The serial underestimation of Zuckerberg as too vulnerable, derivative, or incompetent; of the durability of his core constituencies and sources of advantage; of the undislodgable nature of his personal power; of his grinding unsentimentality about the past; of his desperation to survive; and of his attritional willingness to absorb ridicule and retaliation, is not only an equally apt description of his Roman spirit animal. It is also, perhaps, a better way to understand his attempts to win AI. ...
Thus in his first board meeting after the release of Llama 4, he announced a plan of regime change. The 11 weeks that followed demonstrated the sheer power of his position as the undeposable head of a galactically resource-rich empire. He subsumed FAIR into a new organization, Meta Superintelligence Labs (MSL), and reset the market on compensation for top AI recruits to hundreds of millions or billions of dollars per head. He paid $14.3 billion for a stake in Scale AI and installed its founder, Alexandr Wang, as Meta’s head of AI. He recruited ChatGPT co-creator Shengjia Zhao, former GitHub chief Nat Friedman, Safe Superintelligence co-founder Daniel Gross, and picked off dozens of other defectors from OpenAI, Anthropic, and Google DeepMind. He cut hundreds of employees from the older organizations, and LeCun soon left. Within one quarter, MSL was up and running.
Less than three months after the release of Llama 4, in other words, Zuckerberg had dumped the previous decade-plus of his AI people and strategy, recruited new leadership, new chief scientists, and new core researchers, and announced a new capital expenditure of hundreds of billions of dollars. By April 2026, MSL released its first closed-weight model, Muse Spark, and deployed it through Facebook, Instagram, WhatsApp, and the Meta AI app, which powers the glasses. It was greeted as the cautious return of Zuckerberg to AI, but not yet to the frontier.
From this vantagepoint, it doesn’t matter that since June [2026], Muse Spark has overtaken Google’s Gemini, vaulting Meta back to the frontier pack. It matters little that Muse, now released, was received somewhat ecstatically as perhaps the best consumer AI agent on the market, or that expectations are increasingly giddy for Watermelon, the forthcoming model. It doesn’t matter that Zuckerberg is taking on billions in new debt and spending up to $145 billion this year, including on four data centers of a gigawatt or more, one of which is planned to scale to five. It matters even less that he positioned Meta in a recent manifesto, “The Future Is For Everyone,” as more open, optimistic, and empowering of ordinary people than its relentlessly apocalyptic competitors. He’s either placed the wrong bet, and is reliving the Metaverse; or he is too late, and he’s already lost. ...
Yet perhaps it is that simple, intolerable fact—that new technologies are always being invented; what remains broken is us—which requires a scapegoat. And who could ask for a better one than [Zuckerberg]? He is, for millions of people, a hideous mirror of both modern technological capitalism, and the resistance of history and human nature to perfectibility by it. He is seen at once as shrewd and naive, calculating and lucky, too powerful and too craven. Unlike the Boomers and Gen X, he is the only trillion-dollar founder of his generation, and Millennials are cringe. He is Jewish and uncool, yet impervious to humiliation. He is, in one man, the personification of responsibility for alt-right populism; leftwing authoritarianism; the death of journalism; polarization; political propaganda; political censorship; the dangers of free speech; human rights violations; teenage depression; body dysmorphia; and male loneliness. He is an emperor for life. He is, in a word, an easy mark.
Excerpt:
The serial underestimation of Zuckerberg as too vulnerable, derivative, or incompetent; of the durability of his core constituencies and sources of advantage; of the undislodgable nature of his personal power; of his grinding unsentimentality about the past; of his desperation to survive; and of his attritional willingness to absorb ridicule and retaliation, is not only an equally apt description of his Roman spirit animal. It is also, perhaps, a better way to understand his attempts to win AI. ...
Thus in his first board meeting after the release of Llama 4, he announced a plan of regime change. The 11 weeks that followed demonstrated the sheer power of his position as the undeposable head of a galactically resource-rich empire. He subsumed FAIR into a new organization, Meta Superintelligence Labs (MSL), and reset the market on compensation for top AI recruits to hundreds of millions or billions of dollars per head. He paid $14.3 billion for a stake in Scale AI and installed its founder, Alexandr Wang, as Meta’s head of AI. He recruited ChatGPT co-creator Shengjia Zhao, former GitHub chief Nat Friedman, Safe Superintelligence co-founder Daniel Gross, and picked off dozens of other defectors from OpenAI, Anthropic, and Google DeepMind. He cut hundreds of employees from the older organizations, and LeCun soon left. Within one quarter, MSL was up and running.
Less than three months after the release of Llama 4, in other words, Zuckerberg had dumped the previous decade-plus of his AI people and strategy, recruited new leadership, new chief scientists, and new core researchers, and announced a new capital expenditure of hundreds of billions of dollars. By April 2026, MSL released its first closed-weight model, Muse Spark, and deployed it through Facebook, Instagram, WhatsApp, and the Meta AI app, which powers the glasses. It was greeted as the cautious return of Zuckerberg to AI, but not yet to the frontier.
From this vantagepoint, it doesn’t matter that since June [2026], Muse Spark has overtaken Google’s Gemini, vaulting Meta back to the frontier pack. It matters little that Muse, now released, was received somewhat ecstatically as perhaps the best consumer AI agent on the market, or that expectations are increasingly giddy for Watermelon, the forthcoming model. It doesn’t matter that Zuckerberg is taking on billions in new debt and spending up to $145 billion this year, including on four data centers of a gigawatt or more, one of which is planned to scale to five. It matters even less that he positioned Meta in a recent manifesto, “The Future Is For Everyone,” as more open, optimistic, and empowering of ordinary people than its relentlessly apocalyptic competitors. He’s either placed the wrong bet, and is reliving the Metaverse; or he is too late, and he’s already lost. ...
Yet perhaps it is that simple, intolerable fact—that new technologies are always being invented; what remains broken is us—which requires a scapegoat. And who could ask for a better one than [Zuckerberg]? He is, for millions of people, a hideous mirror of both modern technological capitalism, and the resistance of history and human nature to perfectibility by it. He is seen at once as shrewd and naive, calculating and lucky, too powerful and too craven. Unlike the Boomers and Gen X, he is the only trillion-dollar founder of his generation, and Millennials are cringe. He is Jewish and uncool, yet impervious to humiliation. He is, in one man, the personification of responsibility for alt-right populism; leftwing authoritarianism; the death of journalism; polarization; political propaganda; political censorship; the dangers of free speech; human rights violations; teenage depression; body dysmorphia; and male loneliness. He is an emperor for life. He is, in a word, an easy mark.