Meta Cuts 10% of Reality Labs Staff Amid Metaverse Pivot to AI
Meta discontinues Horizon Workrooms and lays off 1,000+ Reality Labs employees as company shifts focus from metaverse to artificial intelligence development.

The artificial intelligence arms race has shifted battlegrounds. While 2024 and 2025 were defined by the scramble for silicon—specifically Nvidia’s GPUs—2026 is shaping up to be the year of the gigawatt. In a move that fundamentally alters the energy landscape for the tech sector, Meta has signed definitive agreements to secure up to 6.6 gigawatts (GW) of nuclear power. This massive commitment, involving partnerships with Vistra, TerraPower, and Oklo, underscores a critical reality: the path to Artificial General Intelligence (AGI) is paved not just with code, but with reliable, baseload electricity.
For industry observers, this announcement is more than a procurement deal; it is a signal that the physical constraints of AI infrastructure are now the primary bottleneck for growth. As hyperscalers like Meta, Microsoft, and Google scale their operations to train next-generation models, the intermittent nature of traditional renewables—wind and solar—is proving insufficient for the 24/7 energy demands of massive data centers.
Meta’s strategy is notable for its diversification. Rather than betting on a single provider or technology, the company has constructed a portfolio that balances immediate supply from existing reactors with long-term bets on next-generation Small Modular Reactors (SMRs). The agreements effectively split into three distinct categories: existing baseload updates, advanced sodium-cooled reactors, and micro-reactor campuses.
The deal structure reveals a timeline designed to ramp up power availability in sync with the deployment of future AI clusters, specifically the "Prometheus" supercluster in Ohio.
Table 1: Meta’s Nuclear Energy Partnerships and Capacity
| Partner | Technology Type | Capacity Commitment | Expected Timeline |
|---|---|---|---|
| Vistra Corp | Existing Nuclear Capacity | ~2.1 GW | Immediate / Ongoing |
| TerraPower | Natrium (Sodium-Cooled) | ~2.8 GW (Total Potential) | 2032 (Initial Units) |
| Oklo | Aurora Powerhouse (Micro-reactor) | ~1.2 GW | 2030 (First Phase) |
| Various | Uprates & Grid Improvements | ~500 MW | 2027-2029 |
Vistra: The Immediate Solution
The cornerstone of the immediate power supply comes from Vistra Corp. Meta has secured a 20-year Power Purchase Agreement (PPA) tied to the Beaver Valley plant in Pennsylvania and the Perry and Davis-Besse plants in Ohio. Crucially, this deal is not just about siphoning existing power; it involves funding "uprates"—technical modifications to existing reactors that increase their total output. This allows Meta to bring new capacity online without the decade-long regulatory hurdles associated with building new plants from scratch.
TerraPower and Oklo: The Future Bet
The longer-term components of the deal rely on advanced nuclear technologies. The partnership with TerraPower, backed by Bill Gates, focuses on deploying Natrium reactors. These sodium-cooled fast reactors are designed to be safer and more efficient than traditional light-water reactors. Meta has committed to funding the development of two initial units (690 MW) with rights to six additional units by 2035. Similarly, the deal with Oklo involves constructing a nuclear campus in Pike County, Ohio, utilizing their Aurora Powerhouse design, which targets a 2030 operational start.
The narrative surrounding the AI boom is undergoing a rapid transformation. For the past two years, the market's attention was monopolized by Nvidia and the supply of H100 and Blackwell chips. However, the deployment of these chips has revealed a stark physical reality: AI data centers are voracious energy consumers.
A standard data center might consume 30-50 megawatts (MW). In contrast, AI training clusters are now approaching the gigawatt scale—equivalent to the power consumption of a mid-sized city. The "Prometheus" supercluster in New Albany, Ohio, which Meta is aggressively developing, is expected to require at least 1 GW of power.
This energy density renders traditional renewable strategies complicated. While solar and wind are the cheapest forms of new energy, they are intermittent. AI training runs cannot be paused when the sun sets or the wind dies down. Batteries can bridge short gaps, but for gigawatt-scale "firming" (making variable power reliable), the cost becomes prohibitive. This economic physics drives the pivot to nuclear, which offers carbon-free energy with a capacity factor exceeding 90%.
While Meta’s announcement highlights nuclear power, the broader implications ripple through the entire commodities market. The "real" AI boom, as emerging market analysis suggests, is moving downstream into the physical infrastructure required to transmit this power.
Key Infrastructure Constraints:
The complexity of these physical constraints explains why Meta’s deal is geographically concentrated. By focusing on Ohio and Pennsylvania (the PJM Interconnection grid), Meta is locating its compute power near existing generation assets to minimize transmission bottlenecks.
Meta’s 6.6 GW acquisition places immense pressure on its competitors. In the zero-sum game of grid capacity, power secured by Meta is power unavailable to Google, Microsoft, or Amazon.
First-Mover Advantage in Energy
Historically, "first-mover advantage" in tech referred to launching a product. In the AI era, it refers to securing a utility contract. The U.S. power grid is currently clogged with "interconnection queues"—lists of projects waiting for permission to connect to the grid. By signing deals with Vistra for existing capacity and Oklo/TerraPower for dedicated behind-the-meter or co-located generation, Meta is effectively bypassing parts of this queue.
Sustainability vs. Reality
This move also signals a maturation in corporate sustainability goals. The industry is moving away from "matching" energy use with Renewable Energy Credits (RECs)—which often paper over the fact that a data center is running on coal or gas at night—toward "24/7 Carbon-Free Energy" (CFE). Nuclear is currently the only scalable technology that can satisfy the 24/7 CFE requirement for gigawatt-scale loads.
The scale of this energy procurement offers a glimpse into the size of the AI models Meta intends to train. A 6.6 GW portfolio is roughly enough energy to power 5 million American homes. Allocating this magnitude of resources to computational tasks suggests that Meta views its future AI models not as software updates, but as industrial-scale infrastructure projects.
As we look toward the latter half of the decade, the differentiating factor between leading AI labs may no longer be just talent or chip supply, but the ability to keep the lights on. With this historic nuclear agreement, Meta has ensured that its pursuit of AGI will not be throttled by a lack of power, setting a new standard for what it means to be an AI infrastructure company.