We have the capital from our business to do this
In an era when the architecture of intelligence itself is being contested, Meta has committed hundreds of billions of dollars to building data centres of unprecedented scale — physical monuments to the belief that whoever controls the most powerful AI infrastructure will shape the next chapter of human civilization. Mark Zuckerberg's announcement positions Meta not merely as a technology company expanding its capabilities, but as a civilization-scale actor placing a long bet on superintelligence. The wager is underwritten by a formidable advertising empire, yet its returns remain as speculative as the machines it hopes to build.
- Meta is pouring hundreds of billions into AI data centres — including a five-gigawatt facility called Hyperion — in a race to outpace OpenAI and Google before the competitive window closes.
- The company reorganized its entire AI division under a new Superintelligence Labs unit, installing high-profile executives from Scale AI and GitHub to signal seriousness and attract elite talent.
- Internal turbulence complicates the ambition: the flagship open-source Llama 4 model has stumbled, key staff have departed, and leadership is reportedly weighing whether to abandon its most powerful open-source model entirely.
- Meta raised its 2025 capital expenditure to as much as $72 billion and is betting that AI-powered ad tools, a consumer AI app, and smart glasses will eventually convert infrastructure into profit — though analysts warn the timeline is deeply uncertain.
- Markets offered a measured endorsement — shares rose one percent on the news — but the deeper question remains whether this is visionary infrastructure-building or an arms race with no clear finish line.
Mark Zuckerberg announced Monday that Meta would commit hundreds of billions of dollars to a fleet of massive AI data centres aimed at achieving superintelligence — machines capable of outthinking humans across a wide range of tasks. The first facility, Prometheus, is set to come online in 2026 at multigigawatt scale; a second, Hyperion, will eventually reach five gigawatts. Zuckerberg offered a vivid sense of the physical scale involved, noting that a single one of Meta's new titan clusters would cover a significant portion of Manhattan. Industry analysts expect Meta to become the first AI laboratory to operate a gigawatt-plus supercluster.
The announcement reflects an intensifying competition across the technology sector. Meta recently reorganized its AI operations under a new division called Superintelligence Labs, led by Alexandr Wang, former CEO of Scale AI, and Nat Friedman, former head of GitHub — a leadership shuffle accompanied by a $14.3 billion investment in Scale. Zuckerberg grounded the spending in Meta's financial strength, pointing to nearly $165 billion in revenue last year and a resilient advertising business. The company had already raised its 2025 capital expenditure guidance to between $64 billion and $72 billion. Markets responded with modest approval: shares rose one percent on the day and are up more than 20 percent year to date.
Yet the path forward is far from smooth. Meta has faced setbacks with its open-source Llama 4 model and lost key personnel, and senior figures within Superintelligence Labs have reportedly discussed abandoning the company's most powerful open-source model in favour of a proprietary alternative. The company is now counting on new revenue streams — a Meta AI app, AI-driven advertising tools, and smart glasses — to eventually justify the investment. Analysts note that while AI has already begun lifting Meta's ad business, the scale of this commitment is oriented toward a longer horizon, one that could take years to pay off and may never follow a predictable timeline. The central question is whether Meta is laying the foundation for a new era of computing, or simply joining an arms race whose endpoint no one can yet see.
Mark Zuckerberg announced on Monday that Meta would commit hundreds of billions of dollars to building a fleet of massive artificial intelligence data centres designed to pursue superintelligence—machines capable of outthinking humans across a broad range of tasks. The declaration marks an escalation in what has become a high-stakes competition among tech giants, each racing to secure the talent and infrastructure needed to lead the next generation of AI development.
The company's first facility, called Prometheus, is scheduled to come online in 2026 and will operate at multigigawatt scale. A second data centre, Hyperion, will eventually scale up to five gigawatts of capacity. Zuckerberg, speaking through his Threads social media platform, emphasized the sheer physical footprint of these operations: a single one of Meta's new titan clusters, he noted, would cover a significant portion of Manhattan. Industry analysts have already flagged Meta as the likely first AI laboratory to bring a gigawatt-plus supercluster into operation.
The spending spree reflects a broader pattern across the technology sector. OpenAI, Google, and other major players have recently signed high-profile deals and distributed multimillion-dollar compensation packages to lure top engineering talent away from competitors. Meta has been particularly aggressive in this regard. Last month, the company reorganized its artificial intelligence operations under a new division called Superintelligence Labs, placing it under the leadership of Alexandr Wang, the former CEO of Scale AI, and Nat Friedman, who previously ran GitHub. Meta had invested $14.3 billion in Scale as part of this leadership shuffle.
Zuckerberg justified the enormous capital commitment by pointing to Meta's financial strength. The company generated nearly $165 billion in revenue last year, and its core advertising business remains robust. "We have the capital from our business to do this," he said. In April, Meta had already raised its capital expenditure guidance for 2025 to between $64 billion and $72 billion, a figure designed to shore up the company's competitive position against rivals like OpenAI and Google. The stock market appeared to accept the rationale: Meta shares rose 1 percent on the announcement, and the stock has climbed more than 20 percent year to date.
Yet the investment carries significant uncertainty. Meta has experienced setbacks in its artificial intelligence efforts, including problems with its open-source Llama 4 model and the departure of key personnel. The company is now betting that Superintelligence Labs will generate new revenue streams through a Meta AI application, image-to-video advertising tools, and smart glasses. According to reporting from the New York Times, senior members of the unit have even discussed abandoning Behemoth, Meta's most powerful open-source AI model, in favor of developing a proprietary alternative instead.
Analysts acknowledge both the logic and the risk in Meta's approach. Gil Luria, an analyst at D.A. Davidson, noted that artificial intelligence has already begun to boost Meta's advertising business by enabling the company to sell more ads at higher prices. But at the scale Meta is now pursuing, he said, the investment is oriented toward a longer-term competition—one aimed at developing the leading AI model, a goal that could take years to achieve and may not produce returns on any predictable timeline. The question hanging over the entire enterprise is whether the infrastructure Meta is building will eventually justify its cost, or whether the company is simply following its competitors into an arms race with an uncertain endpoint.
Citas Notables
We have the capital from our business to do this— Mark Zuckerberg, Meta CEO
At this scale, the investment is more oriented to the long-term competition to have the leading AI model, which could take time to materialize— Gil Luria, D.A. Davidson analyst
La Conversación del Hearth Otra perspectiva de la historia
Why is Meta spending this much money right now, when the payoff from AI is still unclear?
Because the alternative—falling behind in the race for superintelligence—feels riskier to Zuckerberg than the capital expenditure itself. If OpenAI or Google crack this first, Meta's advertising business, which funds everything else, becomes vulnerable.
But Meta's advertising business is already strong. Why not wait and see which approach actually works?
That's the trap. By the time you know which approach works, the leader will have already locked in the talent, the data, and the infrastructure advantages. You can't catch up from behind in a technology race like this.
So this is about optionality—about not being the company that bet wrong?
Partly. But it's also about what Zuckerberg believes superintelligence will be worth. If he's right that it's the defining technology of the next decade, then spending hundreds of billions now looks cheap in retrospect.
And if he's wrong?
Then Meta has built the world's most expensive infrastructure for a technology that didn't materialize the way anyone expected. But by then, the money will already be spent.