Meta has converted Ohio Facebook site into first phase of AI campus

Meta has converted its New Albany, Ohio, Facebook data center into the first phase of Prometheus, an AI-focused supercluster it plans to scale to a gigawatt, the company said in a blog post. Data Center Dynamics reported the post on Oct. 23.

The site now holds hundreds of thousands of GPUs spread across multiple interconnected buildings, and it has already been used to train Meta's Muse foundation models. The halls are a mix of traditional and AI-optimized data center buildings plus rapid deployment structures, all connected. At full build, Prometheus is planned to comprise roughly one million distributed GPUs.

For funds and family offices underwriting AI campuses, the disclosure puts the diligence weight on megawatts delivered and heat rejected rather than floorplate. Meta runs its GB300 deployment on a platform it calls Clemente, which can use either its modular Air-Assisted Liquid Cooling or Facility Liquid Cooling. That fork likely decides how much of the mechanical plant lands on the tenant's spec rather than the shell the landlord financed.

The post does not say how much of the gigawatt is live or what the conversion cost. The hardware mix is Nvidia GB200 and GB300 chips, with Vera Rubin "arriving soon," according to Meta.

The GB200 silicon sits in Meta's Catalina rack architecture. In the Ariel variant, a two-rack configuration, 36 GPUs per rack are linked by NVLink into a unified 72-GPU domain. Each Blackwell GPU pairs with a Grace CPU and two ConnectX-7 400G NICs at a 1:1 CPU-to-GPU ratio, supporting 800G of I/O per GPU.

GB300 raises HBM capacity per GPU by 1.5x, from 192GB to 288GB, and dense FP4 throughput by a similar factor while roughly doubling softmax acceleration, Meta said. Nvidia's ConnectX-8 SuperNICs deliver 800G per GPU natively, which Meta said makes the Ariel-era capability the baseline.

At the building level, each Ethernet-based fabric interconnects up to 20,736 GPUs through a multi-layer topology. Those fabrics are then joined by a Backend Aggregation network spanning 2km to over 100km, built with Arista 7800-series chassis running deep-buffered Broadcom Jericho3 silicon and delivering multiple Pbps of bandwidth per region pair, Meta said.