While most of the world was distracted by the glitz of new gadgets at CES last week, Mark Zuckerberg was busy rewriting the blueprints of the global energy grid.
On Monday, Meta officially launched Meta Compute a new, high-level organization dedicated to one thing: securing the raw physical power needed to win the AI arms race.
Zuckerberg’s announcement wasn’t just a business update; it was a declaration of war. He isn’t just building faster apps; he’s building a computational empire measured in gigawatts, not just gigabytes. If 2024 was about the models (Llama), and 2025 was about the agents, 2026 is officially the year of the Infrastructure.
Here is the breakdown of Meta’s massive pivot and why the “Metaverse” company is suddenly acting like a power utility.
1. The Numbers are Staggering: Tens to Hundreds of Gigawatts
Zuckerberg revealed that Meta Compute plans to build “tens of gigawatts” of AI compute capacity this decade, with the long-term goal of hitting hundreds of gigawatts.
To put that in perspective: One gigawatt can power roughly 750,000 homes. Meta is talking about building the energy equivalent of several large countries just to run its AI. This is a scale of infrastructure that dwarfs almost anything attempted in the history of the private sector.
By the end of 2026, Meta aims to have over 10 GW of total capacity online. They’ve already committed to spending over $600 billion in the U.S. alone through 2028 to make this happen.
2. “Prometheus” and the Nuclear Bet
You can’t run a gigawatt-scale data center on standard grid power you’d black out the neighborhood. To solve the energy crisis, Meta is going nuclear.
Just last week, the company signed landmark deals for 6.6 GW of nuclear power with providers like Vistra, TerraPower (Bill Gates’ company), and Oklo (backed by Sam Altman). This power will fuel Prometheus, Meta’s massive 1-gigawatt AI supercluster in New Albany, Ohio, which is scheduled to come online later this year. Prometheus isn’t just a data center; it’s a city-sized brain designed to train the next generation of “Superintelligent” models.
3. The Dream Team: Janardhan and Gross
To lead this charge, Zuckerberg has assembled a “war cabinet” for Meta Compute:
- Santosh Janardhan: The veteran head of infrastructure who will oversee the technical stack—from custom silicon (MTIA chips) to the cooling systems in the data centers.
- Daniel Gross: The former co-founder of Safe Superintelligence and a legendary AI investor. His job is the “long game” securing the land, the supply chains for millions of chips, and the government partnerships needed to build at this scale.
- Dina Powell McCormick: A former top-level government advisor and banking executive who just joined as President and Vice Chair to handle the complex geopolitical and financial side of building gigawatt-scale projects.
4. The Pivot: Farewell to (some of) the Metaverse
The launch of Meta Compute comes with a cold reality check. This week, Meta also announced layoffs of roughly 10% of its Reality Labs division (about 1,500 people).
While Zuckerberg hasn’t abandoned the idea of smart glasses or VR, the message is clear: the blank checks for the “Metaverse” are over. The money is being reallocated to the “Meta Compute” engine. Meta has realized that before you can build a virtual world, you have to build the superintelligent AI that can generate it and that requires power.
5. Custom Silicon: Breaking the Nvidia Dependency
Nvidia is still the king, but Meta is tired of being its most expensive customer. A key part of the Meta Compute push is the acceleration of MTIA (Meta Training and Inference Accelerator).
Meta is now testing its first custom training chips (developed with TSMC) in production. By controlling the chip, the software, and the data center, Meta can run its AI workloads far more efficiently than its competitors. They aren’t just buying the car; they are building the engine, the fuel, and the road.
The Verdict: Infrastructure is the New Moat
For years, people argued about which AI model was the “smartest.” But as we enter 2026, we’re realizing that smarts can be copied. Infrastructure cannot.
By building its own power plants, its own chips, and tens of gigawatts of data centers, Meta is building a physical “moat” that startups simply cannot cross. It doesn’t matter how smart your new AI startup is if you can’t find a plug to turn it on.
Zuckerberg is betting that in the end, the winner of the AI era won’t be the person with the best code—it will be the person with the most electricity.
What do you think? Is Meta’s $600 billion power play a visionary move, or is this the biggest “CapEx bubble” in history?
Leave a Reply