and the fire is raising the value of what’s inside it.

Because that is what AI actually did to sports this year. It did not touch the game. It could not. What it touched was the apparatus — the trucks dissolved into server racks in Dallas, the clip desks folded into a personalization engine, the editing bays replaced by a reframing API, the local production stack cut down until a 40-percent-haircut rights bid could clear. Every one of those fires makes the un-burnable thing at the center worth more, and the ownership of that center is consolidating — into leagues that produce their own games, into funds that hold the equity, into whichever balance sheets figured out first that the game and the apparatus were never the same asset.

The people writing the memos got the thesis right: the game cannot be synthesized. The next memo is the one that matters — the one that says the game was never the whole product, prices the apparatus at what a model can do it for, and pays for the rights accordingly.

Sports is not insurance against AI. Sports is the property that AI is clearing the land around. If you own the deed, the fire is doing you a favor.

Everyone else should check what, exactly, their margin was made of.