The price tag on Mark Zuckerberg’s bid for ‘superintelligence’ will blow your mind. Will the product?

The atmosphere in Menlo Park in the summer of 2025 became heavy with a particular kind of ambition. The new Meta Superintelligence Labs was being frenetically assembled. Its stated goal, articulated by Mark Zuckerberg, is to build an intelligence surpassing the human, a “superintelligence.” This artifact is framed not as a remote, centralized oracle, but as a “personal superintelligence,” an egalitarian gift for everyone. The name of the first planned AI supercomputer cluster, a multi-gigawatt facility slated for 2026, is “Prometheus.”

The myth of Prometheus is one of enlightenment and hubris, of stealing fire from the gods and suffering the eternal consequences. To name your machine this is to write your own legend before the fact, to cast your venture in the most heroic, and perhaps tragic, terms available. It is a very Californian story: the pursuit of a world-changing gift, shadowed by the risk of overstepping natural limits.

These projects are meant as a contemporary Apollo program, aimed not at the moon, but at the mind.

To pursue this modern myth, Meta began to “upend itself.” The reports suggest crisis. We learn of four distinct AI division overhauls in six months. We learn of an internal memo that spoke of an “AI arms race” that Meta was, until this consolidation, losing. The reorganization was perceived as an act of existential urgency, one with a specific texture, a specific cost. The 2025 capital expenditures were raised as high as $72 billion. Zuckerberg announced plans to spend hundreds of billions of dollars more. In October, a $27 billion deal was struck with Blue Owl to fund a single data center. These are the numbers of the new arms race.

The talent war of 2025 feels less like recruitment and more like a kind of high-stakes, frantic prospecting. By mid-August, Meta had poached more than 50 top researchers, pulling them from OpenAI, Google, xAI, and Anthropic. The compensation packages are beyond generous. We hear of nine-figure sums. We hear of a $1.5 billion offer made to a single AI lab co-founder, an offer that was declined. This is not the quiet, collegial work of a corporate lab. It is a frenzy.

This burst of activity was meant to correct a failure. The pivot came not from the excitement of new discovery, but from a place of dissatisfaction. The Llama 4 family of models, released in mid-2025, had landed with a thud. The Behemoth model, a 2-trillion-parameter research project, was scrapped. The reception was lukewarm. In response, Zuckerberg handpicked a new team. The old AGI Foundations group was dissolved, its staff redistributed. A new, small, elite working group was formed, mysteriously named TBD Lab, led directly by the new chief AI officer, Alexandr Wang.

This TBD Lab is the core, the protected center. When 600 roles were cut from MSL in October to “reduce bureaucracy,” TBD Lab was spared entirely. The rest of the machinery was re-engineered around it: the long-standing FAIR research arm, once as independent as a university, is now an “innovation engine” to feed TBD. A product team under Nat Friedman is tasked with bridging the lab to the market. And an infrastructure team must build the necessary colossal computational backbone.

RELATED: A new study hints what happens when superintelligence gets brain rot — just like us

Photo illustration by Li Hongbo/VCG via Getty Images

The physical scale of this infrastructure might seem to justify the mythic language. We are no longer talking about servers in a rack. We are talking about multi-gigawatt data centers with a physical footprint that would cover a “significant part of Manhattan.” These projects are meant as a contemporary Apollo program, aimed not at the moon, but at the mind.

The new story Meta tells is one of focus. A “leaner, more efficient unit.” A “startup within Meta.” The company even instituted a hiring freeze in October, not to save money, but to let the new structure “jell.” As if the chaos of $72 billion, of nine-figure salaries and $1.5 billion declined offers, of warring cultures and dissolved teams, of data centers meant to cover small cities, would simply set with a little time.

Inside this new, streamlined venture, a cultural story unfolds. The new guard, the expensive hires coaxed away from rivals, collides with the old guard, the Meta veterans who believed in the company’s previous ethos of open-source science. That mindset, which set Meta apart, is now very much in doubt. Zuckerberg has signaled that the most powerful models, the ones that might actually approach “superintelligence,” will not be open-source. They will be kept closed, due to safety concerns, or perhaps due to strategic ones. The shift is palpable. The open-source ideal of sharing gives way to the new, closed, competitive, secretive model of the arms race.

The skepticism from the outside world has its own narrative. Analysts warn, as Business Insider reported, of an AI bubble, of “diluted shareholder value without any clear innovation gains.” We are told, as one enterprise AI expert put it, that investors “aren’t wowed by flashy demos anymore; they want to see revenue.” The grand, Promethean vision of superintelligence runs headlong into the quarterly demand for durable, scaled products.

We are watching a company, and perhaps a culture, wager its identity on a future it can only describe in mythic terms. The question is not whether a machine can be made to think. The question is what we reveal about ourselves, our ambitions, and our anxieties in the attempt. Meta has entered its moment of truth, fueled by sums of money that are nearly as abstract as the goal itself, and driven not by the quest for fire but by a most human motivation nonetheless: a founder’s dread of being left behind.