Photo-Illustration: Intelligencer; Photo: Chris Unger/Getty Images

After ChatGPT drew millions of users to the first mature, widely available LLM-based chatbot, the race was on to build better, bigger, and more versatile AI models for a range of stated and implied purposes: to automate labor; to entertain people; to replace old software; to come up with entirely new types of software; to perhaps accumulate power in less obvious and more ambitious ways; to build some sort of god and hope it doesn’t immediately murder everyone; and, perhaps, to assure markets that your company is on top of things, wherever they’re going.

Building and deploying state-of-the-art generative-AI models is extremely capital intensive, which has resulted in the unsettling spectacle of some of the richest companies on earth broadcasting their intention to buy dominance in a future they’re all saying is inevitable: $75 billion in data-center investment from Google in 2025; $100 billion from Amazon; $80 billion from Microsoft. Each firm has a different connection to the moment and reason to believe it might prevail. Microsoft was an early OpenAI partner. Google produced some of the core research that set things in motion. Amazon is a massive provider of cloud-computing resources.

In contrast, Meta’s AI investment rationale —  the company is planning to spend up to $72 billion on AI in 2025 — has been a little more nakedly about not missing out. As a result, the company’s big pivot to AI has been a bit of a mess: half-baked features crammed into every interface possible; models that can’t catch up to the frontier; fudged benchmarks; confused users; and just downstream, slopified Facebook and Instagram. And while there are good arguments in favor of the company’s emphasis on building open-source AI, the company’s leadership clearly isn’t happy with where it stands in 2025. Which is how you end up with a situation like this, as reported in Wired:

As Mark Zuckerberg staffs up Meta’s new superintelligence lab, he’s offering top research talent pay packages of up to $300 million over four years, with more than $100 million in total compensation for the first year, WIRED has learned. Meta has made at least 10 of these staggeringly high offers to OpenAI staffers.

Zuckerberg’s approach has already worked in the narrow sense that a bunch of OpenAI researchers have taken (reportedly smaller) offers, leading the company’s chief scientist to tell employees that it felt like “someone has broken into our home and stolen something.” OpenAI CEO Sam Altman tried to play it cooler, telling staff, “Missionaries will beat mercenaries,” and downplaying the importance of the employees Meta managed to poach. OpenAI’s chief scientist and Altman both suggested that they’d be looking at compensation, however, which makes sense given that their core employees are being lured away with some of the highest job offers in the history of the capitalist system.

This is a different approach to buying the future, which is, for Meta, both novel and familiar. It’s familiar in that Meta, a massive firm, is mostly made up of properties it acquired for amounts that sounded high at the time — $1 billion for Instagram, $19 billion for WhatsApp — but which were, in hindsight, pretty good deals. It’s novel in that the investment here is in a few people rather than in a rapidly growing user base or a foundational technology. The basic plan for scaling current AI is known and shared among the big firms: large training runs with lots of data but also a greater focus on “reasoning” models and “reinforcement learning,” which have helped models improve performance as older methods produce diminishing returns but which still leave these firms basically trying to outspend one another for GPUs and data-center square-footage.

Perhaps Zuckerberg is betting that these researchers will come up with new paths for scaling, which could, for a time, belong to Meta alone. (Note the branded downward revision of “superintelligence” from a fearsome entity that must be carefully aligned to a Meta assistant available to “everyone.”) Or maybe Zuck’s bet — which is relatively small in the context of his company’s overall investment in AI — is just that he can kneecap his competitors, slowing them down enough that Meta can finally just catch up.