The Shot Clock

The Platform Trap: Why Giving Your Alpha to Frontier AI Labs Is the New Zynga Play In 2011, Zynga was one of the most successful companies in the world. FarmVille was a cultural phenomenon, the IPO was oversubscribed, and the business model looked unstoppable. There was just one line buried in the S-1 that mattered more than all the others: substantially all of the company’s revenue came through the Facebook platform.

A year later, Facebook decided Zynga’s viral mechanics were bad for its users and turned them off. Zynga never recovered. It eventually sold for less than its IPO price, not because it built bad games, but because it built its business inside someone else’s business.

Last week, Alex Karp went on CNBC and said the quiet part out loud. The frontier AI business model, he argued, is IP extraction disguised as a subscription. Enterprises are paying for tokens that create no measurable value while transferring the alpha of their operations internal memos, strategy documents, customer data, workflows into the pipelines of a handful of labs.

He asked the question every buyer should be asking: if the models were as valuable as claimed, why do the labs charge by the token instead of by the outcome? You don’t meter the transformative. You meter the commodity. And you especially don’t hand your proprietary edge to a vendor who reserves the right to enter your business.

That last part isn’t hypothetical anymore. Figma spent months integrating Claude into its products, collaborating with Anthropic on features that put Figma’s design workflows in front of Anthropic’s models. In April, Anthropic’s chief product officer resigned from Figma’s board. Three days later, Anthropic launched Claude Design. Figma’s stock dropped seven percent in a day.

Whatever you call that vertical integration, moving up the stack, competing with your customers it rhymes perfectly with 2012. The platform watched what its best partner did, learned from it, and then became it.

If you want to see where this pattern ends, look at healthcare. Epic didn’t just sell health systems an electronic health record it sold them a platform, and then it watched. Every time a health system built a clever workflow, a smarter order set, a better revenue-cycle configuration, Epic absorbed the best of it into its Foundation System and shipped it to everyone else.

Thousands of health systems spent two decades doing Epic’s product development for free, each one paying handsomely for the privilege. The final move is the one nobody wants to say out loud: a company sitting on that much operational knowledge and that much capital doesn’t need to stay a vendor forever. Pair Epic’s institutional intelligence with private-equity-scale acquisition capability and you get the endgame buying health systems outright, running them on the accumulated best practices of every customer who ever taught the platform how their business worked, and competing directly against the institutions that built its advantage.

The vendor becomes the operator. The platform enters the health system era. Whether Epic ever pulls that trigger matters less than the fact that it could, because the frontier labs are running the same accumulation play right now, across every industry at once, at a speed Epic never had.

So the real question isn’t whether frontier labs will behave like Facebook. Some already are. The question is what the market structure looks like by 2030, because the structure determines who has leverage. There are three plausible futures, and they are not equally likely.

1. The Hegemon Scenario One lab wins. Capability compounds through a data flywheel: every enterprise that plugs in makes the model smarter, which attracts the next enterprise, until one company effectively holds an index fund of every industry’s operational knowledge. This is the scenario Alex Karp is warning about. In this world, every business faces the Zynga choice: accept existential platform risk or accept inferior tools.

2. Commoditization The frontier commoditizes. A dozen to a hundred labs stay within shouting distance of each other, open-weight models keep raising the floor, and no single vendor can extract IP with impunity because customers can walk. Knowledge still concentrates in labs, but competition disciplines them. Value migrates toward orchestration, governance, and the application layer. Tokens become what electricity became: essential, interchangeable, and cheap.

3. Boutique Models Models go boutique. Distillation and cheap inference make it viable for every serious enterprise to run its own fine-tuned, domain-specific models on its own infrastructure. The frontier labs become foundries, like TSMC: enormously important upstream, invisible downstream. Data never leaves the building. Your alpha stays your moat.

My bet is that 2030 looks like a hybrid of the second and third: an oligopoly at the frontier, boutiques at the edge, with the hegemon scenario surviving only as the tail risk everyone regulates against.

But the deeper lesson from Facebook isn’t about which future arrives. It’s that anyone whose alpha flows through someone else’s platform is playing against a shot clock. Zynga had roughly a year between its IPO and the moment Facebook changed the rules. Figma had two months between collaboration and competition. The clock is getting shorter.

The companies that survive won’t be the ones that picked the right lab. They’ll be the ones that never handed over the ball.